{
 "id": "r2f99bf28246",
 "found": true,
 "parent_id": "r5247d91c07d",
 "created_at": "2026-08-31 21:59:52.207568+00:00",
 "rows": [
  {
   "url": "https://en.wikipedia.org/wiki/GLM_(AI)",
   "title": "GLM (AI)",
   "published_at": "2026-08-30T18:59:23",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-02",
   "access_type": null,
   "license": null,
   "significance": "major model family generation starting the GLM-5 series utilizing the DeepSeek Sparse Attention mechanism",
   "organization_evidence": [
    "GLM, short for General Language Model, is a series of open weight large language models developed by Chinese software company Z.ai."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "GLM, short for General Language Model, is a series of open weight large language models developed by Chinese software company Z.ai.",
    "| GLM-5 | February 2026 |  |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "GLM, short for General Language Model, is a series of open weight large language models developed by Chinese software company Z.ai.",
    "Beginning with GLM-5, Z.ai adopted the DeepSeek Sparse Attention mechanism.",
    "| GLM-5 | February 2026 |  |"
   ]
  },
  {
   "url": "https://en.wikipedia.org/wiki/GLM_(AI)",
   "title": "GLM (AI)",
   "published_at": "2026-08-30T18:59:23",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "iteration of the GLM-5 series",
   "organization_evidence": [
    "GLM, short for General Language Model, is a series of open weight large language models developed by Chinese software company Z.ai."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "GLM, short for General Language Model, is a series of open weight large language models developed by Chinese software company Z.ai.",
    "| GLM-5.1 | April 2026 |  |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "GLM, short for General Language Model, is a series of open weight large language models developed by Chinese software company Z.ai.",
    "| GLM-5.1 | April 2026 |  |"
   ]
  },
  {
   "url": "https://en.wikipedia.org/wiki/GLM_(AI)",
   "title": "GLM (AI)",
   "published_at": "2026-08-30T18:59:23",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-06",
   "access_type": null,
   "license": null,
   "significance": "5.2 iteration of the model released in June 2026 was nearly as performant as Claude Opus 4.7 to 4.8",
   "organization_evidence": [
    "GLM, short for General Language Model, is a series of open weight large language models developed by Chinese software company Z.ai."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "GLM, short for General Language Model, is a series of open weight large language models developed by Chinese software company Z.ai.",
    "The company's benchmarks show that the 5.2 iteration of the model released in June 2026 was nearly as performant as Claude Opus 4.7 to 4.8."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "GLM, short for General Language Model, is a series of open weight large language models developed by Chinese software company Z.ai.",
    "The company's benchmarks show that the 5.2 iteration of the model released in June 2026 was nearly as performant as Claude Opus 4.7 to 4.8."
   ]
  },
  {
   "url": "https://en.wikipedia.org/wiki/GLM_(AI)",
   "title": "GLM (AI)",
   "published_at": "2026-08-30T18:59:23",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-08-14",
   "access_type": null,
   "license": null,
   "significance": "major model version whose weights were made available two weeks after its August 14 release",
   "organization_evidence": [
    "GLM, short for General Language Model, is a series of open weight large language models developed by Chinese software company Z.ai."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "GLM, short for General Language Model, is a series of open weight large language models developed by Chinese software company Z.ai.",
    "Z.ai released GLM-5.3 on 14 August 2026 and made the model's weights available two weeks later."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "GLM, short for General Language Model, is a series of open weight large language models developed by Chinese software company Z.ai.",
    "Z.ai released GLM-5.3 on 14 August 2026 and made the model's weights available two weeks later."
   ]
  },
  {
   "url": "https://en.wikipedia.org/wiki/GLM_(AI)",
   "title": "GLM (AI)",
   "published_at": "2026-08-30T18:59:23",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-08-26",
   "access_type": null,
   "license": null,
   "significance": "model incorporating Kimi Delta Attention developed by Moonshot AI and served using AI chips manufactured in China",
   "organization_evidence": [
    "GLM, short for General Language Model, is a series of open weight large language models developed by Chinese software company Z.ai."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "GLM, short for General Language Model, is a series of open weight large language models developed by Chinese software company Z.ai.",
    "Later that month, Z.ai anonymously previewed a model called Ox Alpha on OpenRouter, which was released with weights on 26 August as GLM-5.3-Flash."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "GLM, short for General Language Model, is a series of open weight large language models developed by Chinese software company Z.ai.",
    "Later that month, Z.ai anonymously previewed a model called Ox Alpha on OpenRouter, which was released with weights on 26 August as GLM-5.3-Flash.",
    "Upon full release, Z.ai disclosed that they had been serving the model using AI chips manufactured in China.",
    "GLM-5.3-Flash incorporates Kimi Delta Attention, developed by competing Chinese lab Moonshot AI."
   ]
  },
  {
   "url": "https://en.wikipedia.org/wiki/GLM_5.3_Flash",
   "title": "GLM (AI)",
   "published_at": "2026-08-28T16:54:58",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-08-14",
   "access_type": null,
   "license": null,
   "significance": "Nearly as performant as Claude Opus 4.7 to 4.8 / flagship model series of Z.ai",
   "organization_evidence": [
    "GLM, short for General Language Model, is a series of open weight large language models developed by Chinese software company Z.ai."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Z.ai released GLM-5.3 on 14 August 2026 and made the model's weights available two weeks later.",
    "GLM, short for General Language Model, is a series of open weight large language models developed by Chinese software company Z.ai."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "GLM is the flagship model series of Z.ai, one of the six AI tigers of China.",
    "GLM, short for General Language Model, is a series of open weight large language models developed by Chinese software company Z.ai."
   ]
  },
  {
   "url": "https://en.wikipedia.org/wiki/GLM_5.3_Flash",
   "title": "GLM (AI)",
   "published_at": "2026-08-28T16:54:58",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-08-26",
   "access_type": null,
   "license": null,
   "significance": "Flash version of GLM-5.3 released with weights",
   "organization_evidence": [
    "Later that month, Z.ai anonymously previewed a model called Ox Alpha on OpenRouter, which was released with weights on 26 August as GLM-5.3-Flash.",
    "GLM, short for General Language Model, is a series of open weight large language models developed by Chinese software company Z.ai."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Later that month, Z.ai anonymously previewed a model called Ox Alpha on OpenRouter, which was released with weights on 26 August as GLM-5.3-Flash.",
    "GLM, short for General Language Model, is a series of open weight large language models developed by Chinese software company Z.ai."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Later that month, Z.ai anonymously previewed a model called Ox Alpha on OpenRouter, which was released with weights on 26 August as GLM-5.3-Flash.",
    "GLM, short for General Language Model, is a series of open weight large language models developed by Chinese software company Z.ai."
   ]
  },
  {
   "url": "https://arxiv.org/html/2604.07190v1",
   "title": "The ATOM Report:Measuring the Open Language Model Ecosystem",
   "published_at": null,
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-02",
   "access_type": null,
   "license": null,
   "significance": "A major update of the Qwen model family featuring new variants such as Qwen3.5-4B, Qwen3.5-35B-A3B, Qwen3.5-122B-A10B, and Qwen3.5-397B-A17B that led to massive adoption dominance.",
   "organization_evidence": [
    "The ATOM Report:\nMeasuring the Open Language Model Ecosystem",
    "## 1 Introduction",
    "We document a clear trend where Chinese models overtook their counterparts built in the U.S. in the summer of 2025 and subsequently widened the gap over their western counterparts.",
    "The growth of the Chinese model ecosystem can largely be attributed to Alibaba\u2019s Qwen, which is responsible for almost a billion cumulative downloads by March 2026."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The ATOM Report:\nMeasuring the Open Language Model Ecosystem",
    "## 5 Model Adoption by Organization",
    "### 5.1 Ecosystem Leaders (Qwen, Llama, DeepSeek, Mistral, and OpenAI)",
    "Qwen\u2019s path to leading the ecosystem.",
    "The early adoption numbers of Qwen3.5 (Qwen Team, 2026) since its release in February 2026 are an indication that the dominance of Qwen relative to its peers will continue (see Section 6 and specifically Fig. 15 for early data on Qwen 3.5\u2019s adoption)."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The ATOM Report:\nMeasuring the Open Language Model Ecosystem",
    "## 6 The Relative Adoption Metric (RAM)",
    "### 6.2 Using RAM to Measure New Models",
    "For example, using an April 2, 2026 snapshot of Hugging Face data, the RAM framework makes the February 2026 Qwen3.5 rollout easier to compare across scales."
   ]
  },
  {
   "url": "https://en.wikipedia.org/wiki/ChatGLM",
   "title": "GLM (AI)",
   "published_at": "2026-08-29T17:42:15",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-08-14",
   "access_type": null,
   "license": null,
   "significance": "A major new model-family iteration adopting the DeepSeek Sparse Attention mechanism and making its weights available.",
   "organization_evidence": [
    "GLM, short for General Language Model, is a series of open weight large language models developed by Chinese software company Z.ai."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Z.ai released GLM-5.3 on 14 August 2026 and made the model's weights available two weeks later.",
    "GLM, short for General Language Model, is a series of open weight large language models developed by Chinese software company Z.ai."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Beginning with GLM-5, Z.ai adopted the DeepSeek Sparse Attention mechanism.",
    "Z.ai released GLM-5.3 on 14 August 2026 and made the model's weights available two weeks later.",
    "GLM, short for General Language Model, is a series of open weight large language models developed by Chinese software company Z.ai."
   ]
  },
  {
   "url": "https://en.wikipedia.org/wiki/ChatGLM",
   "title": "GLM (AI)",
   "published_at": "2026-08-29T17:42:15",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-08-26",
   "access_type": null,
   "license": null,
   "significance": "A model incorporating Kimi Delta Attention, developed by competing Chinese lab Moonshot AI, released with weights.",
   "organization_evidence": [
    "GLM, short for General Language Model, is a series of open weight large language models developed by Chinese software company Z.ai."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Later that month, Z.ai anonymously previewed a model called Ox Alpha on OpenRouter, which was released with weights on 26 August as GLM-5.3-Flash.",
    "GLM, short for General Language Model, is a series of open weight large language models developed by Chinese software company Z.ai."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "GLM-5.3-Flash incorporates Kimi Delta Attention, developed by competing Chinese lab Moonshot AI.",
    "Later that month, Z.ai anonymously previewed a model called Ox Alpha on OpenRouter, which was released with weights on 26 August as GLM-5.3-Flash.",
    "GLM, short for General Language Model, is a series of open weight large language models developed by Chinese software company Z.ai."
   ]
  },
  {
   "url": "https://en.wikipedia.org/wiki/GLM_5.3",
   "title": "GLM (AI)",
   "published_at": "2026-08-26T23:05:36",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-08-26",
   "access_type": null,
   "license": null,
   "significance": "released with weights as GLM-5.3-Flash after an anonymous preview on OpenRouter",
   "organization_evidence": [
    "GLM, short for General Language Model, is a series of open weight large language models developed by Chinese software company Z.ai."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Z.ai anonymously previewed a model called Ox Alpha on OpenRouter, which was released with weights on 26 August as GLM-5.3-Flash.",
    "GLM, short for General Language Model, is a series of open weight large language models developed by Chinese software company Z.ai."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Later that month, Z.ai anonymously previewed a model called Ox Alpha on OpenRouter, which was released with weights on 26 August as GLM-5.3-Flash.",
    "GLM, short for General Language Model, is a series of open weight large language models developed by Chinese software company Z.ai."
   ]
  },
  {
   "url": "https://hidekazu-konishi.com/entry/open_weights_llm_release_history_and_timeline.html",
   "title": "Open-Weights LLM Release History and Timeline - Llama, Mistral, Qwen, DeepSeek, and gpt-oss Model Families, Capability Evolution, and Licenses",
   "published_at": "2026-07-26T00:00:00",
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-02-16",
   "access_type": null,
   "license": null,
   "significance": "integrating advances in multimodal learning, architectural efficiency, and large-scale reinforcement learning",
   "organization_evidence": [
    "## Open-Weights LLM Release History and Timeline - Llama, Mistral, Qwen, DeepSeek, and gpt-oss Model Families, Capability Evolution, and Licenses",
    "Here is the chronological timeline of major open-weights model releases across the families above, in ascending order.",
    "| 2026-02-16 | Qwen | **Qwen3.5 is released** under Apache 2.0, starting with the Qwen3.5-397B-A17B Mixture-of-Experts model; additional sizes follow within weeks (122B-A10B, 35B-A3B, and 27B on 2026-02-24, and 9B / 4B / 2B / 0.8B on 2026-03-02). Alibaba describes the generation as integrating advances in multimodal learning, architectural efficiency, and large-scale reinforcement learning. Sources: [QwenLM/Qwen3.6 on GitHub (news log for Qwen3.5 and Qwen3.6)](https://github.com/QwenLM/Qwen3.6), [Qwen3.5 collection on Hugging Face](https://huggingface.co/collections/Qwen/qwen35)."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## Open-Weights LLM Release History and Timeline - Llama, Mistral, Qwen, DeepSeek, and gpt-oss Model Families, Capability Evolution, and Licenses",
    "Here is the chronological timeline of major open-weights model releases across the families above, in ascending order.",
    "| 2026-02-16 | Qwen | **Qwen3.5 is released** under Apache 2.0, starting with the Qwen3.5-397B-A17B Mixture-of-Experts model; additional sizes follow within weeks (122B-A10B, 35B-A3B, and 27B on 2026-02-24, and 9B / 4B / 2B / 0.8B on 2026-03-02). Alibaba describes the generation as integrating advances in multimodal learning, architectural efficiency, and large-scale reinforcement learning. Sources: [QwenLM/Qwen3.6 on GitHub (news log for Qwen3.5 and Qwen3.6)](https://github.com/QwenLM/Qwen3.6), [Qwen3.5 collection on Hugging Face](https://huggingface.co/collections/Qwen/qwen35)."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## Open-Weights LLM Release History and Timeline - Llama, Mistral, Qwen, DeepSeek, and gpt-oss Model Families, Capability Evolution, and Licenses",
    "Here is the chronological timeline of major open-weights model releases across the families above, in ascending order.",
    "| 2026-02-16 | Qwen | **Qwen3.5 is released** under Apache 2.0, starting with the Qwen3.5-397B-A17B Mixture-of-Experts model; additional sizes follow within weeks (122B-A10B, 35B-A3B, and 27B on 2026-02-24, and 9B / 4B / 2B / 0.8B on 2026-03-02). Alibaba describes the generation as integrating advances in multimodal learning, architectural efficiency, and large-scale reinforcement learning. Sources: [QwenLM/Qwen3.6 on GitHub (news log for Qwen3.5 and Qwen3.6)](https://github.com/QwenLM/Qwen3.6), [Qwen3.5 collection on Hugging Face](https://huggingface.co/collections/Qwen/qwen35)."
   ]
  },
  {
   "url": "https://hidekazu-konishi.com/entry/open_weights_llm_release_history_and_timeline.html",
   "title": "Open-Weights LLM Release History and Timeline - Llama, Mistral, Qwen, DeepSeek, and gpt-oss Model Families, Capability Evolution, and Licenses",
   "published_at": "2026-07-26T00:00:00",
   "organization": "Microsoft",
   "model": null,
   "release_date": "2026-03-04",
   "access_type": null,
   "license": null,
   "significance": "a compact open-weight multimodal reasoning model that combines vision understanding with selective chain-of-thought reasoning",
   "organization_evidence": [
    "## Open-Weights LLM Release History and Timeline - Llama, Mistral, Qwen, DeepSeek, and gpt-oss Model Families, Capability Evolution, and Licenses",
    "Here is the chronological timeline of major open-weights model releases across the families above, in ascending order.",
    "| 2026-03-04 | Phi | **Phi-4-reasoning-vision (15B) is released**, a compact open-weight multimodal reasoning model that combines vision understanding with selective chain-of-thought reasoning. Source: [Phi-4-reasoning-vision](https://www.microsoft.com/en-us/research/blog/phi-4-reasoning-vision-and-the-lessons-of-training-a-multimodal-reasoning-model/). |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## Open-Weights LLM Release History and Timeline - Llama, Mistral, Qwen, DeepSeek, and gpt-oss Model Families, Capability Evolution, and Licenses",
    "Here is the chronological timeline of major open-weights model releases across the families above, in ascending order.",
    "| 2026-03-04 | Phi | **Phi-4-reasoning-vision (15B) is released**, a compact open-weight multimodal reasoning model that combines vision understanding with selective chain-of-thought reasoning. Source: [Phi-4-reasoning-vision](https://www.microsoft.com/en-us/research/blog/phi-4-reasoning-vision-and-the-lessons-of-training-a-multimodal-reasoning-model/). |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## Open-Weights LLM Release History and Timeline - Llama, Mistral, Qwen, DeepSeek, and gpt-oss Model Families, Capability Evolution, and Licenses",
    "Here is the chronological timeline of major open-weights model releases across the families above, in ascending order.",
    "| 2026-03-04 | Phi | **Phi-4-reasoning-vision (15B) is released**, a compact open-weight multimodal reasoning model that combines vision understanding with selective chain-of-thought reasoning. Source: [Phi-4-reasoning-vision](https://www.microsoft.com/en-us/research/blog/phi-4-reasoning-vision-and-the-lessons-of-training-a-multimodal-reasoning-model/). |"
   ]
  },
  {
   "url": "https://hidekazu-konishi.com/entry/open_weights_llm_release_history_and_timeline.html",
   "title": "Open-Weights LLM Release History and Timeline - Llama, Mistral, Qwen, DeepSeek, and gpt-oss Model Families, Capability Evolution, and Licenses",
   "published_at": "2026-07-26T00:00:00",
   "organization": "Mistral AI",
   "model": null,
   "release_date": "2026-03-16",
   "access_type": null,
   "license": null,
   "significance": "a sparse Mixture-of-Experts model that unifies reasoning, multimodal understanding, and agentic coding in one model, with a 256K context window and configurable reasoning effort",
   "organization_evidence": [
    "## Open-Weights LLM Release History and Timeline - Llama, Mistral, Qwen, DeepSeek, and gpt-oss Model Families, Capability Evolution, and Licenses",
    "Here is the chronological timeline of major open-weights model releases across the families above, in ascending order.",
    "| 2026-03-16 | Mistral | **Mistral Small 4 is released** under Apache 2.0, a sparse Mixture-of-Experts model that unifies reasoning, multimodal understanding, and agentic coding in one model, with a 256K context window and configurable reasoning effort. Source: [Mistral Small 4](https://mistral.ai/news/mistral-small-4/). |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## Open-Weights LLM Release History and Timeline - Llama, Mistral, Qwen, DeepSeek, and gpt-oss Model Families, Capability Evolution, and Licenses",
    "Here is the chronological timeline of major open-weights model releases across the families above, in ascending order.",
    "| 2026-03-16 | Mistral | **Mistral Small 4 is released** under Apache 2.0, a sparse Mixture-of-Experts model that unifies reasoning, multimodal understanding, and agentic coding in one model, with a 256K context window and configurable reasoning effort. Source: [Mistral Small 4](https://mistral.ai/news/mistral-small-4/). |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## Open-Weights LLM Release History and Timeline - Llama, Mistral, Qwen, DeepSeek, and gpt-oss Model Families, Capability Evolution, and Licenses",
    "Here is the chronological timeline of major open-weights model releases across the families above, in ascending order.",
    "| 2026-03-16 | Mistral | **Mistral Small 4 is released** under Apache 2.0, a sparse Mixture-of-Experts model that unifies reasoning, multimodal understanding, and agentic coding in one model, with a 256K context window and configurable reasoning effort. Source: [Mistral Small 4](https://mistral.ai/news/mistral-small-4/). |"
   ]
  },
  {
   "url": "https://hidekazu-konishi.com/entry/open_weights_llm_release_history_and_timeline.html",
   "title": "Open-Weights LLM Release History and Timeline - Llama, Mistral, Qwen, DeepSeek, and gpt-oss Model Families, Capability Evolution, and Licenses",
   "published_at": "2026-07-26T00:00:00",
   "organization": "Google",
   "model": null,
   "release_date": "2026-04-02",
   "access_type": null,
   "license": null,
   "significance": "edge E2B/E4B sizes plus larger dense and Mixture-of-Experts models",
   "organization_evidence": [
    "## Open-Weights LLM Release History and Timeline - Llama, Mistral, Qwen, DeepSeek, and gpt-oss Model Families, Capability Evolution, and Licenses",
    "Here is the chronological timeline of major open-weights model releases across the families above, in ascending order.",
    "| 2026-04-02 | Gemma | **Gemma 4 is released** (edge E2B/E4B sizes plus larger dense and Mixture-of-Experts models), Google's first Gemma generation published under the Apache 2.0 license rather than the Gemma Terms of Use, with a focus on agentic workflows. Source: [Introducing Gemma 4](https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/). |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## Open-Weights LLM Release History and Timeline - Llama, Mistral, Qwen, DeepSeek, and gpt-oss Model Families, Capability Evolution, and Licenses",
    "Here is the chronological timeline of major open-weights model releases across the families above, in ascending order.",
    "| 2026-04-02 | Gemma | **Gemma 4 is released** (edge E2B/E4B sizes plus larger dense and Mixture-of-Experts models), Google's first Gemma generation published under the Apache 2.0 license rather than the Gemma Terms of Use, with a focus on agentic workflows. Source: [Introducing Gemma 4](https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/). |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## Open-Weights LLM Release History and Timeline - Llama, Mistral, Qwen, DeepSeek, and gpt-oss Model Families, Capability Evolution, and Licenses",
    "Here is the chronological timeline of major open-weights model releases across the families above, in ascending order.",
    "| 2026-04-02 | Gemma | **Gemma 4 is released** (edge E2B/E4B sizes plus larger dense and Mixture-of-Experts models), Google's first Gemma generation published under the Apache 2.0 license rather than the Gemma Terms of Use, with a focus on agentic workflows. Source: [Introducing Gemma 4](https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/). |"
   ]
  },
  {
   "url": "https://hidekazu-konishi.com/entry/open_weights_llm_release_history_and_timeline.html",
   "title": "Open-Weights LLM Release History and Timeline - Llama, Mistral, Qwen, DeepSeek, and gpt-oss Model Families, Capability Evolution, and Licenses",
   "published_at": "2026-07-26T00:00:00",
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-04-16",
   "access_type": null,
   "license": null,
   "significance": "building on Qwen3.5 while prioritizing stability, real-world coding utility, and direct community feedback",
   "organization_evidence": [
    "## Open-Weights LLM Release History and Timeline - Llama, Mistral, Qwen, DeepSeek, and gpt-oss Model Families, Capability Evolution, and Licenses",
    "Here is the chronological timeline of major open-weights model releases across the families above, in ascending order.",
    "| 2026-04-16 | Qwen | **Qwen3.6 is released** under Apache 2.0, with Qwen3.6-35B-A3B (2026-04-16) followed by Qwen3.6-27B (2026-04-22). Alibaba describes the generation as building on Qwen3.5 while prioritizing stability, real-world coding utility, and direct community feedback. Sources: [QwenLM/Qwen3.6 on GitHub](https://github.com/QwenLM/Qwen3.6), [Qwen3.6 collection on Hugging Face](https://huggingface.co/collections/Qwen/qwen36)."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## Open-Weights LLM Release History and Timeline - Llama, Mistral, Qwen, DeepSeek, and gpt-oss Model Families, Capability Evolution, and Licenses",
    "Here is the chronological timeline of major open-weights model releases across the families above, in ascending order.",
    "| 2026-04-16 | Qwen | **Qwen3.6 is released** under Apache 2.0, with Qwen3.6-35B-A3B (2026-04-16) followed by Qwen3.6-27B (2026-04-22). Alibaba describes the generation as building on Qwen3.5 while prioritizing stability, real-world coding utility, and direct community feedback. Sources: [QwenLM/Qwen3.6 on GitHub](https://github.com/QwenLM/Qwen3.6), [Qwen3.6 collection on Hugging Face](https://huggingface.co/collections/Qwen/qwen36)."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## Open-Weights LLM Release History and Timeline - Llama, Mistral, Qwen, DeepSeek, and gpt-oss Model Families, Capability Evolution, and Licenses",
    "Here is the chronological timeline of major open-weights model releases across the families above, in ascending order.",
    "| 2026-04-16 | Qwen | **Qwen3.6 is released** under Apache 2.0, with Qwen3.6-35B-A3B (2026-04-16) followed by Qwen3.6-27B (2026-04-22). Alibaba describes the generation as building on Qwen3.5 while prioritizing stability, real-world coding utility, and direct community feedback. Sources: [QwenLM/Qwen3.6 on GitHub](https://github.com/QwenLM/Qwen3.6), [Qwen3.6 collection on Hugging Face](https://huggingface.co/collections/Qwen/qwen36)."
   ]
  },
  {
   "url": "https://hidekazu-konishi.com/entry/open_weights_llm_release_history_and_timeline.html",
   "title": "Open-Weights LLM Release History and Timeline - Llama, Mistral, Qwen, DeepSeek, and gpt-oss Model Families, Capability Evolution, and Licenses",
   "published_at": "2026-07-26T00:00:00",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "built on token-wise compression and DeepSeek Sparse Attention, and making a 1M-token context window the standard across DeepSeek's official services",
   "organization_evidence": [
    "## Open-Weights LLM Release History and Timeline - Llama, Mistral, Qwen, DeepSeek, and gpt-oss Model Families, Capability Evolution, and Licenses",
    "Here is the chronological timeline of major open-weights model releases across the families above, in ascending order.",
    "| 2026-04-24 | DeepSeek | **DeepSeek-V4 Preview is released with open weights**, in two Mixture-of-Experts variants - DeepSeek-V4-Pro (1.6T total / 49B active parameters) and DeepSeek-V4-Flash (284B total / 13B active) - built on token-wise compression and DeepSeek Sparse Attention, and making a 1M-token context window the standard across DeepSeek's official services. Source: [DeepSeek-V4 Preview Release](https://api-docs.deepseek.com/news/news260424). |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## Open-Weights LLM Release History and Timeline - Llama, Mistral, Qwen, DeepSeek, and gpt-oss Model Families, Capability Evolution, and Licenses",
    "Here is the chronological timeline of major open-weights model releases across the families above, in ascending order.",
    "| 2026-04-24 | DeepSeek | **DeepSeek-V4 Preview is released with open weights**, in two Mixture-of-Experts variants - DeepSeek-V4-Pro (1.6T total / 49B active parameters) and DeepSeek-V4-Flash (284B total / 13B active) - built on token-wise compression and DeepSeek Sparse Attention, and making a 1M-token context window the standard across DeepSeek's official services. Source: [DeepSeek-V4 Preview Release](https://api-docs.deepseek.com/news/news260424). |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## Open-Weights LLM Release History and Timeline - Llama, Mistral, Qwen, DeepSeek, and gpt-oss Model Families, Capability Evolution, and Licenses",
    "Here is the chronological timeline of major open-weights model releases across the families above, in ascending order.",
    "| 2026-04-24 | DeepSeek | **DeepSeek-V4 Preview is released with open weights**, in two Mixture-of-Experts variants - DeepSeek-V4-Pro (1.6T total / 49B active parameters) and DeepSeek-V4-Flash (284B total / 13B active) - built on token-wise compression and DeepSeek Sparse Attention, and making a 1M-token context window the standard across DeepSeek's official services. Source: [DeepSeek-V4 Preview Release](https://api-docs.deepseek.com/news/news260424). |"
   ]
  },
  {
   "url": "https://hidekazu-konishi.com/entry/open_weights_llm_release_history_and_timeline.html",
   "title": "Open-Weights LLM Release History and Timeline - Llama, Mistral, Qwen, DeepSeek, and gpt-oss Model Families, Capability Evolution, and Licenses",
   "published_at": "2026-07-26T00:00:00",
   "organization": "Mistral AI",
   "model": null,
   "release_date": "2026-05-22",
   "access_type": null,
   "license": null,
   "significance": "a 128B dense model with a 256K context window that merges instruction-following, reasoning, and coding in a single set of weights",
   "organization_evidence": [
    "## Open-Weights LLM Release History and Timeline - Llama, Mistral, Qwen, DeepSeek, and gpt-oss Model Families, Capability Evolution, and Licenses",
    "Here is the chronological timeline of major open-weights model releases across the families above, in ascending order.",
    "| 2026-05-22 | Mistral | **Mistral Medium 3.5 is released** as open weights under a modified MIT license - a 128B dense model with a 256K context window that merges instruction-following, reasoning, and coding in a single set of weights, introduced alongside Mistral's Vibe coding environment and remote agents (the product itself is announced as available in public preview). Source: [Remote agents in Vibe. Powered by Mistral Medium 3.5.](https://mistral.ai/news/vibe-remote-agents-mistral-medium-3-5/). |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## Open-Weights LLM Release History and Timeline - Llama, Mistral, Qwen, DeepSeek, and gpt-oss Model Families, Capability Evolution, and Licenses",
    "Here is the chronological timeline of major open-weights model releases across the families above, in ascending order.",
    "| 2026-05-22 | Mistral | **Mistral Medium 3.5 is released** as open weights under a modified MIT license - a 128B dense model with a 256K context window that merges instruction-following, reasoning, and coding in a single set of weights, introduced alongside Mistral's Vibe coding environment and remote agents (the product itself is announced as available in public preview). Source: [Remote agents in Vibe. Powered by Mistral Medium 3.5.](https://mistral.ai/news/vibe-remote-agents-mistral-medium-3-5/). |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## Open-Weights LLM Release History and Timeline - Llama, Mistral, Qwen, DeepSeek, and gpt-oss Model Families, Capability Evolution, and Licenses",
    "Here is the chronological timeline of major open-weights model releases across the families above, in ascending order.",
    "| 2026-05-22 | Mistral | **Mistral Medium 3.5 is released** as open weights under a modified MIT license - a 128B dense model with a 256K context window that merges instruction-following, reasoning, and coding in a single set of weights, introduced alongside Mistral's Vibe coding environment and remote agents (the product itself is announced as available in public preview). Source: [Remote agents in Vibe. Powered by Mistral Medium 3.5.](https://mistral.ai/news/vibe-remote-agents-mistral-medium-3-5/). |"
   ]
  },
  {
   "url": "https://edgen.tech/vi/news/post/kimi-k3-open-weight-release-draws-100k-downloads-erases-us-ai-lead",
   "title": "Kimi K3 open-weight release draws 100k downloads, erases US AI lead",
   "published_at": "2026-08-17T03:11:14",
   "organization": "Moonshot AI",
   "model": "Kimi K3",
   "release_date": "2026-07-27",
   "access_type": "open-weight",
   "license": "charges MaaS providers and cloud vendors with revenue above $20 million over 12 consecutive months, while leaving end-user applications free",
   "significance": "a 2.8-trillion-parameter open-weight model, matched or surpassed Anthropic's Fable 5 on key benchmarks at roughly half the cost, erasing the closed-source lead",
   "organization_evidence": [
    "Kimi K3 open-weight release draws 100k downloads, erases US AI lead",
    "Moonshot AI's Kimi K3, a 2.8-trillion-parameter open-weight model, matched or surpassed Anthropic's Fable 5 on key benchmarks at roughly half the cost, erasing the closed-source lead."
   ],
   "model_evidence": [
    "Kimi K3 open-weight release draws 100k downloads, erases US AI lead",
    "Moonshot AI's Kimi K3, a 2.8-trillion-parameter open-weight model, matched or surpassed Anthropic's Fable 5 on key benchmarks at roughly half the cost, erasing the closed-source lead."
   ],
   "release_date_evidence": [
    "Kimi K3 open-weight release draws 100k downloads, erases US AI lead",
    "The model's July 27 release on Hugging Face drew roughly 100,000 downloads in its first 24 hours, and demand was so high that Moonshot temporarily blocked new consumer sign-ups."
   ],
   "access_type_evidence": [
    "Kimi K3 open-weight release draws 100k downloads, erases US AI lead",
    "Moonshot AI's Kimi K3, a 2.8-trillion-parameter open-weight model, matched or surpassed Anthropic's Fable 5 on key benchmarks at roughly half the cost, erasing the closed-source lead."
   ],
   "license_evidence": [
    "Kimi K3 open-weight release draws 100k downloads, erases US AI lead",
    "Kimi K3 introduced a new license that charges MaaS providers and cloud vendors with revenue above $20 million over 12 consecutive months, while leaving end-user applications free."
   ],
   "significance_evidence": [
    "Kimi K3 open-weight release draws 100k downloads, erases US AI lead",
    "Moonshot AI's Kimi K3, a 2.8-trillion-parameter open-weight model, matched or surpassed Anthropic's Fable 5 on key benchmarks at roughly half the cost, erasing the closed-source lead."
   ]
  },
  {
   "url": "https://en.wikipedia.org/wiki/Ox_Alpha",
   "title": "GLM (AI)",
   "published_at": "2026-08-26T23:05:36",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-08-26",
   "access_type": null,
   "license": null,
   "significance": "major model released with weights on OpenRouter as GLM-5.3-Flash",
   "organization_evidence": [
    "GLM, short for General Language Model, is a series of open weight large language models developed by Chinese software company Z.ai."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "GLM, short for General Language Model, is a series of open weight large language models developed by Chinese software company Z.ai."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Later that month, Z.ai anonymously previewed a model called Ox Alpha on OpenRouter, which was released with weights on 26 August as GLM-5.3-Flash.",
    "GLM, short for General Language Model, is a series of open weight large language models developed by Chinese software company Z.ai."
   ]
  },
  {
   "url": "https://qtithow.com/2026/07/comparing-120-open-source-llms-best.html",
   "title": "Comparing 120+ Open-Source LLMs: The Best Models in 2026",
   "published_at": null,
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "Mixture-of-experts model with 1.6 trillion parameters activating 49 billion tokens, scoring 87.5% on MMLU-Pro, 90.1% on GPQA Diamond, and 93.5% on LiveCodeBench with native 1-million-token context",
   "organization_evidence": [
    "Comparing 120+ Open-Source LLMs: The Best Models in 2026",
    "Recent releases show the shift clearly. **DeepSeek V4 Pro** (released April 24, 2026), **GLM-5.1** from Z.ai, **Kimi K2.6** from Moonshot AI, and **Qwen3.5** from Alibaba now rival the best proprietary LLMs\u2014including Claude Opus 4.8, GPT-5.5, and Gemini 3.1 Pro.",
    "| DeepSeek V4 Pro | DeepSeek | 1.6T (49B active) | 1M | 87.5% MMLU-Pro | 90.1% GPQA | 93.5% LiveCodeBench | | | MIT | $0.35 | $0.70 | April 2026 |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Comparing 120+ Open-Source LLMs: The Best Models in 2026",
    "Recent releases show the shift clearly. **DeepSeek V4 Pro** (released April 24, 2026), **GLM-5.1** from Z.ai, **Kimi K2.6** from Moonshot AI, and **Qwen3.5** from Alibaba now rival the best proprietary LLMs\u2014including Claude Opus 4.8, GPT-5.5, and Gemini 3.1 Pro.",
    "| DeepSeek V4 Pro | DeepSeek | 1.6T (49B active) | 1M | 87.5% MMLU-Pro | 90.1% GPQA | 93.5% LiveCodeBench | | | MIT | $0.35 | $0.70 | April 2026 |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Comparing 120+ Open-Source LLMs: The Best Models in 2026",
    "**DeepSeek V4 Pro** (April 24, 2026) is the new leader.",
    "This mixture-of-experts model with 1.6 trillion parameters activates just 49 billion tokens.",
    "It scores 87.5% on MMLU-Pro, 90.1% on GPQA Diamond, and 93.5% on LiveCodeBench."
   ]
  },
  {
   "url": "https://qtithow.com/2026/07/comparing-120-open-source-llms-best.html",
   "title": "Comparing 120+ Open-Source LLMs: The Best Models in 2026",
   "published_at": null,
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-04-01",
   "access_type": null,
   "license": null,
   "significance": "Leads SWE-Bench Pro at 58.4% among open models, mixture-of-experts with 754 billion parameters trained on Huawei Ascend chips",
   "organization_evidence": [
    "Comparing 120+ Open-Source LLMs: The Best Models in 2026",
    "Recent releases show the shift clearly. **DeepSeek V4 Pro** (released April 24, 2026), **GLM-5.1** from Z.ai, **Kimi K2.6** from Moonshot AI, and **Qwen3.5** from Alibaba now rival the best proprietary LLMs\u2014including Claude Opus 4.8, GPT-5.5, and Gemini 3.1 Pro.",
    "| GLM-5.1 | Z.ai | 754B | 200K | 91.7% MMLU | 85.7% GPQA | 58.4% SWE-Bench Pro | | | MIT | $0.30 | $2.15 | April 2026 |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Comparing 120+ Open-Source LLMs: The Best Models in 2026",
    "Recent releases show the shift clearly. **DeepSeek V4 Pro** (released April 24, 2026), **GLM-5.1** from Z.ai, **Kimi K2.6** from Moonshot AI, and **Qwen3.5** from Alibaba now rival the best proprietary LLMs\u2014including Claude Opus 4.8, GPT-5.5, and Gemini 3.1 Pro.",
    "| GLM-5.1 | Z.ai | 754B | 200K | 91.7% MMLU | 85.7% GPQA | 58.4% SWE-Bench Pro | | | MIT | $0.30 | $2.15 | April 2026 |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Comparing 120+ Open-Source LLMs: The Best Models in 2026",
    "**GLM-5.1** from Z.ai (formerly Zhipu) leads SWE-Bench Pro at 58.4%\u2014beating GPT-5.4 (57.7%) and Claude Opus 4.6 (57.3%).",
    "This mixture-of-experts with 754 billion parameters was trained entirely on Huawei Ascend chips and ships under MIT license."
   ]
  },
  {
   "url": "https://qtithow.com/2026/07/comparing-120-open-source-llms-best.html",
   "title": "Comparing 120+ Open-Source LLMs: The Best Models in 2026",
   "published_at": null,
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-04-01",
   "access_type": null,
   "license": null,
   "significance": "Ranks second for open performance with 92% on HumanEval, 90.5% on GPQA Diamond, 96.4% on AIME 2026, with 256K context and native video input",
   "organization_evidence": [
    "Comparing 120+ Open-Source LLMs: The Best Models in 2026",
    "Recent releases show the shift clearly. **DeepSeek V4 Pro** (released April 24, 2026), **GLM-5.1** from Z.ai, **Kimi K2.6** from Moonshot AI, and **Qwen3.5** from Alibaba now rival the best proprietary LLMs\u2014including Claude Opus 4.8, GPT-5.5, and Gemini 3.1 Pro.",
    "| Kimi K2.6 | Moonshot AI | 1T (32B active) | 256K | 84.6% MMLU-Pro | 90.5% GPQA | 92% HumanEval | | | Modified MIT | $0.28 | $1.10 | April 2026 |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Comparing 120+ Open-Source LLMs: The Best Models in 2026",
    "Recent releases show the shift clearly. **DeepSeek V4 Pro** (released April 24, 2026), **GLM-5.1** from Z.ai, **Kimi K2.6** from Moonshot AI, and **Qwen3.5** from Alibaba now rival the best proprietary LLMs\u2014including Claude Opus 4.8, GPT-5.5, and Gemini 3.1 Pro.",
    "| Kimi K2.6 | Moonshot AI | 1T (32B active) | 256K | 84.6% MMLU-Pro | 90.5% GPQA | 92% HumanEval | | | Modified MIT | $0.28 | $1.10 | April 2026 |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Comparing 120+ Open-Source LLMs: The Best Models in 2026",
    "**Kimi K2.6** from Moonshot AI ranks second for open performance: 92% on HumanEval, 90.5% on GPQA Diamond, 96.4% on AIME 2026, with 256K context and native video input."
   ]
  },
  {
   "url": "https://qtithow.com/2026/07/comparing-120-open-source-llms-best.html",
   "title": "Comparing 120+ Open-Source LLMs: The Best Models in 2026",
   "published_at": null,
   "organization": "Alibaba",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Rivals the best proprietary LLMs",
   "organization_evidence": [
    "Comparing 120+ Open-Source LLMs: The Best Models in 2026",
    "Recent releases show the shift clearly. **DeepSeek V4 Pro** (released April 24, 2026), **GLM-5.1** from Z.ai, **Kimi K2.6** from Moonshot AI, and **Qwen3.5** from Alibaba now rival the best proprietary LLMs\u2014including Claude Opus 4.8, GPT-5.5, and Gemini 3.1 Pro."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Comparing 120+ Open-Source LLMs: The Best Models in 2026",
    "Recent releases show the shift clearly. **DeepSeek V4 Pro** (released April 24, 2026), **GLM-5.1** from Z.ai, **Kimi K2.6** from Moonshot AI, and **Qwen3.5** from Alibaba now rival the best proprietary LLMs\u2014including Claude Opus 4.8, GPT-5.5, and Gemini 3.1 Pro."
   ]
  },
  {
   "url": "https://tech-insider.org/au/open-source-ai-model-wave-2026",
   "title": "Open Source AI Models: 9 Launches in 12 Days [2026]",
   "published_at": "2026-08-01T01:03:53",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "a 975-billion-parameter multimodal model",
   "organization_evidence": [
    "# Open Source AI Models: 9 Launches in 12 Days [2026]",
    "Thinking Machines Lab, founded by former OpenAI chief technology officer Mira Murati, released **Inkling** on July 15, 2026: a 975-billion-parameter multimodal model under a fully permissive Apache 2.0 license, arriving in the same stretch as new open-weight drops from Moonshot AI, Zhipu AI, Google DeepMind and Mistral AI.",
    "## The Full Release Wave, Model by Model",
    "| Model | Developer | Release Date | Total Params | Active Params | License |",
    "| Inkling | Thinking Machines Lab | July 15, 2026 | 975B | 41B | Apache 2.0 |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Open Source AI Models: 9 Launches in 12 Days [2026]",
    "Thinking Machines Lab, founded by former OpenAI chief technology officer Mira Murati, released **Inkling** on July 15, 2026: a 975-billion-parameter multimodal model under a fully permissive Apache 2.0 license, arriving in the same stretch as new open-weight drops from Moonshot AI, Zhipu AI, Google DeepMind and Mistral AI.",
    "## The Full Release Wave, Model by Model",
    "| Model | Developer | Release Date | Total Params | Active Params | License |",
    "| Inkling | Thinking Machines Lab | July 15, 2026 | 975B | 41B | Apache 2.0 |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Open Source AI Models: 9 Launches in 12 Days [2026]",
    "Thinking Machines Lab, founded by former OpenAI chief technology officer Mira Murati, released **Inkling** on July 15, 2026: a 975-billion-parameter multimodal model under a fully permissive Apache 2.0 license, arriving in the same stretch as new open-weight drops from Moonshot AI, Zhipu AI, Google DeepMind and Mistral AI."
   ]
  },
  {
   "url": "https://tech-insider.org/au/open-source-ai-model-wave-2026",
   "title": "Open Source AI Models: 9 Launches in 12 Days [2026]",
   "published_at": "2026-08-01T01:03:53",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-16",
   "access_type": null,
   "license": null,
   "significance": "2.8 trillion total parameters and 104 billion active",
   "organization_evidence": [
    "# Open Source AI Models: 9 Launches in 12 Days [2026]",
    "Thinking Machines Lab, founded by former OpenAI chief technology officer Mira Murati, released **Inkling** on July 15, 2026: a 975-billion-parameter multimodal model under a fully permissive Apache 2.0 license, arriving in the same stretch as new open-weight drops from Moonshot AI, Zhipu AI, Google DeepMind and Mistral AI.",
    "## The Full Release Wave, Model by Model",
    "| Model | Developer | Release Date | Total Params | Active Params | License |",
    "| Kimi K3 | Moonshot AI | July 16, 2026 (weights July 27) | 2.8T | 104B | Open weights |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Open Source AI Models: 9 Launches in 12 Days [2026]",
    "## The Full Release Wave, Model by Model",
    "| Model | Developer | Release Date | Total Params | Active Params | License |",
    "| Kimi K3 | Moonshot AI | July 16, 2026 (weights July 27) | 2.8T | 104B | Open weights |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Open Source AI Models: 9 Launches in 12 Days [2026]",
    "Moonshot AI\u2019s [Kimi K3 launched on July 16 with 2.8 trillion total parameters and 104 billion active](/au/kimi-k3-launch-2026/) \u2013 a scale that led multiple trackers to describe the model, unveiled in July 2026, as the world\u2019s largest open-weight model to date \u2013 continuing a release cadence that has also pushed Moonshot\u2019s own valuation sharply higher this year, as covered in our report on [Moonshot\u2019s push toward a $30 billion valuation](/au/moonshot-ai-kimi-valuation-2026/)."
   ]
  },
  {
   "url": "https://tech-insider.org/au/open-source-ai-model-wave-2026",
   "title": "Open Source AI Models: 9 Launches in 12 Days [2026]",
   "published_at": "2026-08-01T01:03:53",
   "organization": "Zhipu AI",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "it posted a 62.1% score on SWE-bench Pro, carries a 1-million-token context window",
   "organization_evidence": [
    "# Open Source AI Models: 9 Launches in 12 Days [2026]",
    "Thinking Machines Lab, founded by former OpenAI chief technology officer Mira Murati, released **Inkling** on July 15, 2026: a 975-billion-parameter multimodal model under a fully permissive Apache 2.0 license, arriving in the same stretch as new open-weight drops from Moonshot AI, Zhipu AI, Google DeepMind and Mistral AI.",
    "## The Full Release Wave, Model by Model",
    "| Model | Developer | Release Date | Total Params | Active Params | License |",
    "| GLM-5.2 | Zhipu AI (Z.ai) | June 13, 2026 | 744B | 40B | MIT |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Open Source AI Models: 9 Launches in 12 Days [2026]",
    "## The Full Release Wave, Model by Model",
    "| Model | Developer | Release Date | Total Params | Active Params | License |",
    "| GLM-5.2 | Zhipu AI (Z.ai) | June 13, 2026 | 744B | 40B | MIT |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Open Source AI Models: 9 Launches in 12 Days [2026]",
    "Zhipu AI\u2019s GLM-5.2, released June 13 under an MIT license, has already been covered in detail on this site: it posted a 62.1% score on SWE-bench Pro, carries a 1-million-token context window, and became, according to industry benchmark trackers, [the strongest open-weight model on the Artificial Analysis Intelligence Index](/au/glm-5-2-tops-open-weight-ai-models-2026/), trailing only Claude Fable 5, Claude Opus 4.8 and GPT-5.5 among all models, closed or open."
   ]
  },
  {
   "url": "https://tech-insider.org/au/open-source-ai-model-wave-2026",
   "title": "Open Source AI Models: 9 Launches in 12 Days [2026]",
   "published_at": "2026-08-01T01:03:53",
   "organization": "Google DeepMind",
   "model": null,
   "release_date": "2026-07",
   "access_type": null,
   "license": null,
   "significance": "27-billion-parameter build sits at the opposite end, aimed at teams that want something they can run on a single high-end GPU",
   "organization_evidence": [
    "# Open Source AI Models: 9 Launches in 12 Days [2026]",
    "Thinking Machines Lab, founded by former OpenAI chief technology officer Mira Murati, released **Inkling** on July 15, 2026: a 975-billion-parameter multimodal model under a fully permissive Apache 2.0 license, arriving in the same stretch as new open-weight drops from Moonshot AI, Zhipu AI, Google DeepMind and Mistral AI.",
    "## The Full Release Wave, Model by Model",
    "| Model | Developer | Release Date | Total Params | Active Params | License |",
    "| Gemma 3 (27B) | Google DeepMind | July 2026 | 27B | n/a (dense) | Open license |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Open Source AI Models: 9 Launches in 12 Days [2026]",
    "## The Full Release Wave, Model by Model",
    "| Model | Developer | Release Date | Total Params | Active Params | License |",
    "| Gemma 3 (27B) | Google DeepMind | July 2026 | 27B | n/a (dense) | Open license |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Open Source AI Models: 9 Launches in 12 Days [2026]",
    "The spread is wide. Kimi K3\u2019s 2.8 trillion total parameters make it the largest model in the group by a wide margin, while Gemma 3\u2019s 27-billion-parameter build sits at the opposite end, aimed at teams that want something they can run on a single high-end GPU rather than a data center rack."
   ]
  },
  {
   "url": "https://tech-insider.org/au/open-source-ai-model-wave-2026",
   "title": "Open Source AI Models: 9 Launches in 12 Days [2026]",
   "published_at": "2026-08-01T01:03:53",
   "organization": "Mistral AI",
   "model": null,
   "release_date": "2026-07",
   "access_type": null,
   "license": null,
   "significance": "24 billion parameters, positioning it as a lighter, cheaper-to-run option rather than a frontier flagship",
   "organization_evidence": [
    "# Open Source AI Models: 9 Launches in 12 Days [2026]",
    "Thinking Machines Lab, founded by former OpenAI chief technology officer Mira Murati, released **Inkling** on July 15, 2026: a 975-billion-parameter multimodal model under a fully permissive Apache 2.0 license, arriving in the same stretch as new open-weight drops from Moonshot AI, Zhipu AI, Google DeepMind and Mistral AI.",
    "## The Full Release Wave, Model by Model",
    "| Model | Developer | Release Date | Total Params | Active Params | License |",
    "| Mistral Medium 3 | Mistral AI | July 2026 | 24B | n/a (dense) | Open weights |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Open Source AI Models: 9 Launches in 12 Days [2026]",
    "## The Full Release Wave, Model by Model",
    "| Model | Developer | Release Date | Total Params | Active Params | License |",
    "| Mistral Medium 3 | Mistral AI | July 2026 | 24B | n/a (dense) | Open weights |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Open Source AI Models: 9 Launches in 12 Days [2026]",
    "Mistral AI, which has spent much of 2026 diversifying beyond pure chat models (including the robotics-focused release detailed in our story on [Mistral\u2019s push toward a $23 billion valuation](/au/mistral-robotics-model-valuation-2026/)), added Mistral Medium 3 to the July wave at 24 billion parameters, positioning it as a lighter, cheaper-to-run option rather than a frontier flagship \u2013 the latest step after a rapid-fire 2026 that saw Mistral ship the fully free, Apache-2.0-licensed Mistral 3 family (7B, 30B and 96B) on February 10, a 64k-context Mistral 3.1 update in March, and the 119-billion-parameter, 6.5-billion-active Mistral Small 4 on March 20."
   ]
  },
  {
   "url": "https://fazm.ai/t/latest-open-source-llm-releases-2026",
   "title": "Fazm - Claude Code & Codex GUI for macOS | Persistent Sessions",
   "published_at": "2026-06-20T17:35:27",
   "organization": "MiniMax",
   "model": "MiniMax M3",
   "release_date": "2026-06-01",
   "access_type": "open-weight",
   "license": "Conditional (review terms)",
   "significance": "First open-weight model pairing frontier coding, 1M context, and native multimodality",
   "organization_evidence": [
    "# Latest open source LLM releases, 2026",
    "| Model | Lab | Shipped | Parameters | Context | License |",
    "| MiniMax M3First open-weight model pairing frontier coding, 1M context, and native multimodality. Vendor-reported 59.0% SWE-Bench Pro. Weights staged to Hugging Face after launch. | MiniMax | Jun 1, 2026 | Undisclosed (MoE) | 1M | Conditional (review terms) |"
   ],
   "model_evidence": [
    "# Latest open source LLM releases, 2026",
    "| Model | Lab | Shipped | Parameters | Context | License |",
    "| MiniMax M3First open-weight model pairing frontier coding, 1M context, and native multimodality. Vendor-reported 59.0% SWE-Bench Pro. Weights staged to Hugging Face after launch. | MiniMax | Jun 1, 2026 | Undisclosed (MoE) | 1M | Conditional (review terms) |"
   ],
   "release_date_evidence": [
    "# Latest open source LLM releases, 2026",
    "| Model | Lab | Shipped | Parameters | Context | License |",
    "| MiniMax M3First open-weight model pairing frontier coding, 1M context, and native multimodality. Vendor-reported 59.0% SWE-Bench Pro. Weights staged to Hugging Face after launch. | MiniMax | Jun 1, 2026 | Undisclosed (MoE) | 1M | Conditional (review terms) |"
   ],
   "access_type_evidence": [
    "# Latest open source LLM releases, 2026",
    "A running list of the open-weight models that have shipped this year, in release order, with the facts that age well: ship date, parameter count, context window, and license.",
    "| MiniMax M3First open-weight model pairing frontier coding, 1M context, and native multimodality. Vendor-reported 59.0% SWE-Bench Pro. Weights staged to Hugging Face after launch. | MiniMax | Jun 1, 2026 | Undisclosed (MoE) | 1M | Conditional (review terms) |"
   ],
   "license_evidence": [
    "| Model | Lab | Shipped | Parameters | Context | License |",
    "| MiniMax M3First open-weight model pairing frontier coding, 1M context, and native multimodality. Vendor-reported 59.0% SWE-Bench Pro. Weights staged to Hugging Face after launch. | MiniMax | Jun 1, 2026 | Undisclosed (MoE) | 1M | Conditional (review terms) |"
   ],
   "significance_evidence": [
    "# Latest open source LLM releases, 2026",
    "| Model | Lab | Shipped | Parameters | Context | License |",
    "| MiniMax M3First open-weight model pairing frontier coding, 1M context, and native multimodality. Vendor-reported 59.0% SWE-Bench Pro. Weights staged to Hugging Face after launch. | MiniMax | Jun 1, 2026 | Undisclosed (MoE) | 1M | Conditional (review terms) |"
   ]
  },
  {
   "url": "https://fazm.ai/t/latest-open-source-llm-releases-2026",
   "title": "Fazm - Claude Code & Codex GUI for macOS | Persistent Sessions",
   "published_at": "2026-06-20T17:35:27",
   "organization": "Z.ai (Zhipu AI)",
   "model": "GLM-5.2",
   "release_date": "2026-06-12",
   "access_type": "open weight",
   "license": "Open weight",
   "significance": "Successor to GLM-5.1 positioned for long-horizon agentic engineering",
   "organization_evidence": [
    "# Latest open source LLM releases, 2026",
    "| Model | Lab | Shipped | Parameters | Context | License |",
    "| GLM-5.2Successor to GLM-5.1 (vendor-reported 58.4% SWE-Bench Pro). Positioned for long-horizon agentic engineering. | Z.ai (Zhipu AI) | Jun 12-13, 2026 | MoE | Long-horizon agentic | Open weight |"
   ],
   "model_evidence": [
    "# Latest open source LLM releases, 2026",
    "| Model | Lab | Shipped | Parameters | Context | License |",
    "| GLM-5.2Successor to GLM-5.1 (vendor-reported 58.4% SWE-Bench Pro). Positioned for long-horizon agentic engineering. | Z.ai (Zhipu AI) | Jun 12-13, 2026 | MoE | Long-horizon agentic | Open weight |"
   ],
   "release_date_evidence": [
    "# Latest open source LLM releases, 2026",
    "| Model | Lab | Shipped | Parameters | Context | License |",
    "| GLM-5.2Successor to GLM-5.1 (vendor-reported 58.4% SWE-Bench Pro). Positioned for long-horizon agentic engineering. | Z.ai (Zhipu AI) | Jun 12-13, 2026 | MoE | Long-horizon agentic | Open weight |"
   ],
   "access_type_evidence": [
    "# Latest open source LLM releases, 2026",
    "A running list of the open-weight models that have shipped this year, in release order, with the facts that age well: ship date, parameter count, context window, and license.",
    "| Model | Lab | Shipped | Parameters | Context | License |",
    "| GLM-5.2Successor to GLM-5.1 (vendor-reported 58.4% SWE-Bench Pro). Positioned for long-horizon agentic engineering. | Z.ai (Zhipu AI) | Jun 12-13, 2026 | MoE | Long-horizon agentic | Open weight |"
   ],
   "license_evidence": [
    "# Latest open source LLM releases, 2026",
    "| Model | Lab | Shipped | Parameters | Context | License |",
    "| GLM-5.2Successor to GLM-5.1 (vendor-reported 58.4% SWE-Bench Pro). Positioned for long-horizon agentic engineering. | Z.ai (Zhipu AI) | Jun 12-13, 2026 | MoE | Long-horizon agentic | Open weight |"
   ],
   "significance_evidence": [
    "# Latest open source LLM releases, 2026",
    "| Model | Lab | Shipped | Parameters | Context | License |",
    "| GLM-5.2Successor to GLM-5.1 (vendor-reported 58.4% SWE-Bench Pro). Positioned for long-horizon agentic engineering. | Z.ai (Zhipu AI) | Jun 12-13, 2026 | MoE | Long-horizon agentic | Open weight |"
   ]
  },
  {
   "url": "https://fazm.ai/t/latest-open-source-llm-releases-2026",
   "title": "Fazm - Claude Code & Codex GUI for macOS | Persistent Sessions",
   "published_at": "2026-06-20T17:35:27",
   "organization": "Moonshot AI",
   "model": "Kimi K2.7 Code",
   "release_date": "2026-06-12",
   "access_type": "open-weight",
   "license": "Modified MIT",
   "significance": "Coding-focused refresh of the K2 line",
   "organization_evidence": [
    "# Latest open source LLM releases, 2026",
    "| Model | Lab | Shipped | Parameters | Context | License |",
    "| Kimi K2.7 CodeCoding-focused refresh of the K2 line shipped the same week as GLM-5.2. | Moonshot AI | Jun 12-13, 2026 | ~1T total (MoE) | Agentic coding | Modified MIT |"
   ],
   "model_evidence": [
    "# Latest open source LLM releases, 2026",
    "| Model | Lab | Shipped | Parameters | Context | License |",
    "| Kimi K2.7 CodeCoding-focused refresh of the K2 line shipped the same week as GLM-5.2. | Moonshot AI | Jun 12-13, 2026 | ~1T total (MoE) | Agentic coding | Modified MIT |"
   ],
   "release_date_evidence": [
    "# Latest open source LLM releases, 2026",
    "| Model | Lab | Shipped | Parameters | Context | License |",
    "| Kimi K2.7 CodeCoding-focused refresh of the K2 line shipped the same week as GLM-5.2. | Moonshot AI | Jun 12-13, 2026 | ~1T total (MoE) | Agentic coding | Modified MIT |"
   ],
   "access_type_evidence": [
    "# Latest open source LLM releases, 2026",
    "A running list of the open-weight models that have shipped this year, in release order, with the facts that age well: ship date, parameter count, context window, and license.",
    "| Model | Lab | Shipped | Parameters | Context | License |",
    "| Kimi K2.7 CodeCoding-focused refresh of the K2 line shipped the same week as GLM-5.2. | Moonshot AI | Jun 12-13, 2026 | ~1T total (MoE) | Agentic coding | Modified MIT |"
   ],
   "license_evidence": [
    "# Latest open source LLM releases, 2026",
    "| Model | Lab | Shipped | Parameters | Context | License |",
    "| Kimi K2.7 CodeCoding-focused refresh of the K2 line shipped the same week as GLM-5.2. | Moonshot AI | Jun 12-13, 2026 | ~1T total (MoE) | Agentic coding | Modified MIT |"
   ],
   "significance_evidence": [
    "# Latest open source LLM releases, 2026",
    "| Model | Lab | Shipped | Parameters | Context | License |",
    "| Kimi K2.7 CodeCoding-focused refresh of the K2 line shipped the same week as GLM-5.2. | Moonshot AI | Jun 12-13, 2026 | ~1T total (MoE) | Agentic coding | Modified MIT |"
   ]
  },
  {
   "url": "https://fazm.ai/t/latest-open-source-llm-releases-2026",
   "title": "Fazm - Claude Code & Codex GUI for macOS | Persistent Sessions",
   "published_at": "2026-06-20T17:35:27",
   "organization": "DeepSeek",
   "model": "DeepSeek V4",
   "release_date": "2026-04-24",
   "access_type": "open-weight",
   "license": "MIT",
   "significance": "Two MoE sizes, native 1M context, trained on 32T+ tokens",
   "organization_evidence": [
    "# Latest open source LLM releases, 2026",
    "| Model | Lab | Shipped | Parameters | Context | License |",
    "| DeepSeek V4Two MoE sizes, native 1M context, trained on 32T+ tokens. The clean-license frontier MoE of the spring. | DeepSeek | Apr 24, 2026 (preview) | V4-Pro 1.6T/49B active; V4-Flash 284B/13B active | 1M | MIT |"
   ],
   "model_evidence": [
    "# Latest open source LLM releases, 2026",
    "| Model | Lab | Shipped | Parameters | Context | License |",
    "| DeepSeek V4Two MoE sizes, native 1M context, trained on 32T+ tokens. The clean-license frontier MoE of the spring. | DeepSeek | Apr 24, 2026 (preview) | V4-Pro 1.6T/49B active; V4-Flash 284B/13B active | 1M | MIT |"
   ],
   "release_date_evidence": [
    "# Latest open source LLM releases, 2026",
    "| Model | Lab | Shipped | Parameters | Context | License |",
    "| DeepSeek V4Two MoE sizes, native 1M context, trained on 32T+ tokens. The clean-license frontier MoE of the spring. | DeepSeek | Apr 24, 2026 (preview) | V4-Pro 1.6T/49B active; V4-Flash 284B/13B active | 1M | MIT |"
   ],
   "access_type_evidence": [
    "# Latest open source LLM releases, 2026",
    "A running list of the open-weight models that have shipped this year, in release order, with the facts that age well: ship date, parameter count, context window, and license.",
    "| Model | Lab | Shipped | Parameters | Context | License |",
    "| DeepSeek V4Two MoE sizes, native 1M context, trained on 32T+ tokens. The clean-license frontier MoE of the spring. | DeepSeek | Apr 24, 2026 (preview) | V4-Pro 1.6T/49B active; V4-Flash 284B/13B active | 1M | MIT |"
   ],
   "license_evidence": [
    "# Latest open source LLM releases, 2026",
    "| Model | Lab | Shipped | Parameters | Context | License |",
    "| DeepSeek V4Two MoE sizes, native 1M context, trained on 32T+ tokens. The clean-license frontier MoE of the spring. | DeepSeek | Apr 24, 2026 (preview) | V4-Pro 1.6T/49B active; V4-Flash 284B/13B active | 1M | MIT |"
   ],
   "significance_evidence": [
    "# Latest open source LLM releases, 2026",
    "| Model | Lab | Shipped | Parameters | Context | License |",
    "| DeepSeek V4Two MoE sizes, native 1M context, trained on 32T+ tokens. The clean-license frontier MoE of the spring. | DeepSeek | Apr 24, 2026 (preview) | V4-Pro 1.6T/49B active; V4-Flash 284B/13B active | 1M | MIT |"
   ]
  },
  {
   "url": "https://fazm.ai/t/latest-open-source-llm-releases-2026",
   "title": "Fazm - Claude Code & Codex GUI for macOS | Persistent Sessions",
   "published_at": "2026-06-20T17:35:27",
   "organization": "Moonshot AI",
   "model": "Kimi K2.6",
   "release_date": "2026-04-20",
   "access_type": "open-weight",
   "license": "Modified MIT",
   "significance": "Native multimodal agentic model",
   "organization_evidence": [
    "# Latest open source LLM releases, 2026",
    "| Model | Lab | Shipped | Parameters | Context | License |",
    "| Kimi K2.6Native multimodal agentic model. MIT-equivalent below 100M MAU / $20M monthly revenue, attribution clause above. | Moonshot AI | Apr 20, 2026 | 1T total / 32B active (MoE) | Agentic | Modified MIT |"
   ],
   "model_evidence": [
    "# Latest open source LLM releases, 2026",
    "| Model | Lab | Shipped | Parameters | Context | License |",
    "| Kimi K2.6Native multimodal agentic model. MIT-equivalent below 100M MAU / $20M monthly revenue, attribution clause above. | Moonshot AI | Apr 20, 2026 | 1T total / 32B active (MoE) | Agentic | Modified MIT |"
   ],
   "release_date_evidence": [
    "# Latest open source LLM releases, 2026",
    "| Model | Lab | Shipped | Parameters | Context | License |",
    "| Kimi K2.6Native multimodal agentic model. MIT-equivalent below 100M MAU / $20M monthly revenue, attribution clause above. | Moonshot AI | Apr 20, 2026 | 1T total / 32B active (MoE) | Agentic | Modified MIT |"
   ],
   "access_type_evidence": [
    "# Latest open source LLM releases, 2026",
    "A running list of the open-weight models that have shipped this year, in release order, with the facts that age well: ship date, parameter count, context window, and license.",
    "| Model | Lab | Shipped | Parameters | Context | License |",
    "| Kimi K2.6Native multimodal agentic model. MIT-equivalent below 100M MAU / $20M monthly revenue, attribution clause above. | Moonshot AI | Apr 20, 2026 | 1T total / 32B active (MoE) | Agentic | Modified MIT |"
   ],
   "license_evidence": [
    "# Latest open source LLM releases, 2026",
    "| Model | Lab | Shipped | Parameters | Context | License |",
    "| Kimi K2.6Native multimodal agentic model. MIT-equivalent below 100M MAU / $20M monthly revenue, attribution clause above. | Moonshot AI | Apr 20, 2026 | 1T total / 32B active (MoE) | Agentic | Modified MIT |"
   ],
   "significance_evidence": [
    "# Latest open source LLM releases, 2026",
    "| Model | Lab | Shipped | Parameters | Context | License |",
    "| Kimi K2.6Native multimodal agentic model. MIT-equivalent below 100M MAU / $20M monthly revenue, attribution clause above. | Moonshot AI | Apr 20, 2026 | 1T total / 32B active (MoE) | Agentic | Modified MIT |"
   ]
  },
  {
   "url": "https://fazm.ai/t/latest-open-source-llm-releases-2026",
   "title": "Fazm - Claude Code & Codex GUI for macOS | Persistent Sessions",
   "published_at": "2026-06-20T17:35:27",
   "organization": "Mistral",
   "model": "Mistral Medium 3.5",
   "release_date": "2026-04-29",
   "access_type": "open-weight",
   "license": "Modified MIT",
   "significance": "Dense coder, permissive for individuals and most companies",
   "organization_evidence": [
    "# Latest open source LLM releases, 2026",
    "| Model | Lab | Shipped | Parameters | Context | License |",
    "| Mistral Medium 3.5Dense coder. Permissive for individuals and most companies; read the revenue clause before large-scale commercial deployment. | Mistral | Apr 29, 2026 | 128B dense | 256K | Modified MIT |"
   ],
   "model_evidence": [
    "# Latest open source LLM releases, 2026",
    "| Model | Lab | Shipped | Parameters | Context | License |",
    "| Mistral Medium 3.5Dense coder. Permissive for individuals and most companies; read the revenue clause before large-scale commercial deployment. | Mistral | Apr 29, 2026 | 128B dense | 256K | Modified MIT |"
   ],
   "release_date_evidence": [
    "# Latest open source LLM releases, 2026",
    "| Model | Lab | Shipped | Parameters | Context | License |",
    "| Mistral Medium 3.5Dense coder. Permissive for individuals and most companies; read the revenue clause before large-scale commercial deployment. | Mistral | Apr 29, 2026 | 128B dense | 256K | Modified MIT |"
   ],
   "access_type_evidence": [
    "# Latest open source LLM releases, 2026",
    "A running list of the open-weight models that have shipped this year, in release order, with the facts that age well: ship date, parameter count, context window, and license.",
    "| Model | Lab | Shipped | Parameters | Context | License |",
    "| Mistral Medium 3.5Dense coder. Permissive for individuals and most companies; read the revenue clause before large-scale commercial deployment. | Mistral | Apr 29, 2026 | 128B dense | 256K | Modified MIT |"
   ],
   "license_evidence": [
    "# Latest open source LLM releases, 2026",
    "| Model | Lab | Shipped | Parameters | Context | License |",
    "| Mistral Medium 3.5Dense coder. Permissive for individuals and most companies; read the revenue clause before large-scale commercial deployment. | Mistral | Apr 29, 2026 | 128B dense | 256K | Modified MIT |"
   ],
   "significance_evidence": [
    "# Latest open source LLM releases, 2026",
    "| Model | Lab | Shipped | Parameters | Context | License |",
    "| Mistral Medium 3.5Dense coder. Permissive for individuals and most companies; read the revenue clause before large-scale commercial deployment. | Mistral | Apr 29, 2026 | 128B dense | 256K | Modified MIT |"
   ]
  },
  {
   "url": "https://fazm.ai/t/latest-open-source-llm-releases-2026",
   "title": "Fazm - Claude Code & Codex GUI for macOS | Persistent Sessions",
   "published_at": "2026-06-20T17:35:27",
   "organization": "Google",
   "model": "Google Gemma 4",
   "release_date": "2026-04",
   "access_type": "open-weight",
   "license": "Apache 2.0",
   "significance": "31B dense runs on a single H100; native function calling, 100+ languages",
   "organization_evidence": [
    "# Latest open source LLM releases, 2026",
    "| Model | Lab | Shipped | Parameters | Context | License |",
    "| Google Gemma 431B dense runs on a single H100; 26B MoE gives near-4B serving cost. Native function calling, 100+ languages. | Google | Apr 2026 | 31B dense; 26B A4B MoE | 256K | Apache 2.0 |"
   ],
   "model_evidence": [
    "# Latest open source LLM releases, 2026",
    "| Model | Lab | Shipped | Parameters | Context | License |",
    "| Google Gemma 431B dense runs on a single H100; 26B MoE gives near-4B serving cost. Native function calling, 100+ languages. | Google | Apr 2026 | 31B dense; 26B A4B MoE | 256K | Apache 2.0 |"
   ],
   "release_date_evidence": [
    "# Latest open source LLM releases, 2026",
    "| Model | Lab | Shipped | Parameters | Context | License |",
    "| Google Gemma 431B dense runs on a single H100; 26B MoE gives near-4B serving cost. Native function calling, 100+ languages. | Google | Apr 2026 | 31B dense; 26B A4B MoE | 256K | Apache 2.0 |"
   ],
   "access_type_evidence": [
    "# Latest open source LLM releases, 2026",
    "A running list of the open-weight models that have shipped this year, in release order, with the facts that age well: ship date, parameter count, context window, and license.",
    "| Model | Lab | Shipped | Parameters | Context | License |",
    "| Google Gemma 431B dense runs on a single H100; 26B MoE gives near-4B serving cost. Native function calling, 100+ languages. | Google | Apr 2026 | 31B dense; 26B A4B MoE | 256K | Apache 2.0 |"
   ],
   "license_evidence": [
    "# Latest open source LLM releases, 2026",
    "| Model | Lab | Shipped | Parameters | Context | License |",
    "| Google Gemma 431B dense runs on a single H100; 26B MoE gives near-4B serving cost. Native function calling, 100+ languages. | Google | Apr 2026 | 31B dense; 26B A4B MoE | 256K | Apache 2.0 |"
   ],
   "significance_evidence": [
    "# Latest open source LLM releases, 2026",
    "| Model | Lab | Shipped | Parameters | Context | License |",
    "| Google Gemma 431B dense runs on a single H100; 26B MoE gives near-4B serving cost. Native function calling, 100+ languages. | Google | Apr 2026 | 31B dense; 26B A4B MoE | 256K | Apache 2.0 |"
   ]
  },
  {
   "url": "https://fazm.ai/t/latest-open-source-llm-releases-2026",
   "title": "Fazm - Claude Code & Codex GUI for macOS | Persistent Sessions",
   "published_at": "2026-06-20T17:35:27",
   "organization": "Alibaba",
   "model": "Qwen 3.5",
   "release_date": "2026-02",
   "access_type": "open-weight",
   "license": "Apache 2.0",
   "significance": "Hybrid Gated Delta + sparse MoE with native vision-language",
   "organization_evidence": [
    "# Latest open source LLM releases, 2026",
    "| Model | Lab | Shipped | Parameters | Context | License |",
    "| Qwen 3.5Hybrid Gated Delta + sparse MoE with native vision-language. Qwen3-Coder is SOTA among open models on agentic coding. | Alibaba | Feb 2026 | Family incl. Coder-480B-A35B | Long | Apache 2.0 |"
   ],
   "model_evidence": [
    "# Latest open source LLM releases, 2026",
    "| Model | Lab | Shipped | Parameters | Context | License |",
    "| Qwen 3.5Hybrid Gated Delta + sparse MoE with native vision-language. Qwen3-Coder is SOTA among open models on agentic coding. | Alibaba | Feb 2026 | Family incl. Coder-480B-A35B | Long | Apache 2.0 |"
   ],
   "release_date_evidence": [
    "# Latest open source LLM releases, 2026",
    "| Model | Lab | Shipped | Parameters | Context | License |",
    "| Qwen 3.5Hybrid Gated Delta + sparse MoE with native vision-language. Qwen3-Coder is SOTA among open models on agentic coding. | Alibaba | Feb 2026 | Family incl. Coder-480B-A35B | Long | Apache 2.0 |"
   ],
   "access_type_evidence": [
    "# Latest open source LLM releases, 2026",
    "A running list of the open-weight models that have shipped this year, in release order, with the facts that age well: ship date, parameter count, context window, and license.",
    "| Model | Lab | Shipped | Parameters | Context | License |",
    "| Qwen 3.5Hybrid Gated Delta + sparse MoE with native vision-language. Qwen3-Coder is SOTA among open models on agentic coding. | Alibaba | Feb 2026 | Family incl. Coder-480B-A35B | Long | Apache 2.0 |"
   ],
   "license_evidence": [
    "# Latest open source LLM releases, 2026",
    "| Model | Lab | Shipped | Parameters | Context | License |",
    "| Qwen 3.5Hybrid Gated Delta + sparse MoE with native vision-language. Qwen3-Coder is SOTA among open models on agentic coding. | Alibaba | Feb 2026 | Family incl. Coder-480B-A35B | Long | Apache 2.0 |"
   ],
   "significance_evidence": [
    "# Latest open source LLM releases, 2026",
    "| Model | Lab | Shipped | Parameters | Context | License |",
    "| Qwen 3.5Hybrid Gated Delta + sparse MoE with native vision-language. Qwen3-Coder is SOTA among open models on agentic coding. | Alibaba | Feb 2026 | Family incl. Coder-480B-A35B | Long | Apache 2.0 |"
   ]
  },
  {
   "url": "https://fazm.ai/t/latest-open-source-llm-releases-2026",
   "title": "Fazm - Claude Code & Codex GUI for macOS | Persistent Sessions",
   "published_at": "2026-06-20T17:35:27",
   "organization": "Meta",
   "model": "Llama 4 Scout",
   "release_date": "2026",
   "access_type": "open-weight",
   "license": "Meta custom (700M MAU clause)",
   "significance": "Industry-leading context window",
   "organization_evidence": [
    "# Latest open source LLM releases, 2026",
    "| Model | Lab | Shipped | Parameters | Context | License |",
    "| Llama 4 ScoutIndustry-leading context window. The license is not OSI-approved and carries a monthly-active-user ceiling. | Meta | 2026 | 109B MoE / 17B active (16 experts) | 10M | Meta custom (700M MAU clause) |"
   ],
   "model_evidence": [
    "# Latest open source LLM releases, 2026",
    "| Model | Lab | Shipped | Parameters | Context | License |",
    "| Llama 4 ScoutIndustry-leading context window. The license is not OSI-approved and carries a monthly-active-user ceiling. | Meta | 2026 | 109B MoE / 17B active (16 experts) | 10M | Meta custom (700M MAU clause) |"
   ],
   "release_date_evidence": [
    "# Latest open source LLM releases, 2026",
    "| Model | Lab | Shipped | Parameters | Context | License |",
    "| Llama 4 ScoutIndustry-leading context window. The license is not OSI-approved and carries a monthly-active-user ceiling. | Meta | 2026 | 109B MoE / 17B active (16 experts) | 10M | Meta custom (700M MAU clause) |"
   ],
   "access_type_evidence": [
    "# Latest open source LLM releases, 2026",
    "A running list of the open-weight models that have shipped this year, in release order, with the facts that age well: ship date, parameter count, context window, and license.",
    "| Model | Lab | Shipped | Parameters | Context | License |",
    "| Llama 4 ScoutIndustry-leading context window. The license is not OSI-approved and carries a monthly-active-user ceiling. | Meta | 2026 | 109B MoE / 17B active (16 experts) | 10M | Meta custom (700M MAU clause) |"
   ],
   "license_evidence": [
    "# Latest open source LLM releases, 2026",
    "| Model | Lab | Shipped | Parameters | Context | License |",
    "| Llama 4 ScoutIndustry-leading context window. The license is not OSI-approved and carries a monthly-active-user ceiling. | Meta | 2026 | 109B MoE / 17B active (16 experts) | 10M | Meta custom (700M MAU clause) |"
   ],
   "significance_evidence": [
    "# Latest open source LLM releases, 2026",
    "| Model | Lab | Shipped | Parameters | Context | License |",
    "| Llama 4 ScoutIndustry-leading context window. The license is not OSI-approved and carries a monthly-active-user ceiling. | Meta | 2026 | 109B MoE / 17B active (16 experts) | 10M | Meta custom (700M MAU clause) |"
   ]
  },
  {
   "url": "https://best-ai.news/ai-model-releases-2026",
   "title": "Latest AI Models 2026: 99 Released \u2014 Complete Catalog",
   "published_at": "2026-08-24T09:16:53",
   "organization": "Meta",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "a 30B open-weights, multimodal agentic model built for always-on local agent workflows, small enough to run on a single consumer GPU",
   "organization_evidence": [
    "Latest AI Models 2026 \u2014 99 Released",
    "## August 202621 models",
    "[by Meta](/ai-brands/meta)",
    "Meta's Muse Glimmer \u2014 a 30B open-weights, multimodal agentic model built for always-on local agent workflows, small enough to run on a single consumer GPU."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Latest AI Models 2026 \u2014 99 Released",
    "## August 202621 models",
    "[by Meta](/ai-brands/meta)",
    "Meta's Muse Glimmer \u2014 a 30B open-weights, multimodal agentic model built for always-on local agent workflows, small enough to run on a single consumer GPU."
   ]
  },
  {
   "url": "https://best-ai.news/ai-model-releases-2026",
   "title": "Latest AI Models 2026: 99 Released \u2014 Complete Catalog",
   "published_at": "2026-08-24T09:16:53",
   "organization": "NVIDIA",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "the next major generation of the Nemotron open model family, following the Nemotron 3 and Nemotron 3.5 releases, aimed at reasoning-heavy and agentic enterprise AI workloads",
   "organization_evidence": [
    "Latest AI Models 2026 \u2014 99 Released",
    "## August 202621 models",
    "[by NVIDIA](/ai-brands/nvidia)",
    "NVIDIA's Nemotron 4 \u2014 the next major generation of the Nemotron open model family, following the Nemotron 3 and Nemotron 3.5 releases, aimed at reasoning-heavy and agentic enterprise AI workloads."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Latest AI Models 2026 \u2014 99 Released",
    "## August 202621 models",
    "[by NVIDIA](/ai-brands/nvidia)",
    "NVIDIA's Nemotron 4 \u2014 the next major generation of the Nemotron open model family, following the Nemotron 3 and Nemotron 3.5 releases, aimed at reasoning-heavy and agentic enterprise AI workloads."
   ]
  },
  {
   "url": "https://best-ai.news/ai-model-releases-2026",
   "title": "Latest AI Models 2026: 99 Released \u2014 Complete Catalog",
   "published_at": "2026-08-24T09:16:53",
   "organization": "IBM",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "family of open, Apache 2.0 licensed language, vision and embedding models aimed at enterprise workloads",
   "organization_evidence": [
    "Latest AI Models 2026 \u2014 99 Released",
    "## July 202615 models",
    "[by IBM](/ai-brands/ibm)",
    "Granite is IBM's family of open, Apache 2.0 licensed language, vision and embedding models aimed at enterprise workloads."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Latest AI Models 2026 \u2014 99 Released",
    "## July 202615 models",
    "[by IBM](/ai-brands/ibm)",
    "Granite is IBM's family of open, Apache 2.0 licensed language, vision and embedding models aimed at enterprise workloads."
   ]
  },
  {
   "url": "https://best-ai.news/ai-model-releases-2026",
   "title": "Latest AI Models 2026: 99 Released \u2014 Complete Catalog",
   "published_at": "2026-08-24T09:16:53",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-01",
   "access_type": null,
   "license": null,
   "significance": "Thinking Machines Lab's first model trained from scratch: a 975B-parameter open-weights multimodal Mixture-of-Experts with 41B active parameters and controllable thinking effort",
   "organization_evidence": [
    "Latest AI Models 2026 \u2014 99 Released",
    "## July 202615 models",
    "[by Thinking Machines Lab](/ai-brands/thinking-machines)",
    "Inkling \u2014 Thinking Machines Lab's first model trained from scratch: a 975B-parameter open-weights multimodal Mixture-of-Experts with 41B active parameters and controllable thinking effort, released under Apache 2.0 in July 2026."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Latest AI Models 2026 \u2014 99 Released",
    "## July 202615 models",
    "[by Thinking Machines Lab](/ai-brands/thinking-machines)",
    "Inkling \u2014 Thinking Machines Lab's first model trained from scratch: a 975B-parameter open-weights multimodal Mixture-of-Experts with 41B active parameters and controllable thinking effort, released under Apache 2.0 in July 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Latest AI Models 2026 \u2014 99 Released",
    "## July 202615 models",
    "[by Thinking Machines Lab](/ai-brands/thinking-machines)",
    "Inkling \u2014 Thinking Machines Lab's first model trained from scratch: a 975B-parameter open-weights multimodal Mixture-of-Experts with 41B active parameters and controllable thinking effort, released under Apache 2.0 in July 2026."
   ]
  },
  {
   "url": "https://best-ai.news/ai-model-releases-2026",
   "title": "Latest AI Models 2026: 99 Released \u2014 Complete Catalog",
   "published_at": "2026-08-24T09:16:53",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "next-generation Kimi model, reported to close the gap with Anthropic's Opus 4.8 and shipped as one of the largest open-source frontier models",
   "organization_evidence": [
    "Latest AI Models 2026 \u2014 99 Released",
    "## July 202615 models",
    "[by Moonshot AI](/ai-brands/moonshot-ai)",
    "Kimi K3 \u2014 Moonshot AI's next-generation Kimi model, reported to close the gap with Anthropic's Opus 4.8 and shipped as one of the largest open-source frontier models."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Latest AI Models 2026 \u2014 99 Released",
    "## July 202615 models",
    "[by Moonshot AI](/ai-brands/moonshot-ai)",
    "Kimi K3 \u2014 Moonshot AI's next-generation Kimi model, reported to close the gap with Anthropic's Opus 4.8 and shipped as one of the largest open-source frontier models."
   ]
  },
  {
   "url": "https://best-ai.news/ai-model-releases-2026",
   "title": "Latest AI Models 2026: 99 Released \u2014 Complete Catalog",
   "published_at": "2026-08-24T09:16:53",
   "organization": "Meituan",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "1.6-trillion-parameter open-source agentic coding model with a 1M-token context window. The first trillion-parameter model claimed to be fully pre-trained AND served on domestic Chinese AI chips",
   "organization_evidence": [
    "Latest AI Models 2026 \u2014 99 Released",
    "## June 202613 models",
    "[by Meituan](/ai-brands/meituan)",
    "Meituan's 1.6-trillion-parameter open-source agentic coding model with a 1M-token context window (June 2026). The first trillion-parameter model claimed to be fully pre-trained AND served on domestic Chinese AI chips."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Latest AI Models 2026 \u2014 99 Released",
    "## June 202613 models",
    "[by Meituan](/ai-brands/meituan)",
    "Meituan's 1.6-trillion-parameter open-source agentic coding model with a 1M-token context window (June 2026). The first trillion-parameter model claimed to be fully pre-trained AND served on domestic Chinese AI chips."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Latest AI Models 2026 \u2014 99 Released",
    "## June 202613 models",
    "[by Meituan](/ai-brands/meituan)",
    "Meituan's 1.6-trillion-parameter open-source agentic coding model with a 1M-token context window (June 2026). The first trillion-parameter model claimed to be fully pre-trained AND served on domestic Chinese AI chips."
   ]
  },
  {
   "url": "https://best-ai.news/ai-model-releases-2026",
   "title": "Latest AI Models 2026: 99 Released \u2014 Complete Catalog",
   "published_at": "2026-08-24T09:16:53",
   "organization": "Zhipu AI",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "open-weight Chinese LLMs (GLM-4.5/4.6/5/5.1/5.2 plus GLM-OCR, Air, Flash and Turbo variants) known for strong coding and long-context performance under permissive (MIT) licenses",
   "organization_evidence": [
    "Latest AI Models 2026 \u2014 99 Released",
    "## June 202613 models",
    "[by Zhipu AI](/ai-brands/zhipu-ai)",
    "The GLM (General Language Model) family from Zhipu AI / Z.ai \u2014 open-weight Chinese LLMs (GLM-4.5/4.6/5/5.1/5.2 plus GLM-OCR, Air, Flash and Turbo variants) known for strong coding and long-context performance under permissive (MIT) licenses."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Latest AI Models 2026 \u2014 99 Released",
    "## June 202613 models",
    "[by Zhipu AI](/ai-brands/zhipu-ai)",
    "The GLM (General Language Model) family from Zhipu AI / Z.ai \u2014 open-weight Chinese LLMs (GLM-4.5/4.6/5/5.1/5.2 plus GLM-OCR, Air, Flash and Turbo variants) known for strong coding and long-context performance under permissive (MIT) licenses."
   ]
  },
  {
   "url": "https://best-ai.news/ai-model-releases-2026",
   "title": "Latest AI Models 2026: 99 Released \u2014 Complete Catalog",
   "published_at": "2026-08-24T09:16:53",
   "organization": "Google",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "a compact, efficient multimodal LLM designed to run on a single GPU",
   "organization_evidence": [
    "Latest AI Models 2026 \u2014 99 Released",
    "## June 202613 models",
    "[by Google](/ai-brands/google)",
    "Google's 12-billion-parameter open model in the Gemma 4 family \u2014 a compact, efficient multimodal LLM designed to run on a single GPU."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Latest AI Models 2026 \u2014 99 Released",
    "## June 202613 models",
    "[by Google](/ai-brands/google)",
    "Google's 12-billion-parameter open model in the Gemma 4 family \u2014 a compact, efficient multimodal LLM designed to run on a single GPU."
   ]
  },
  {
   "url": "https://best-ai.news/ai-model-releases-2026",
   "title": "Latest AI Models 2026: 99 Released \u2014 Complete Catalog",
   "published_at": "2026-08-24T09:16:53",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "flagship 1T-parameter open-weight LLM featuring 262K context window, long-horizon coding with up to 300 sub-agent swarms and 4,000 coordinated steps",
   "organization_evidence": [
    "Latest AI Models 2026 \u2014 99 Released",
    "## April 20265 models",
    "[by Moonshot AI](/ai-brands/moonshot-ai)",
    "Moonshot AI's flagship 1T-parameter open-weight LLM featuring 262K context window, long-horizon coding with up to 300 sub-agent swarms and 4,000 coordinated steps."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Latest AI Models 2026 \u2014 99 Released",
    "## April 20265 models",
    "[by Moonshot AI](/ai-brands/moonshot-ai)",
    "Moonshot AI's flagship 1T-parameter open-weight LLM featuring 262K context window, long-horizon coding with up to 300 sub-agent swarms and 4,000 coordinated steps."
   ]
  },
  {
   "url": "https://best-ai.news/ai-model-releases-2026",
   "title": "Latest AI Models 2026: 99 Released \u2014 Complete Catalog",
   "published_at": "2026-08-24T09:16:53",
   "organization": "Google",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Available in 4 sizes (E2B, E4B, 26B MoE, 31B Dense) with advanced reasoning, agentic workflows, vision, audio, 256K context, 140+ languages",
   "organization_evidence": [
    "Latest AI Models 2026 \u2014 99 Released",
    "## April 20265 models",
    "[by Google](/ai-brands/google)",
    "Google DeepMind's most capable open model family. Available in 4 sizes (E2B, E4B, 26B MoE, 31B Dense) with advanced reasoning, agentic workflows, vision, audio, 256K context, 140+ languages. Apache 2.0 license. Runs on devices from phones to H100 GPUs."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Latest AI Models 2026 \u2014 99 Released",
    "## April 20265 models",
    "[by Google](/ai-brands/google)",
    "Google DeepMind's most capable open model family. Available in 4 sizes (E2B, E4B, 26B MoE, 31B Dense) with advanced reasoning, agentic workflows, vision, audio, 256K context, 140+ languages. Apache 2.0 license. Runs on devices from phones to H100 GPUs."
   ]
  },
  {
   "url": "https://openrouter.ai/blog/insights/the-open-weight-models-that-matter-june-2026/",
   "title": "The Open Weight Models that Matter: June 2026",
   "published_at": "2026-06-27T00:00:00",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "The first open-weight model that teams immediately dropped into real agentic pipelines as a plausible substitute for an Anthropic- or OpenAI-class frontier model.",
   "organization_evidence": [
    "The Open Weight Models that Matter: June 2026",
    "## 1. [DeepSeek V4 Flash](https://openrouter.ai/deepseek/deepseek-v4-flash) \u2014 The first to cross the agentic rubicon",
    "Adoption has also been driven largely by price; DeepSeek\u2019s first-party API lists Flash at $0.14 / $0.28 per million tokens (in/out), but **does** retain data for training purposes."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The Open Weight Models that Matter: June 2026",
    "## 1. [DeepSeek V4 Flash](https://openrouter.ai/deepseek/deepseek-v4-flash) \u2014 The first to cross the agentic rubicon",
    "The model was released April 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The Open Weight Models that Matter: June 2026",
    "## 1. [DeepSeek V4 Flash](https://openrouter.ai/deepseek/deepseek-v4-flash) \u2014 The first to cross the agentic rubicon",
    "DeepSeek V4 Flash is the first open-weight model that teams immediately dropped into real agentic pipelines as a plausible substitute for an Anthropic- or OpenAI-class frontier model."
   ]
  },
  {
   "url": "https://openrouter.ai/blog/insights/the-open-weight-models-that-matter-june-2026/",
   "title": "The Open Weight Models that Matter: June 2026",
   "published_at": "2026-06-27T00:00:00",
   "organization": "GLM",
   "model": null,
   "release_date": "2026-06",
   "access_type": null,
   "license": null,
   "significance": "Breaking through on planning quality and long-horizon coding.",
   "organization_evidence": [
    "The Open Weight Models that Matter: June 2026",
    "## 2. [GLM 5.2](https://openrouter.ai/z-ai/glm-5.2) \u2014 The open model that makes Opus-style agentic coding portable",
    "GLM 5.2 was released in mid-June but the early read is very strong."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The Open Weight Models that Matter: June 2026",
    "## 2. [GLM 5.2](https://openrouter.ai/z-ai/glm-5.2) \u2014 The open model that makes Opus-style agentic coding portable",
    "GLM 5.2 was released in mid-June but the early read is very strong."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The Open Weight Models that Matter: June 2026",
    "## 2. [GLM 5.2](https://openrouter.ai/z-ai/glm-5.2) \u2014 The open model that makes Opus-style agentic coding portable",
    "Where DeepSeek broke through on price, GLM 5.2 appears to be breaking through on planning quality and long-horizon coding."
   ]
  },
  {
   "url": "https://campustechnology.com/articles/2026/08/26/meta-releases-muse-glimmer-an-open-weight-ai-model-designed-to-run-on-consumer-hardware.aspx",
   "title": "Meta Releases Muse Glimmer, an Open-Weight AI Model Designed to Run on Consumer Hardware -- Campus Technology",
   "published_at": "2026-08-26T00:00:00",
   "organization": "Meta",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "a 30-billion-parameter, open-weight model optimized for local, always-on AI agents",
   "organization_evidence": [
    "### Meta Releases Muse Glimmer, an Open-Weight AI Model Designed to Run on Consumer Hardware",
    "Meta describes Glimmer as a 30-billion-parameter, open-weight model optimized for local, always-on AI agents."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "### Meta Releases Muse Glimmer, an Open-Weight AI Model Designed to Run on Consumer Hardware",
    "Meta describes Glimmer as a 30-billion-parameter, open-weight model optimized for local, always-on AI agents."
   ]
  },
  {
   "url": "https://fazm.ai/blog/new-llm-releases-april-2026",
   "title": "New LLM Releases April 2026: Every Major Model Launch This Month",
   "published_at": "2026-04-11T00:00:00",
   "organization": "Google",
   "model": null,
   "release_date": "2026-04-02",
   "access_type": null,
   "license": null,
   "significance": "Four Gemma 4 variants under Apache 2.0 with benchmark scores outperforming models 20 times its size, supporting 256K context, native vision and audio processing, and over 140 languages",
   "organization_evidence": [
    "Roundup, April 2026",
    "Google shipped four Gemma 4 variants under Apache 2.0."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Roundup, April 2026",
    "Google released the Gemma 4 family on April 2 under Apache 2.0, delivering four models purpose-built for different deployment scenarios:"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Roundup, April 2026",
    "Google shipped four Gemma 4 variants under Apache 2.0.",
    "Google released the Gemma 4 family on April 2 under Apache 2.0, delivering four models purpose-built for different deployment scenarios:",
    "- **Gemma 4 31B Dense** the flagship, with benchmark scores that outperform models 20 times its size",
    "- **Gemma 4 26B MoE** mixture-of-experts variant for efficient inference",
    "- **Gemma 4 E4B** consumer GPU and edge deployment",
    "- **Gemma 4 E2B** smartphones and Raspberry Pi devices",
    "All four models support 256K context windows, native vision and audio processing, and fluency in over 140 languages. They are purpose-built for advanced reasoning and agentic workflows."
   ]
  },
  {
   "url": "https://fazm.ai/blog/new-llm-releases-april-2026",
   "title": "New LLM Releases April 2026: Every Major Model Launch This Month",
   "published_at": "2026-04-11T00:00:00",
   "organization": "Zhipu AI",
   "model": null,
   "release_date": "2026-04-01",
   "access_type": null,
   "license": null,
   "significance": "744B MoE parameter model (40B active), 200K context window, beating Claude Opus 4.6 and GPT-5.4 on SWE-Bench Pro",
   "organization_evidence": [
    "Roundup, April 2026",
    "Zhipu published GLM-5.1 under MIT."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Roundup, April 2026",
    "## Quick Reference: April 2026 LLM Releases",
    "| Model | Company | Release Date | Parameters | Context | License | Pricing /1M tokens |",
    "| GLM-5.1 | Zhipu AI | Early Apr | 744B MoE (40B active) | 200K | MIT | Free (open weights) |",
    "Zhipu AI released GLM-5.1 under the MIT license, a 744-billion parameter mixture-of-experts model with 40 billion parameters active per forward pass and a 200K context window."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Roundup, April 2026",
    "## Quick Reference: April 2026 LLM Releases",
    "| Model | Company | Release Date | Parameters | Context | License | Pricing /1M tokens |",
    "| GLM-5.1 | Zhipu AI | Early Apr | 744B MoE (40B active) | 200K | MIT | Free (open weights) |",
    "Zhipu AI released GLM-5.1 under the MIT license, a 744-billion parameter mixture-of-experts model with 40 billion parameters active per forward pass and a 200K context window.",
    "The headline claim: on SWE-Bench Pro, GLM-5.1 reportedly beat both Claude Opus 4.6 and GPT-5.4."
   ]
  },
  {
   "url": "https://fazm.ai/blog/new-llm-releases-april-2026",
   "title": "New LLM Releases April 2026: Every Major Model Launch This Month",
   "published_at": "2026-04-11T00:00:00",
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-04-01",
   "access_type": null,
   "license": null,
   "significance": "1 million token context window targeting agentic coding workflows",
   "organization_evidence": [
    "Roundup, April 2026",
    "Alibaba shipped Qwen 3.6-Plus."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Roundup, April 2026",
    "## Quick Reference: April 2026 LLM Releases",
    "| Model | Company | Release Date | Parameters | Context | License | Pricing /1M tokens |",
    "| Qwen 3.6-Plus | Alibaba | Early Apr | Undisclosed | 1M | Open | Free (open weights) |",
    "Alibaba's Qwen 3.6-Plus targets agentic coding workflows with a 1 million token context window."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Roundup, April 2026",
    "Alibaba shipped Qwen 3.6-Plus.",
    "## Quick Reference: April 2026 LLM Releases",
    "| Model | Company | Release Date | Parameters | Context | License | Pricing /1M tokens |",
    "| Qwen 3.6-Plus | Alibaba | Early Apr | Undisclosed | 1M | Open | Free (open weights) |",
    "Alibaba's Qwen 3.6-Plus targets agentic coding workflows with a 1 million token context window."
   ]
  },
  {
   "url": "https://fazm.ai/blog/new-llm-releases-april-2026",
   "title": "New LLM Releases April 2026: Every Major Model Launch This Month",
   "published_at": "2026-04-11T00:00:00",
   "organization": "Meta",
   "model": null,
   "release_date": "2026-04-01",
   "access_type": null,
   "license": null,
   "significance": "Mixture-of-experts models with native multimodal capabilities and a 10M token context window on Llama 4 Scout",
   "organization_evidence": [
    "Roundup, April 2026",
    "Meta shipped Llama 4 Scout and Maverick."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Roundup, April 2026",
    "## Quick Reference: April 2026 LLM Releases",
    "| Model | Company | Release Date | Parameters | Context | License | Pricing /1M tokens |",
    "| Llama 4 Scout | Meta | Apr (rolling) | Undisclosed | 10M | Llama License | Free (open weights) |",
    "Meta's Llama 4 family includes two headline models:"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Roundup, April 2026",
    "Meta shipped Llama 4 Scout and Maverick.",
    "## Quick Reference: April 2026 LLM Releases",
    "| Model | Company | Release Date | Parameters | Context | License | Pricing /1M tokens |",
    "| Llama 4 Scout | Meta | Apr (rolling) | Undisclosed | 10M | Llama License | Free (open weights) |",
    "Meta's Llama 4 family includes two headline models:",
    "- **Llama 4 Scout** with a 10 million token context window, the largest of any model released this month",
    "- **Llama 4 Maverick** with 400 billion parameters, 1 million token context, and native multimodal capabilities",
    "Both models use a mixture-of-experts architecture and are natively multimodal from training (not bolted-on vision after the fact)."
   ]
  },
  {
   "url": "https://fazm.ai/blog/new-llm-releases-april-2026",
   "title": "New LLM Releases April 2026: Every Major Model Launch This Month",
   "published_at": "2026-04-11T00:00:00",
   "organization": "Arcee AI",
   "model": null,
   "release_date": "2026-04-01",
   "access_type": null,
   "license": null,
   "significance": "400 billion parameter model designed for enterprise use cases under Apache 2.0",
   "organization_evidence": [
    "Roundup, April 2026",
    "Arcee dropped Trinity at 400B parameters."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Roundup, April 2026",
    "## Quick Reference: April 2026 LLM Releases",
    "| Model | Company | Release Date | Parameters | Context | License | Pricing /1M tokens |",
    "| Arcee Trinity | Arcee AI | Early Apr | 400B | TBA | Apache 2.0 | Free (open weights) |",
    "Arcee AI released Trinity, a 400 billion parameter model under Apache 2.0."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Roundup, April 2026",
    "Arcee dropped Trinity at 400B parameters.",
    "## Quick Reference: April 2026 LLM Releases",
    "| Model | Company | Release Date | Parameters | Context | License | Pricing /1M tokens |",
    "| Arcee Trinity | Arcee AI | Early Apr | 400B | TBA | Apache 2.0 | Free (open weights) |",
    "Arcee AI released Trinity, a 400 billion parameter model under Apache 2.0."
   ]
  },
  {
   "url": "https://edgen.tech/tr/news/post/kimi-k3-open-weight-release-draws-100k-downloads-erases-us-ai-lead",
   "title": "Kimi K3 open-weight release draws 100k downloads, erases US AI lead",
   "published_at": "2026-08-17T03:11:14",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-27",
   "access_type": null,
   "license": null,
   "significance": "matched or surpassed Anthropic's Fable 5 on key benchmarks at roughly half the cost, erasing the closed-source lead",
   "organization_evidence": [
    "Kimi K3 open-weight release draws 100k downloads, erases US AI lead",
    "Moonshot AI's Kimi K3, a 2.8-trillion-parameter open-weight model, matched or surpassed Anthropic's Fable 5 on key benchmarks at roughly half the cost, erasing the closed-source lead."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Kimi K3 open-weight release draws 100k downloads, erases US AI lead",
    "The model's July 27 release on Hugging Face drew roughly 100,000 downloads in its first 24 hours, and demand was so high that Moonshot temporarily blocked new consumer sign-ups."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Kimi K3 open-weight release draws 100k downloads, erases US AI lead",
    "Moonshot AI's Kimi K3, a 2.8-trillion-parameter open-weight model, matched or surpassed Anthropic's Fable 5 on key benchmarks at roughly half the cost, erasing the closed-source lead."
   ]
  },
  {
   "url": "https://rits.shanghai.nyu.edu/ai/poolside-releases-laguna-s-2-1-a-118b-open-weight-coding-model",
   "title": "Poolside Releases Laguna S 2.1, a 118B Open-Weight Coding Model",
   "published_at": null,
   "organization": "Poolside",
   "model": "Laguna S 2.1",
   "release_date": "2026-07-21",
   "access_type": "open-weight",
   "license": "OpenMDW-1.1",
   "significance": "a 118-billion-parameter Mixture-of-Experts model built for agentic coding that activates only 8 billion parameters per token",
   "organization_evidence": [
    "# Poolside Releases Laguna S 2.1, a 118B Open-Weight Coding Model",
    "**On July 21, 2026, San Francisco-based Poolside released Laguna S 2.1** \u2014 a 118-billion-parameter Mixture-of-Experts model built for agentic coding that activates only 8 billion parameters per token."
   ],
   "model_evidence": [
    "# Poolside Releases Laguna S 2.1, a 118B Open-Weight Coding Model",
    "**On July 21, 2026, San Francisco-based Poolside released Laguna S 2.1** \u2014 a 118-billion-parameter Mixture-of-Experts model built for agentic coding that activates only 8 billion parameters per token."
   ],
   "release_date_evidence": [
    "# Poolside Releases Laguna S 2.1, a 118B Open-Weight Coding Model",
    "**On July 21, 2026, San Francisco-based Poolside released Laguna S 2.1** \u2014 a 118-billion-parameter Mixture-of-Experts model built for agentic coding that activates only 8 billion parameters per token."
   ],
   "access_type_evidence": [
    "# Poolside Releases Laguna S 2.1, a 118B Open-Weight Coding Model",
    "The weights are on Hugging Face under the permissive OpenMDW-1.1 license."
   ],
   "license_evidence": [
    "# Poolside Releases Laguna S 2.1, a 118B Open-Weight Coding Model",
    "The weights are on Hugging Face under the permissive OpenMDW-1.1 license."
   ],
   "significance_evidence": [
    "# Poolside Releases Laguna S 2.1, a 118B Open-Weight Coding Model",
    "**On July 21, 2026, San Francisco-based Poolside released Laguna S 2.1** \u2014 a 118-billion-parameter Mixture-of-Experts model built for agentic coding that activates only 8 billion parameters per token."
   ]
  },
  {
   "url": "https://llm-releases.com",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Tencent Hunyuan",
   "model": null,
   "release_date": "2026-08-20",
   "access_type": null,
   "license": null,
   "significance": "outperforming open heavyweights such as DeepSeek-V4-Pro and Kimi K2.6 on translation quality",
   "organization_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Hy-MT2-30B-A3B\nAvailable\nTencent HunyuanOpen weights\nThe flagship of Tencent Hunyuan's Hy-MT2 family of 'fast-thinking' multilingual machine-translation models, open-weighted on Hugging Face on Aug 20 2026."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Hy-MT2-30B-A3B\nAvailable\nTencent HunyuanOpen weights\nThe flagship of Tencent Hunyuan's Hy-MT2 family of 'fast-thinking' multilingual machine-translation models, open-weighted on Hugging Face on Aug 20 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "Tencent reports it outperforming open heavyweights such as DeepSeek-V4-Pro and Kimi K2.6 on translation quality, with even the smaller 1.8B sibling (Hy-MT2-1.8B, released the same day) beating commercial APIs from Microsoft and Doubao \u2014 vendor figures, unverified at launch."
   ]
  },
  {
   "url": "https://llm-releases.com",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "DeepReinforce (Ornith)",
   "model": null,
   "release_date": "2026-08-19",
   "access_type": null,
   "license": null,
   "significance": "a 9B-parameter dense coding/agent model trained with the family's self-improving task-and-scaffold RL loop",
   "organization_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Ornith-1.5-9B\nAvailable\nDeepReinforce (Ornith)Open weights\nThe smallest model in DeepReinforce's Ornith-1.5 family (released 2026-08-19, MIT, weights on Hugging Face): a 9B-parameter dense coding/agent model trained with the family's self-improving task-and-scaffold RL loop, and shipped with a quantized 'Ornith-1.5-9B-Mobile' build that runs on iPhone and Android."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Ornith-1.5-9B\nAvailable\nDeepReinforce (Ornith)Open weights\nThe smallest model in DeepReinforce's Ornith-1.5 family (released 2026-08-19, MIT, weights on Hugging Face): a 9B-parameter dense coding/agent model trained with the family's self-improving task-and-scaffold RL loop, and shipped with a quantized 'Ornith-1.5-9B-Mobile' build that runs on iPhone and Android."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Ornith-1.5-9B\nAvailable\nDeepReinforce (Ornith)Open weights\nThe smallest model in DeepReinforce's Ornith-1.5 family (released 2026-08-19, MIT, weights on Hugging Face): a 9B-parameter dense coding/agent model trained with the family's self-improving task-and-scaffold RL loop, and shipped with a quantized 'Ornith-1.5-9B-Mobile' build that runs on iPhone and Android."
   ]
  },
  {
   "url": "https://llm-releases.com",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "DeepReinforce (Ornith)",
   "model": null,
   "release_date": "2026-08-19",
   "access_type": null,
   "license": null,
   "significance": "a 35B-parameter Mixture-of-Experts that activates ~3B parameters per token",
   "organization_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Ornith-1.5-35B-A3B\nAvailable\nDeepReinforce (Ornith)Open weights\nThe mid-size model in DeepReinforce's Ornith-1.5 family (released 2026-08-19, MIT, weights on Hugging Face): a 35B-parameter Mixture-of-Experts that activates ~3B parameters per token, trained with the same self-improving task-and-scaffold generation loop as the 397B flagship."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Ornith-1.5-35B-A3B\nAvailable\nDeepReinforce (Ornith)Open weights\nThe mid-size model in DeepReinforce's Ornith-1.5 family (released 2026-08-19, MIT, weights on Hugging Face): a 35B-parameter Mixture-of-Experts that activates ~3B parameters per token, trained with the same self-improving task-and-scaffold generation loop as the 397B flagship."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Ornith-1.5-35B-A3B\nAvailable\nDeepReinforce (Ornith)Open weights\nThe mid-size model in DeepReinforce's Ornith-1.5 family (released 2026-08-19, MIT, weights on Hugging Face): a 35B-parameter Mixture-of-Experts that activates ~3B parameters per token, trained with the same self-improving task-and-scaffold generation loop as the 397B flagship."
   ]
  },
  {
   "url": "https://llm-releases.com",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "DeepReinforce (Ornith)",
   "model": null,
   "release_date": "2026-08-19",
   "access_type": null,
   "license": null,
   "significance": "a ~397B-parameter Mixture-of-Experts coding/agent model trained with a self-improving RL loop",
   "organization_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Ornith-1.5-397B\nAvailable\nDeepReinforce (Ornith)FrontierOpen weights\nThe flagship of DeepReinforce's Ornith-1.5 family, released 2026-08-19 under the MIT license with weights on Hugging Face."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Ornith-1.5-397B\nAvailable\nDeepReinforce (Ornith)FrontierOpen weights\nThe flagship of DeepReinforce's Ornith-1.5 family, released 2026-08-19 under the MIT license with weights on Hugging Face."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "A ~397B-parameter Mixture-of-Experts coding/agent model (per-token active count not disclosed) trained with a self-improving RL loop: rather than fixed human-curated tasks, the system proposes progressively harder tasks itself, generates a task-specific orchestration scaffold for each, and produces the solution rollouts used for reinforcement learning, with reward propagating across all three stages (all optimized with GRPO)."
   ]
  },
  {
   "url": "https://llm-releases.com",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Alibaba (Qwen)",
   "model": null,
   "release_date": "2026-08-14",
   "access_type": null,
   "license": null,
   "significance": "A 27B dense model (~28B counting the ~1B vision encoder) with 64 layers, hidden size 5,120, and a 248,320-token vocabulary",
   "organization_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Qwen3.8-27B\nAvailable\nAlibaba (Qwen)Open weights\nThe open-weight, single-GPU sibling of Qwen3.8-Max, published by Alibaba on Hugging Face on Aug 14 2026 under Apache 2.0 \u2014 the smaller open release Alibaba had promised alongside the closed Qwen3.8-Max flagship."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Qwen3.8-27B\nAvailable\nAlibaba (Qwen)Open weights\nThe open-weight, single-GPU sibling of Qwen3.8-Max, published by Alibaba on Hugging Face on Aug 14 2026 under Apache 2.0 \u2014 the smaller open release Alibaba had promised alongside the closed Qwen3.8-Max flagship."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "A 27B dense model (~28B counting the ~1B vision encoder) with 64 layers, hidden size 5,120, and a 248,320-token vocabulary."
   ]
  },
  {
   "url": "https://llm-releases.com",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Dots Studio (rednote-hilab)",
   "model": null,
   "release_date": "2026-08-14",
   "access_type": null,
   "license": null,
   "significance": "A Mixture-of-Experts model with ~280B total parameters and ~16B active per token",
   "organization_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Dots3-Note Preview\nPreview\nDots Studio (rednote-hilab)Open weights\nThe first open-weight release in the dots3 series from Dots Studio (rednote-hilab), the AI lab of Xiaohongshu (RedNote), open-weighted on Hugging Face on Aug 14 2026 under Apache-2.0."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Dots3-Note Preview\nPreview\nDots Studio (rednote-hilab)Open weights\nThe first open-weight release in the dots3 series from Dots Studio (rednote-hilab), the AI lab of Xiaohongshu (RedNote), open-weighted on Hugging Face on Aug 14 2026 under Apache-2.0."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "A Mixture-of-Experts model with ~280B total parameters and ~16B active per token, carrying a 512K-token context and multimodal understanding across text, vision, and audio (text output)."
   ]
  },
  {
   "url": "https://llm-releases.com",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Cohere",
   "model": null,
   "release_date": "2026-08-12",
   "access_type": null,
   "license": null,
   "significance": "A 2.4B-parameter VLM combining a custom 400M native-resolution vision encoder, a 2B language model on the Command A+ architecture, and a projector",
   "organization_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### North Micro Vision Instruct\nAvailable\nCohereOpen source\nCohere's compact document-focused vision-language model, published Aug 12 2026 under Apache 2.0."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### North Micro Vision Instruct\nAvailable\nCohereOpen source\nCohere's compact document-focused vision-language model, published Aug 12 2026 under Apache 2.0."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "A 2.4B-parameter VLM combining a custom 400M native-resolution vision encoder, a 2B language model on the Command A+ architecture, and a projector; it preserves aspect ratio for images up to 1654x2339px (an A4 page at 200 dpi)."
   ]
  },
  {
   "url": "https://llm-releases.com",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Liquid AI",
   "model": null,
   "release_date": "2026-08-12",
   "access_type": null,
   "license": null,
   "significance": "a 3.1B-parameter VLM built on the LFM2.5-2.6B text base with an integrated SigLIP2 400M NaFlex vision encoder",
   "organization_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### LFM2.5-VL-3B\nAvailable\nLiquid AIOpen weights\nLiquid AI's edge vision-language model, released Aug 12 2026 \u2014 a 3.1B-parameter VLM built on the LFM2.5-2.6B text base with an integrated SigLIP2 400M NaFlex vision encoder."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### LFM2.5-VL-3B\nAvailable\nLiquid AIOpen weights\nLiquid AI's edge vision-language model, released Aug 12 2026 \u2014 a 3.1B-parameter VLM built on the LFM2.5-2.6B text base with an integrated SigLIP2 400M NaFlex vision encoder."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### LFM2.5-VL-3B\nAvailable\nLiquid AIOpen weights\nLiquid AI's edge vision-language model, released Aug 12 2026 \u2014 a 3.1B-parameter VLM built on the LFM2.5-2.6B text base with an integrated SigLIP2 400M NaFlex vision encoder."
   ]
  },
  {
   "url": "https://llm-releases.com",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "NVIDIA",
   "model": null,
   "release_date": "2026-08-11",
   "access_type": null,
   "license": null,
   "significance": "A hybrid Mamba-2 + MoE + Attention design with ~31.6B total and ~3.6B active parameters and a 1M-token context",
   "organization_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Nemotron 3.5 Lightning\nAvailable\nNVIDIAOpen weights\nNVIDIA's efficient open Mixture-of-Experts model, released Aug 11 2026 for long-running agents."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Nemotron 3.5 Lightning\nAvailable\nNVIDIAOpen weights\nNVIDIA's efficient open Mixture-of-Experts model, released Aug 11 2026 for long-running agents."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "A hybrid Mamba-2 + MoE + Attention design with ~31.6B total and ~3.6B active parameters and a 1M-token context, shipped alongside the NeMo Switchyard model router."
   ]
  },
  {
   "url": "https://llm-releases.com",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Meta AI",
   "model": null,
   "release_date": "2026-08-10",
   "access_type": null,
   "license": null,
   "significance": "A ~30B-parameter dense causal transformer (about 29.6B parameters across 52 layers) paired with a ~1.8B ViT-G/14 perception encoder",
   "organization_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Muse Glimmer\nAvailable\nMeta AIOpen weights\nMeta's first open-weight agentic model, released Aug 10, 2026 under an Apache 2.0 license \u2014 Meta's return to open weights after the closed Muse Spark line."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Muse Glimmer\nAvailable\nMeta AIOpen weights\nMeta's first open-weight agentic model, released Aug 10, 2026 under an Apache 2.0 license \u2014 Meta's return to open weights after the closed Muse Spark line."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "A ~30B-parameter dense causal transformer (about 29.6B parameters across 52 layers) paired with a ~1.8B ViT-G/14 perception encoder, so it accepts interleaved text and images and returns text across more than 100 languages."
   ]
  },
  {
   "url": "https://llm-releases.com",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Liquid AI",
   "model": null,
   "release_date": "2026-08-06",
   "access_type": null,
   "license": null,
   "significance": "a 2.69B-parameter dense model, distinct from the LFM2.5-8B-A1B MoE",
   "organization_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### LFM2.5-2.6B\nAvailable\nLiquid AIOpen weights\nLiquid AI's on-device agentic model, released Aug 6 2026 (surfaced on hosted platforms ~Aug 11) \u2014 a 2.69B-parameter dense model, distinct from the LFM2.5-8B-A1B MoE."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### LFM2.5-2.6B\nAvailable\nLiquid AIOpen weights\nLiquid AI's on-device agentic model, released Aug 6 2026 (surfaced on hosted platforms ~Aug 11) \u2014 a 2.69B-parameter dense model, distinct from the LFM2.5-8B-A1B MoE."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### LFM2.5-2.6B\nAvailable\nLiquid AIOpen weights\nLiquid AI's on-device agentic model, released Aug 6 2026 (surfaced on hosted platforms ~Aug 11) \u2014 a 2.69B-parameter dense model, distinct from the LFM2.5-8B-A1B MoE."
   ]
  },
  {
   "url": "https://llm-releases.com",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Ant Group (inclusionAI)",
   "model": null,
   "release_date": "2026-08-06",
   "access_type": null,
   "license": null,
   "significance": "A sparse Mixture-of-Experts model with 7.9B total parameters and only ~1.3B active per token",
   "organization_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Ling-3.0-tiny\nAvailable\nAnt Group (inclusionAI)Open weights\nThe smallest member of Ant Group inclusionAI's Ling 3.0 family, open-weighted on Hugging Face on Aug 6 2026 under the MIT license \u2014 distinct from the (API-only at launch) Ling-3.0-flash."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Ling-3.0-tiny\nAvailable\nAnt Group (inclusionAI)Open weights\nThe smallest member of Ant Group inclusionAI's Ling 3.0 family, open-weighted on Hugging Face on Aug 6 2026 under the MIT license \u2014 distinct from the (API-only at launch) Ling-3.0-flash."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "A sparse Mixture-of-Experts model with 7.9B total parameters and only ~1.3B active per token: 128 routed experts with 8 routed plus 1 shared expert active per token, using the same 3:1 alternating stack of Kimi Delta Attention (KDA, linear) and Multi-head Latent Attention (MLA) layers as the rest of the family, for a 262,144-token (256K) context."
   ]
  },
  {
   "url": "https://llm-releases.com",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Poolside",
   "model": null,
   "release_date": "2026-05-22",
   "access_type": null,
   "license": null,
   "significance": "a 118B-total / 8B-active Mixture-of-Experts that activates only ~6.8% of its parameters per token",
   "organization_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Laguna S 2.1\nAvailable\nPoolsideOpen source\nPoolside's open-weight agentic-coding model and a scale-up of the Laguna XS family (same pre-training data as XS 2.1): a 118B-total / 8B-active Mixture-of-Experts that activates only ~6.8% of its parameters per token, giving larger-model behavior while staying cheap to serve, with a 1M-token context in both thinking and no-thinking modes."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "Trained in under nine weeks on 4,096 NVIDIA H200 GPUs (pre-training began 22 May 2026); first Poolside model with RL in FP8."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Laguna S 2.1\nAvailable\nPoolsideOpen source\nPoolside's open-weight agentic-coding model and a scale-up of the Laguna XS family (same pre-training data as XS 2.1): a 118B-total / 8B-active Mixture-of-Experts that activates only ~6.8% of its parameters per token, giving larger-model behavior while staying cheap to serve, with a 1M-token context in both thinking and no-thinking modes."
   ]
  },
  {
   "url": "https://llm-releases.com",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-16",
   "access_type": null,
   "license": null,
   "significance": "a 2.8T-parameter MoE (896 experts, 16 active per token) using Kimi Delta Attention and Attention Residuals",
   "organization_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Kimi K3\nAvailable\nMoonshot AIFrontierOpen weights\nMoonshot's flagship open-weight agentic model and the largest open model released to date: a 2.8T-parameter MoE (896 experts, 16 active per token) using Kimi Delta Attention and Attention Residuals, with native multimodal input and a 1M-token context."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "Launched via API on Jul 16, 2026 at $3/$15 per Mtok (cached input $0.30); full open weights published to Hugging Face on Jul 26, 2026 \u2014 a day ahead of the announced Jul 27 target \u2014 under a Modified MIT license, making it freely downloadable and self-hostable."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Kimi K3\nAvailable\nMoonshot AIFrontierOpen weights\nMoonshot's flagship open-weight agentic model and the largest open model released to date: a 2.8T-parameter MoE (896 experts, 16 active per token) using Kimi Delta Attention and Attention Residuals, with native multimodal input and a 1M-token context."
   ]
  },
  {
   "url": "https://llm-releases.com",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "a natively multimodal Mixture-of-Experts with 975B total / 41B active parameters that reasons across text, image, and audio inputs and emits text",
   "organization_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Inkling\nAvailable\nThinking Machines LabFrontierOpen source\nThinking Machines Lab's first model and the leading U.S. open-weights release: a natively multimodal Mixture-of-Experts with 975B total / 41B active parameters that reasons across text, image, and audio inputs and emits text."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Inkling\nAvailable\nThinking Machines LabFrontierOpen source\nThinking Machines Lab's first model and the leading U.S. open-weights release: a natively multimodal Mixture-of-Experts with 975B total / 41B active parameters that reasons across text, image, and audio inputs and emits text."
   ]
  },
  {
   "url": "https://llm-releases.com",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Tencent Hunyuan",
   "model": null,
   "release_date": "2026-07-06",
   "access_type": null,
   "license": null,
   "significance": "A 295B-total / 21B-active Transformer MoE with an additional 3.8B multi-token-prediction (MTP) layer and a 256K-token context",
   "organization_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Hunyuan Hy3\nAvailable\nTencent HunyuanFrontierOpen source\nThe general-availability release of Tencent's third-generation Hunyuan (Hunyuan 3.0), officially launched and open-sourced on July 6, 2026 after April's \"Hy3 preview\"."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Hunyuan Hy3\nAvailable\nTencent HunyuanFrontierOpen source\nThe general-availability release of Tencent's third-generation Hunyuan (Hunyuan 3.0), officially launched and open-sourced on July 6, 2026 after April's \"Hy3 preview\"."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "A 295B-total / 21B-active Transformer MoE with an additional 3.8B multi-token-prediction (MTP) layer and a 256K-token context, offering three selectable inference modes that blend fast and slow thinking."
   ]
  },
  {
   "url": "https://llm-releases.com",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "NVIDIA",
   "model": null,
   "release_date": "2026-07-06",
   "access_type": null,
   "license": null,
   "significance": "a compressed variant of Nemotron-3-Super-120B-A12B produced with \"Iterative Puzzle\"",
   "organization_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Nemotron-Labs-3-Puzzle-75B-A9B\nAvailable\nNVIDIAOpen weights\nA deployment-optimized open-weight model from NVIDIA, released July 6, 2026 \u2014 a compressed variant of Nemotron-3-Super-120B-A12B produced with \"Iterative Puzzle\", a post-training compression framework that jointly prunes MoE experts, active-parameter budget, and Mamba state to boost inference efficiency while preserving accuracy."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Nemotron-Labs-3-Puzzle-75B-A9B\nAvailable\nNVIDIAOpen weights\nA deployment-optimized open-weight model from NVIDIA, released July 6, 2026 \u2014 a compressed variant of Nemotron-3-Super-120B-A12B produced with \"Iterative Puzzle\", a post-training compression framework that jointly prunes MoE experts, active-parameter budget, and Mamba state to boost inference efficiency while preserving accuracy."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Nemotron-Labs-3-Puzzle-75B-A9B\nAvailable\nNVIDIAOpen weights\nA deployment-optimized open-weight model from NVIDIA, released July 6, 2026 \u2014 a compressed variant of Nemotron-3-Super-120B-A12B produced with \"Iterative Puzzle\", a post-training compression framework that jointly prunes MoE experts, active-parameter budget, and Mamba state to boost inference efficiency while preserving accuracy."
   ]
  },
  {
   "url": "https://llm-releases.com",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Poolside",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "a 33B-total / 3B-active Mixture-of-Experts built for agentic coding and long-horizon work on a local machine",
   "organization_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Laguna XS 2.1\nAvailable\nPoolsideOpen source\nPoolside's open-weight small coding model: a 33B-total / 3B-active Mixture-of-Experts built for agentic coding and long-horizon work on a local machine, served at 256K context."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Laguna XS 2.1\nAvailable\nPoolsideOpen source\nPoolside's open-weight small coding model: a 33B-total / 3B-active Mixture-of-Experts built for agentic coding and long-horizon work on a local machine, served at 256K context."
   ]
  },
  {
   "url": "https://llm-releases.com",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Meituan (LongCat)",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "a 1.6-trillion-parameter Mixture-of-Experts model (~48B active per token, dynamically routed between ~33B and ~56B) with a 1M-token context, built for agentic coding",
   "organization_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### LongCat-2.0\nAvailable\nMeituan (LongCat)FrontierOpen source\nMeituan's open-weight flagship: a 1.6-trillion-parameter Mixture-of-Experts model (~48B active per token, dynamically routed between ~33B and ~56B) with a 1M-token context, built for agentic coding."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### LongCat-2.0\nAvailable\nMeituan (LongCat)FrontierOpen source\nMeituan's open-weight flagship: a 1.6-trillion-parameter Mixture-of-Experts model (~48B active per token, dynamically routed between ~33B and ~56B) with a 1M-token context, built for agentic coding."
   ]
  },
  {
   "url": "https://llm-releases.com",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Google DeepMind",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "An open-weight text-diffusion model built on the Gemma 4 26B-A4B MoE backbone (25.2B total / 3.8B active)",
   "organization_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### DiffusionGemma 26B-A4B\nAvailable\nGoogle DeepMindOpen source\nAn open-weight text-diffusion model built on the Gemma 4 26B-A4B MoE backbone (25.2B total / 3.8B active)."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### DiffusionGemma 26B-A4B\nAvailable\nGoogle DeepMindOpen source\nAn open-weight text-diffusion model built on the Gemma 4 26B-A4B MoE backbone (25.2B total / 3.8B active)."
   ]
  },
  {
   "url": "https://llm-releases.com",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Cohere",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "A 30B-total / 3B-active MoE for agentic coding with a 256K context and up to 64K output, sized to run locally for enterprise coding agents",
   "organization_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### North Mini Code 1.0\nAvailable\nCohereOpen source\nCohere's first developer-focused model and the first in its North family of code agents."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "A 30B-total / 3B-active MoE for agentic coding with a 256K context and up to 64K output, sized to run locally for enterprise coding agents."
   ]
  },
  {
   "url": "https://llm-releases.com",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Google DeepMind",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "A dense 12B member of the Gemma 4 family with a unified, encoder-free multimodal architecture",
   "organization_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Gemma 4 12B\nAvailable\nGoogle DeepMindOpen source\nA dense 12B member of the Gemma 4 family with a unified, encoder-free multimodal architecture: vision and audio are projected straight into the LLM backbone."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Gemma 4 12B\nAvailable\nGoogle DeepMindOpen source\nA dense 12B member of the Gemma 4 family with a unified, encoder-free multimodal architecture: vision and audio are projected straight into the LLM backbone."
   ]
  },
  {
   "url": "https://llm-releases.com",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Nex AGI",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Nex AGI's open-weight agentic flagship, post-trained on Qwen3.5-397B-A17B (397B total / ~17B active MoE)",
   "organization_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Nex-N2-Pro\nAvailable\nNex AGIOpen source\nNex AGI's open-weight agentic flagship, post-trained on Qwen3.5-397B-A17B (397B total / ~17B active MoE) by the Shanghai Innovation Institute-led Nex alliance."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Nex-N2-Pro\nAvailable\nNex AGIOpen source\nNex AGI's open-weight agentic flagship, post-trained on Qwen3.5-397B-A17B (397B total / ~17B active MoE) by the Shanghai Innovation Institute-led Nex alliance."
   ]
  },
  {
   "url": "https://llm-releases.com",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "StepFun",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "a ~196B-total / ~11B-active vision-language model with native image and video understanding, a 256K context, and selectable reasoning tiers",
   "organization_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Step-3.7-Flash\nAvailable\nStepFunOpen source\nStepFun's high-efficiency multimodal sparse-MoE successor to Step-3.5-Flash: a ~196B-total / ~11B-active vision-language model with native image and video understanding, a 256K context, and selectable reasoning tiers (high/medium/low)."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Step-3.7-Flash\nAvailable\nStepFunOpen source\nStepFun's high-efficiency multimodal sparse-MoE successor to Step-3.5-Flash: a ~196B-total / ~11B-active vision-language model with native image and video understanding, a 256K context, and selectable reasoning tiers (high/medium/low)."
   ]
  },
  {
   "url": "https://llm-releases.com",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Liquid AI",
   "model": null,
   "release_date": "2026-05-28",
   "access_type": null,
   "license": null,
   "significance": "8.3B total parameters with only ~1.5B active per forward pass (32 experts, 4 active per token)",
   "organization_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### LFM2.5-8B-A1B\nAvailable\nLiquid AIOpen weights\nLiquid AI's on-device Mixture-of-Experts model: 8.3B total parameters with only ~1.5B active per forward pass (32 experts, 4 active per token)."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "Released May 28 2026 under the LFM Open License; caught in a July catalog-gap sweep."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### LFM2.5-8B-A1B\nAvailable\nLiquid AIOpen weights\nLiquid AI's on-device Mixture-of-Experts model: 8.3B total parameters with only ~1.5B active per forward pass (32 experts, 4 active per token)."
   ]
  },
  {
   "url": "https://llm-releases.com",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Alibaba (Qwen)",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Dense 27B that punches far above its weight on agentic coding \u2014 easy to self-host on a single GPU node",
   "organization_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Qwen3.6-27B\nAvailable\nAlibaba (Qwen)Open source\nDense 27B that punches far above its weight on agentic coding \u2014 easy to self-host on a single GPU node."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Qwen3.6-27B\nAvailable\nAlibaba (Qwen)Open source\nDense 27B that punches far above its weight on agentic coding \u2014 easy to self-host on a single GPU node."
   ]
  },
  {
   "url": "https://llm-releases.com",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "DeepSeek",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Efficient V4 companion model with 284B total / 13B active parameters and the same one-million-token context window",
   "organization_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### DeepSeek V4-Flash\nAvailable\nDeepSeekOpen source\nEfficient V4 companion model with 284B total / 13B active parameters and the same one-million-token context window."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### DeepSeek V4-Flash\nAvailable\nDeepSeekOpen source\nEfficient V4 companion model with 284B total / 13B active parameters and the same one-million-token context window."
   ]
  },
  {
   "url": "https://llm-releases.com",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "DeepSeek",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Preview-series sparse MoE flagship with a one-million-token context window and 1.6T total / 49B active parameters",
   "organization_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### DeepSeek V4-Pro\nAvailable\nDeepSeekFrontierOpen source\nPreview-series sparse MoE flagship with a one-million-token context window and 1.6T total / 49B active parameters."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### DeepSeek V4-Pro\nAvailable\nDeepSeekFrontierOpen source\nPreview-series sparse MoE flagship with a one-million-token context window and 1.6T total / 49B active parameters."
   ]
  },
  {
   "url": "https://llm-releases.com",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Tencent Hunyuan",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "A 295B-total / 21B-active Transformer MoE",
   "organization_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Hunyuan Hy3-preview\nAvailable\nTencent HunyuanFrontierOpen weights\nTencent's third-generation Hunyuan, rebuilt from scratch in ~90 days and open-sourced as the \"Hy3 preview\"."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "A 295B-total / 21B-active Transformer MoE (80 layers, 192 experts with top-8 routing, plus a 3.8B multi-token-prediction layer) with a 256K-token context, positioned as a leading open reasoning-and-agent model for its size with strong cost efficiency."
   ]
  },
  {
   "url": "https://llm-releases.com",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Xiaomi (MiMo)",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "a 1.02T-parameter Mixture-of-Experts model with ~42B active parameters, a hybrid-attention architecture, and a 1M-token context window",
   "organization_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### MiMo-V2.5-Pro\nAvailable\nXiaomi (MiMo)FrontierOpen source\nXiaomi's open-weight flagship: a 1.02T-parameter Mixture-of-Experts model with ~42B active parameters, a hybrid-attention architecture, and a 1M-token context window."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### MiMo-V2.5-Pro\nAvailable\nXiaomi (MiMo)FrontierOpen source\nXiaomi's open-weight flagship: a 1.02T-parameter Mixture-of-Experts model with ~42B active parameters, a hybrid-attention architecture, and a 1M-token context window."
   ]
  },
  {
   "url": "https://llm-releases.com",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Xiaomi (MiMo)",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "~310B total parameters with ~15B active, trained on ~48T tokens, with a 1M-token context window",
   "organization_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### MiMo-V2.5\nAvailable\nXiaomi (MiMo)Open source\nXiaomi's open-weight sparse-MoE model: ~310B total parameters with ~15B active, trained on ~48T tokens, with a 1M-token context window."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### MiMo-V2.5\nAvailable\nXiaomi (MiMo)Open source\nXiaomi's open-weight sparse-MoE model: ~310B total parameters with ~15B active, trained on ~48T tokens, with a 1M-token context window."
   ]
  },
  {
   "url": "https://llm-releases.com",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Google DeepMind",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Google DeepMind's Gemma 4 advanced-reasoning open model for personal computers",
   "organization_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Gemma 4 31B\nAvailable\nGoogle DeepMindOpen source\nGoogle DeepMind's Gemma 4 advanced-reasoning open model for personal computers, part of the April 2026 Gemma 4 family."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Gemma 4 31B\nAvailable\nGoogle DeepMindOpen source\nGoogle DeepMind's Gemma 4 advanced-reasoning open model for personal computers, part of the April 2026 Gemma 4 family."
   ]
  },
  {
   "url": "https://llm-releases.com",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Mistral AI",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "A 119B-total / ~6B-active Mixture-of-Experts (128 experts, 4 active per token) with native text+image input, a 256K context, and a configurable reasoning_effort toggle",
   "organization_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Mistral Small 4\nAvailable\nMistral AIOpen source\nMistral's March 2026 Small release: the first Mistral model to unify reasoning (Magistral), multimodal understanding (Pixtral), and agentic coding (Devstral) into one Apache 2.0 model."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "A 119B-total / ~6B-active Mixture-of-Experts (128 experts, 4 active per token) with native text+image input, a 256K context, and a configurable reasoning_effort toggle for fast or deep responses."
   ]
  },
  {
   "url": "https://llm-releases.com",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "StepFun",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "StepFun's Apache-licensed sparse MoE model for fast agentic execution, coding, math, browsing, and tool-use workflows",
   "organization_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Step-3.5-Flash\nAvailable\nStepFunOpen source\nStepFun's Apache-licensed sparse MoE model for fast agentic execution, coding, math, browsing, and tool-use workflows."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Step-3.5-Flash\nAvailable\nStepFunOpen source\nStepFun's Apache-licensed sparse MoE model for fast agentic execution, coding, math, browsing, and tool-use workflows."
   ]
  },
  {
   "url": "https://llm-releases.com",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Sarvam AI",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Apache-licensed Indian-context MoE from Sarvam AI, optimized for reasoning, coding, agentic tasks, and 22 Indian languages",
   "organization_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Sarvam-105B\nAvailable\nSarvam AIOpen source\nApache-licensed Indian-context MoE from Sarvam AI, optimized for reasoning, coding, agentic tasks, and 22 Indian languages."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Sarvam-105B\nAvailable\nSarvam AIOpen source\nApache-licensed Indian-context MoE from Sarvam AI, optimized for reasoning, coding, agentic tasks, and 22 Indian languages."
   ]
  },
  {
   "url": "https://llm-releases.com",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Alibaba (Qwen)",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "A dense 9B with native vision, a 262K-token context, and the Qwen3.5 family's unified hybrid thinking / non-thinking mode",
   "organization_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Qwen3.5-9B\nAvailable\nAlibaba (Qwen)Open source\nThe flagship of Alibaba's small dense Qwen3.5 models."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "A dense 9B with native vision, a 262K-token context, and the Qwen3.5 family's unified hybrid thinking / non-thinking mode."
   ]
  },
  {
   "url": "https://llm-releases.com",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Alibaba (Qwen)",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "A dense 4B in Alibaba's small Qwen3.5 family",
   "organization_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Qwen3.5-4B\nAvailable\nAlibaba (Qwen)Open source\nA dense 4B in Alibaba's small Qwen3.5 family, rated by Artificial Analysis as the most intelligent model under 5B parameters at launch \u2014 outscoring several 7B\u20139B peers despite roughly half the parameters."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Qwen3.5-4B\nAvailable\nAlibaba (Qwen)Open source\nA dense 4B in Alibaba's small Qwen3.5 family, rated by Artificial Analysis as the most intelligent model under 5B parameters at launch \u2014 outscoring several 7B\u20139B peers despite roughly half the parameters."
   ]
  },
  {
   "url": "https://llm-releases.com",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Alibaba (Qwen)",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "A dense 2B Qwen3.5 model built for high-throughput, low-latency edge and on-device use",
   "organization_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Qwen3.5-2B\nAvailable\nAlibaba (Qwen)Open source\nA dense 2B Qwen3.5 model built for high-throughput, low-latency edge and on-device use."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Qwen3.5-2B\nAvailable\nAlibaba (Qwen)Open source\nA dense 2B Qwen3.5 model built for high-throughput, low-latency edge and on-device use."
   ]
  },
  {
   "url": "https://llm-releases.com",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Alibaba (Qwen)",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "a dense 0.8B designed for the most constrained on-device deployments, operating in non-thinking (instruct) mode by default",
   "organization_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Qwen3.5-0.8B\nAvailable\nAlibaba (Qwen)Open source\nThe smallest Qwen3.5 model \u2014 a dense 0.8B designed for the most constrained on-device deployments, operating in non-thinking (instruct) mode by default."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Qwen3.5-0.8B\nAvailable\nAlibaba (Qwen)Open source\nThe smallest Qwen3.5 model \u2014 a dense 0.8B designed for the most constrained on-device deployments, operating in non-thinking (instruct) mode by default."
   ]
  },
  {
   "url": "https://llm-releases.com",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Alibaba (Qwen)",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Native vision-language MoE supporting 201 languages with a 1M-token context",
   "organization_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Qwen3.5-397B\nAvailable\nAlibaba (Qwen)FrontierOpen source\nNative vision-language MoE supporting 201 languages with a 1M-token context."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Qwen3.5-397B\nAvailable\nAlibaba (Qwen)FrontierOpen source\nNative vision-language MoE supporting 201 languages with a 1M-token context."
   ]
  },
  {
   "url": "https://llm-releases.com",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Z.ai (Zhipu AI)",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Z.ai flagship for complex systems engineering and long-horizon agentic tasks, scaling the GLM line to 744B total / 40B active parameters",
   "organization_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### GLM-5\nAvailable\nZ.ai (Zhipu AI)FrontierOpen source\nZ.ai flagship for complex systems engineering and long-horizon agentic tasks, scaling the GLM line to 744B total / 40B active parameters."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### GLM-5\nAvailable\nZ.ai (Zhipu AI)FrontierOpen source\nZ.ai flagship for complex systems engineering and long-horizon agentic tasks, scaling the GLM line to 744B total / 40B active parameters."
   ]
  },
  {
   "url": "https://llm-releases.com",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Alibaba (Qwen)",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Apache-licensed Qwen3-Next coding-agent model with 80B total / 3B active parameters, 256K context, and long-horizon tool-use training",
   "organization_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Qwen3-Coder-Next\nAvailable\nAlibaba (Qwen)Open source\nApache-licensed Qwen3-Next coding-agent model with 80B total / 3B active parameters, 256K context, and long-horizon tool-use training."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Qwen3-Coder-Next\nAvailable\nAlibaba (Qwen)Open source\nApache-licensed Qwen3-Next coding-agent model with 80B total / 3B active parameters, 256K context, and long-horizon tool-use training."
   ]
  },
  {
   "url": "https://llm-releases.com",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Z.ai (Zhipu AI)",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Coding-focused GLM release with improved multilingual agentic coding, terminal tasks, tool use, and interface generation",
   "organization_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### GLM-4.7\nAvailable\nZ.ai (Zhipu AI)FrontierOpen source\nCoding-focused GLM release with improved multilingual agentic coding, terminal tasks, tool use, and interface generation."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### GLM-4.7\nAvailable\nZ.ai (Zhipu AI)FrontierOpen source\nCoding-focused GLM release with improved multilingual agentic coding, terminal tasks, tool use, and interface generation."
   ]
  },
  {
   "url": "https://llm-releases.com",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Allen Institute for AI (Ai2)",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Ai2's fully open thinking model with public weights, code, data, checkpoints, and training details across the OLMo 3 pipeline",
   "organization_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### OLMo 3 Think 32B\nAvailable\nAllen Institute for AI (Ai2)Open source\nAi2's fully open thinking model with public weights, code, data, checkpoints, and training details across the OLMo 3 pipeline."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### OLMo 3 Think 32B\nAvailable\nAllen Institute for AI (Ai2)Open source\nAi2's fully open thinking model with public weights, code, data, checkpoints, and training details across the OLMo 3 pipeline."
   ]
  },
  {
   "url": "https://llm-releases.com",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Z.ai (Zhipu AI)",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Open 106B-class vision-language model with native multimodal function calling for visual agents",
   "organization_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### GLM-4.6V\nAvailable\nZ.ai (Zhipu AI)Open source\nOpen 106B-class vision-language model with native multimodal function calling for visual agents."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### GLM-4.6V\nAvailable\nZ.ai (Zhipu AI)Open source\nOpen 106B-class vision-language model with native multimodal function calling for visual agents."
   ]
  },
  {
   "url": "https://llm-releases.com",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "DeepSeek",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Reasoning-first agent model that adds DeepSeek Sparse Attention and thinking directly inside tool-use workflows",
   "organization_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### DeepSeek-V3.2\nAvailable\nDeepSeekFrontierOpen source\nReasoning-first agent model that adds DeepSeek Sparse Attention and thinking directly inside tool-use workflows."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### DeepSeek-V3.2\nAvailable\nDeepSeekFrontierOpen source\nReasoning-first agent model that adds DeepSeek Sparse Attention and thinking directly inside tool-use workflows."
   ]
  },
  {
   "url": "https://llm-releases.com",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "DeepSeek",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "High-compute reasoning variant of V3.2, positioned for olympiad-level math, programming, and other deep reasoning tasks",
   "organization_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### DeepSeek-V3.2-Speciale\nAvailable\nDeepSeekFrontierOpen source\nHigh-compute reasoning variant of V3.2, positioned for olympiad-level math, programming, and other deep reasoning tasks."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### DeepSeek-V3.2-Speciale\nAvailable\nDeepSeekFrontierOpen source\nHigh-compute reasoning variant of V3.2, positioned for olympiad-level math, programming, and other deep reasoning tasks."
   ]
  },
  {
   "url": "https://llm-releases.com",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Moonshot AI",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "MIT-licensed hybrid linear-attention model using Kimi Delta Attention, built for million-token contexts with much lower KV-cache usage",
   "organization_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Kimi-Linear-48B-A3B-Instruct\nAvailable\nMoonshot AIOpen source\nMIT-licensed hybrid linear-attention model using Kimi Delta Attention, built for million-token contexts with much lower KV-cache usage."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### Kimi-Linear-48B-A3B-Instruct\nAvailable\nMoonshot AIOpen source\nMIT-licensed hybrid linear-attention model using Kimi Delta Attention, built for million-token contexts with much lower KV-cache usage."
   ]
  },
  {
   "url": "https://llm-releases.com",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Z.ai (Zhipu AI)",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Agentic reasoning and coding upgrade over GLM-4.5, expanding the text context window",
   "organization_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### GLM-4.6\nAvailable\nZ.ai (Zhipu AI)FrontierOpen source\nAgentic reasoning and coding upgrade over GLM-4.5, expanding the text context window from 128K to 20"
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The model release tracker",
    "# Every large language model release, sourced and tracked over time.",
    "### GLM-4.6\nAvailable\nZ.ai (Zhipu AI)FrontierOpen source\nAgentic reasoning and coding upgrade over GLM-4.5, expanding the text context window from 128K to 20"
   ]
  },
  {
   "url": "https://digitalapplied.com/blog/thinking-machines-inkling-open-weight-multimodal-launch-2026",
   "title": "Inkling: Murati\u2019s Open-Weight Bet Lands on Hugging Face",
   "published_at": "2026-07-15T23:00:00",
   "organization": "Thinking Machines",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "A 975-billion-parameter open-weight Mixture-of-Experts model that is natively multimodal and features a controllable thinking-effort dial.",
   "organization_evidence": [
    "Thinking Machines Lab released Inkling on July 15, 2026 \u2014 the first in-house model from the startup Mira Murati founded after leaving OpenAI, and a 975-billion-parameter open-weight bet that the future of enterprise AI is customization, not capability leaderboards."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Thinking Machines Lab released Inkling on July 15, 2026 \u2014 the first in-house model from the startup Mira Murati founded after leaving OpenAI, and a 975-billion-parameter open-weight bet that the future of enterprise AI is customization, not capability leaderboards."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Thinking Machines Lab released Inkling on July 15, 2026 \u2014 the first in-house model from the startup Mira Murati founded after leaving OpenAI, and a 975-billion-parameter open-weight bet that the future of enterprise AI is customization, not capability leaderboards.",
    "The specs read like a frontier launch: a Mixture-of-Experts transformer with 41 billion active parameters per token, trained on 45 trillion tokens spanning text, images, audio, and video, released under Apache 2.0 with full weights live on Hugging Face."
   ]
  },
  {
   "url": "http://rits.shanghai.nyu.edu/ai/thinking-machines-releases-inkling-its-first-open-weight-model",
   "title": "Thinking Machines Releases Inkling, Its First Open-Weight Model",
   "published_at": null,
   "organization": "Thinking Machines Lab",
   "model": "Inkling",
   "release_date": "2026-07-15",
   "access_type": "open-weight",
   "license": "Apache 2.0",
   "significance": "975-billion-parameter Mixture-of-Experts model supporting a context window of up to 1 million tokens",
   "organization_evidence": [
    "# Thinking Machines Releases Inkling, Its First Open-Weight Model",
    "**On July 15, 2026, Thinking Machines Lab released Inkling** \u2014 its first model trained from scratch and, notably, its first *open-weight* release."
   ],
   "model_evidence": [
    "# Thinking Machines Releases Inkling, Its First Open-Weight Model",
    "**On July 15, 2026, Thinking Machines Lab released Inkling** \u2014 its first model trained from scratch and, notably, its first *open-weight* release."
   ],
   "release_date_evidence": [
    "**On July 15, 2026, Thinking Machines Lab released Inkling** \u2014 its first model trained from scratch and, notably, its first *open-weight* release."
   ],
   "access_type_evidence": [
    "# Thinking Machines Releases Inkling, Its First Open-Weight Model",
    "**On July 15, 2026, Thinking Machines Lab released Inkling** \u2014 its first model trained from scratch and, notably, its first *open-weight* release."
   ],
   "license_evidence": [
    "Inkling is a 975-billion-parameter Mixture-of-Experts model (41B active) published under the permissive Apache 2.0 license, with full weights on Hugging Face."
   ],
   "significance_evidence": [
    "# Thinking Machines Releases Inkling, Its First Open-Weight Model",
    "Inkling is a 975-billion-parameter Mixture-of-Experts model (41B active) published under the permissive Apache 2.0 license, with full weights on Hugging Face.",
    "Attention alternates between sliding-window and global layers at a 5:1 ratio, and the model supports a context window of up to **1 million tokens**."
   ]
  },
  {
   "url": "https://hostingseekers.com/blog/best-open-source-llms",
   "title": "10 Best Open-Source LLMs in 2026",
   "published_at": "2026-07-24T06:40:49",
   "organization": "Z.ai (formerly Zhipu AI)",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "A Mixture-of-Experts model with 744B total parameters and a 1-million-token context window that ranks highest among open-weight models on the Artificial Analysis Intelligence Index.",
   "organization_evidence": [
    "# 10 Best Open-Source LLMs in 2026",
    "This guide ranks the 10 strongest open-source/open-weight LLMs as of 2026, with real specs, verified licenses, actual pricing, and honest pros and cons for each.",
    "Z.ai (formerly Zhipu AI), a Beijing-based, Tsinghua-spun-out company, released GLM-5.2 on June 13, 2026, as the third major release in its GLM-5 line."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# 10 Best Open-Source LLMs in 2026",
    "This guide ranks the 10 strongest open-source/open-weight LLMs as of 2026, with real specs, verified licenses, actual pricing, and honest pros and cons for each.",
    "Z.ai (formerly Zhipu AI), a Beijing-based, Tsinghua-spun-out company, released GLM-5.2 on June 13, 2026, as the third major release in its GLM-5 line."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# 10 Best Open-Source LLMs in 2026",
    "This guide ranks the 10 strongest open-source/open-weight LLMs as of 2026, with real specs, verified licenses, actual pricing, and honest pros and cons for each.",
    "Z.ai (formerly Zhipu AI), a Beijing-based, Tsinghua-spun-out company, released GLM-5.2 on June 13, 2026, as the third major release in its GLM-5 line.",
    "It\u2019s a Mixture-of-Experts model reported at 744B total parameters with roughly 40B active per token."
   ]
  },
  {
   "url": "https://hostingseekers.com/blog/best-open-source-llms",
   "title": "10 Best Open-Source LLMs in 2026",
   "published_at": "2026-07-24T06:40:49",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-06-12",
   "access_type": null,
   "license": null,
   "significance": "A 1-trillion-parameter MoE model designed for agentic coding with reduced reasoning token overhead and independent MCP benchmark successes.",
   "organization_evidence": [
    "# 10 Best Open-Source LLMs in 2026",
    "This guide ranks the 10 strongest open-source/open-weight LLMs as of 2026, with real specs, verified licenses, actual pricing, and honest pros and cons for each.",
    "Moonshot AI\u2019s Kimi line has kept a fast 2\u20133-month release cadence since the original K2 in July 2025."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# 10 Best Open-Source LLMs in 2026",
    "This guide ranks the 10 strongest open-source/open-weight LLMs as of 2026, with real specs, verified licenses, actual pricing, and honest pros and cons for each.",
    "K2.7 Code, released June 12, 2026, is a 1-trillion-parameter MoE model with 32B active parameters and 384 experts, a 256K context window, and a design focus on cutting \u201cthinking token\u201d overhead\u2014roughly 30% fewer reasoning tokens than K2.6 for comparable or better output."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# 10 Best Open-Source LLMs in 2026",
    "This guide ranks the 10 strongest open-source/open-weight LLMs as of 2026, with real specs, verified licenses, actual pricing, and honest pros and cons for each.",
    "K2.7 Code, released June 12, 2026, is a 1-trillion-parameter MoE model with 32B active parameters and 384 experts, a 256K context window, and a design focus on cutting \u201cthinking token\u201d overhead\u2014roughly 30% fewer reasoning tokens than K2.6 for comparable or better output."
   ]
  },
  {
   "url": "https://hostingseekers.com/blog/best-open-source-llms",
   "title": "10 Best Open-Source LLMs in 2026",
   "published_at": "2026-07-24T06:40:49",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "A two-tier open-weight preview providing advanced reasoning, high context windows, and low pricing.",
   "organization_evidence": [
    "# 10 Best Open-Source LLMs in 2026",
    "This guide ranks the 10 strongest open-source/open-weight LLMs as of 2026, with real specs, verified licenses, actual pricing, and honest pros and cons for each.",
    "DeepSeek shipped V4 as a two-tier open-weight preview on April 24, 2026: V4-Pro (1.6T total parameters, ~49B active) for advanced reasoning and agentic coding, and V4-Flash (284B total, ~13B active) for faster, cheaper inference."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# 10 Best Open-Source LLMs in 2026",
    "This guide ranks the 10 strongest open-source/open-weight LLMs as of 2026, with real specs, verified licenses, actual pricing, and honest pros and cons for each.",
    "DeepSeek shipped V4 as a two-tier open-weight preview on April 24, 2026: V4-Pro (1.6T total parameters, ~49B active) for advanced reasoning and agentic coding, and V4-Flash (284B total, ~13B active) for faster, cheaper inference."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# 10 Best Open-Source LLMs in 2026",
    "This guide ranks the 10 strongest open-source/open-weight LLMs as of 2026, with real specs, verified licenses, actual pricing, and honest pros and cons for each.",
    "DeepSeek shipped V4 as a two-tier open-weight preview on April 24, 2026: V4-Pro (1.6T total parameters, ~49B active) for advanced reasoning and agentic coding, and V4-Flash (284B total, ~13B active) for faster, cheaper inference."
   ]
  },
  {
   "url": "https://hostingseekers.com/blog/best-open-source-llms",
   "title": "10 Best Open-Source LLMs in 2026",
   "published_at": "2026-07-24T06:40:49",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "An open-weight model combining frontier-level coding, a 1M-token context window, and native multimodal inputs.",
   "organization_evidence": [
    "# 10 Best Open-Source LLMs in 2026",
    "This guide ranks the 10 strongest open-source/open-weight LLMs as of 2026, with real specs, verified licenses, actual pricing, and honest pros and cons for each.",
    "MiniMax, a Shanghai-based lab, released M3 on June 1, 2026, positioning it as the first open-weight model to combine frontier-level coding, a 1M-token context window, and native text/image/video input in a single checkpoint."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# 10 Best Open-Source LLMs in 2026",
    "This guide ranks the 10 strongest open-source/open-weight LLMs as of 2026, with real specs, verified licenses, actual pricing, and honest pros and cons for each.",
    "MiniMax, a Shanghai-based lab, released M3 on June 1, 2026, positioning it as the first open-weight model to combine frontier-level coding, a 1M-token context window, and native text/image/video input in a single checkpoint."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# 10 Best Open-Source LLMs in 2026",
    "This guide ranks the 10 strongest open-source/open-weight LLMs as of 2026, with real specs, verified licenses, actual pricing, and honest pros and cons for each.",
    "MiniMax, a Shanghai-based lab, released M3 on June 1, 2026, positioning it as the first open-weight model to combine frontier-level coding, a 1M-token context window, and native text/image/video input in a single checkpoint."
   ]
  },
  {
   "url": "https://hostingseekers.com/blog/best-open-source-llms",
   "title": "10 Best Open-Source LLMs in 2026",
   "published_at": "2026-07-24T06:40:49",
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-04-16",
   "access_type": null,
   "license": null,
   "significance": "A hybrid attention model family featuring a 35B-A3B MoE model and a 27B dense model that excels on agentic coding benchmarks.",
   "organization_evidence": [
    "# 10 Best Open-Source LLMs in 2026",
    "This guide ranks the 10 strongest open-source/open-weight LLMs as of 2026, with real specs, verified licenses, actual pricing, and honest pros and cons for each.",
    "Alibaba\u2019s Qwen team shipped Qwen3.6 in two waves in April 2026: the 35B-A3B MoE model, with 3B active parameters, released April 16, and a dense 27B model, released April 22,"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# 10 Best Open-Source LLMs in 2026",
    "This guide ranks the 10 strongest open-source/open-weight LLMs as of 2026, with real specs, verified licenses, actual pricing, and honest pros and cons for each.",
    "Alibaba\u2019s Qwen team shipped Qwen3.6 in two waves in April 2026: the 35B-A3B MoE model, with 3B active parameters, released April 16, and a dense 27B model, released April 22,"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# 10 Best Open-Source LLMs in 2026",
    "This guide ranks the 10 strongest open-source/open-weight LLMs as of 2026, with real specs, verified licenses, actual pricing, and honest pros and cons for each.",
    "Alibaba\u2019s Qwen team shipped Qwen3.6 in two waves in April 2026: the 35B-A3B MoE model, with 3B active parameters, released April 16, and a dense 27B model, released April 22, that notably outperforms the much larger 397B Qwen3.5 flagship on agentic coding benchmarks while running on a single consumer GPU."
   ]
  },
  {
   "url": "https://hostingseekers.com/blog/best-open-source-llms",
   "title": "10 Best Open-Source LLMs in 2026",
   "published_at": "2026-07-24T06:40:49",
   "organization": "Google DeepMind",
   "model": null,
   "release_date": "2026-04-02",
   "access_type": null,
   "license": null,
   "significance": "A family of models ranging from edge sizes to a 31B dense flagship, released under Apache 2.0 with strong benchmark results.",
   "organization_evidence": [
    "# 10 Best Open-Source LLMs in 2026",
    "This guide ranks the 10 strongest open-source/open-weight LLMs as of 2026, with real specs, verified licenses, actual pricing, and honest pros and cons for each.",
    "Google DeepMind released Gemma 4 on April 2, 2026, in four to five sizes ranging from edge-optimized E2B/E4B models up through a 26B MoE and a 31B dense flagship."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# 10 Best Open-Source LLMs in 2026",
    "This guide ranks the 10 strongest open-source/open-weight LLMs as of 2026, with real specs, verified licenses, actual pricing, and honest pros and cons for each.",
    "Google DeepMind released Gemma 4 on April 2, 2026, in four to five sizes ranging from edge-optimized E2B/E4B models up through a 26B MoE and a 31B dense flagship."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# 10 Best Open-Source LLMs in 2026",
    "This guide ranks the 10 strongest open-source/open-weight LLMs as of 2026, with real specs, verified licenses, actual pricing, and honest pros and cons for each.",
    "Google DeepMind released Gemma 4 on April 2, 2026, in four to five sizes ranging from edge-optimized E2B/E4B models up through a 26B MoE and a 31B dense flagship."
   ]
  },
  {
   "url": "https://techstory.in/thinking-machines-launches-inkling-open-weight-ai-model-to-challenge-global-rivals",
   "title": "Thinking Machines Launches Inkling Open-Weight AI Model to Challenge Global Rivals",
   "published_at": "2026-07-16T17:37:12",
   "organization": "Thinking Machines",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "975 billion parameter open-weight Mixture-of-Experts model that activates 41 billion parameters per task and performs strongly on agent tasks",
   "organization_evidence": [
    "Thinking Machines, the San Francisco-based AI business founded by former OpenAI Chief Technology Officer Mira Murati, has launched its first general-purpose AI model, Inkling, an open-weight system with 975 billion parameters."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Murati announced the debut on X on July 15, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Thinking Machines, the San Francisco-based AI business founded by former OpenAI Chief Technology Officer Mira Murati, has launched its first general-purpose AI model, Inkling, an open-weight system with 975 billion parameters.",
    "The 975 billion parameter count makes Inkling one of the largest open-weight models released to date.",
    "Its architecture uses a Mixture-of-Experts design, which activates only 41 billion parameters at a time for any given task making it more computationally efficient than a dense model of the same total size despite the headline parameter count.",
    "According to Reuters, Inkling performed best on agent-related tasks, which require AI systems to perform a series of activities rather than answer single questions."
   ]
  },
  {
   "url": "https://basedai.co/research/open-weight-inflection",
   "title": "The open-weight inflection",
   "published_at": "2026-07-17T00:00:00",
   "organization": "Moonshot",
   "model": null,
   "release_date": "2026-01-27",
   "access_type": null,
   "license": null,
   "significance": "First leg of the trigger cluster",
   "organization_evidence": [
    "The open-weight inflectionJuly 2026 \u00b7 First editionBasedAI \u2014 Research",
    "A January\u2013February cluster \u2014 Moonshot's Kimi K2.5, Z.ai's GLM-5 (shipped inside the crossover week itself), and Alibaba's Qwen3.5 \u2014 preceded the first break.",
    "| Jan 27 | Kimi K2.5 (Moonshot) | Modified MIT | First leg of the trigger cluster; later the base of Cursor's Composer 2 |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The open-weight inflectionJuly 2026 \u00b7 First editionBasedAI \u2014 Research",
    "3. **The acceleration sits inside a release cadence**: Kimi K2.5 (January 27), GLM-5 (February 11 \u2014 inside the crossover week), Qwen3.5 (February 16), DeepSeek V4 (April 24, MIT, 1M context, agentic-grade), GLM-5.2 and Kimi K2.7 Code (June 12\u201313), then Kimi K3 (July 16, weights promised by July 27)."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The open-weight inflectionJuly 2026 \u00b7 First editionBasedAI \u2014 Research",
    "| Jan 27 | Kimi K2.5 (Moonshot) | Modified MIT | First leg of the trigger cluster; later the base of Cursor's Composer 2 |"
   ]
  },
  {
   "url": "https://basedai.co/research/open-weight-inflection",
   "title": "The open-weight inflection",
   "published_at": "2026-07-17T00:00:00",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-02-11",
   "access_type": null,
   "license": null,
   "significance": "Shipped inside the crossover week",
   "organization_evidence": [
    "The open-weight inflectionJuly 2026 \u00b7 First editionBasedAI \u2014 Research",
    "A January\u2013February cluster \u2014 Moonshot's Kimi K2.5, Z.ai's GLM-5 (shipped inside the crossover week itself), and Alibaba's Qwen3.5 \u2014 preceded the first break.",
    "| Feb 11 | GLM-5 (Z.ai) | MIT | Shipped inside the crossover week |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The open-weight inflectionJuly 2026 \u00b7 First editionBasedAI \u2014 Research",
    "3. **The acceleration sits inside a release cadence**: Kimi K2.5 (January 27), GLM-5 (February 11 \u2014 inside the crossover week), Qwen3.5 (February 16), DeepSeek V4 (April 24, MIT, 1M context, agentic-grade), GLM-5.2 and Kimi K2.7 Code (June 12\u201313), then Kimi K3 (July 16, weights promised by July 27)."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The open-weight inflectionJuly 2026 \u00b7 First editionBasedAI \u2014 Research",
    "| Feb 11 | GLM-5 (Z.ai) | MIT | Shipped inside the crossover week |"
   ]
  },
  {
   "url": "https://basedai.co/research/open-weight-inflection",
   "title": "The open-weight inflection",
   "published_at": "2026-07-17T00:00:00",
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-02-16",
   "access_type": null,
   "license": null,
   "significance": "Third leg; open eight-tier family",
   "organization_evidence": [
    "The open-weight inflectionJuly 2026 \u00b7 First editionBasedAI \u2014 Research",
    "A January\u2013February cluster \u2014 Moonshot's Kimi K2.5, Z.ai's GLM-5 (shipped inside the crossover week itself), and Alibaba's Qwen3.5 \u2014 preceded the first break.",
    "| Feb ~16 | Qwen3.5-397B (Alibaba) | Apache 2.0 | Third leg; open eight-tier family |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The open-weight inflectionJuly 2026 \u00b7 First editionBasedAI \u2014 Research",
    "3. **The acceleration sits inside a release cadence**: Kimi K2.5 (January 27), GLM-5 (February 11 \u2014 inside the crossover week), Qwen3.5 (February 16), DeepSeek V4 (April 24, MIT, 1M context, agentic-grade), GLM-5.2 and Kimi K2.7 Code (June 12\u201313), then Kimi K3 (July 16, weights promised by July 27)."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The open-weight inflectionJuly 2026 \u00b7 First editionBasedAI \u2014 Research",
    "| Feb ~16 | Qwen3.5-397B (Alibaba) | Apache 2.0 | Third leg; open eight-tier family |"
   ]
  },
  {
   "url": "https://basedai.co/research/open-weight-inflection",
   "title": "The open-weight inflection",
   "published_at": "2026-07-17T00:00:00",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "First agentic-grade DeepSeek; second wave",
   "organization_evidence": [
    "The open-weight inflectionJuly 2026 \u00b7 First editionBasedAI \u2014 Research",
    "DeepSeek V4, released April 24 under an MIT license with a 1M-token context, drove a second wave; OpenRouter calls it \"the first DeepSeek model sufficient for agentic workloads.\"",
    "| Apr 24 | DeepSeek V4 (Pro/Flash) | MIT | First agentic-grade DeepSeek; second wave |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The open-weight inflectionJuly 2026 \u00b7 First editionBasedAI \u2014 Research",
    "3. **The acceleration sits inside a release cadence**: Kimi K2.5 (January 27), GLM-5 (February 11 \u2014 inside the crossover week), Qwen3.5 (February 16), DeepSeek V4 (April 24, MIT, 1M context, agentic-grade), GLM-5.2 and Kimi K2.7 Code (June 12\u201313), then Kimi K3 (July 16, weights promised by July 27)."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The open-weight inflectionJuly 2026 \u00b7 First editionBasedAI \u2014 Research",
    "| Apr 24 | DeepSeek V4 (Pro/Flash) | MIT | First agentic-grade DeepSeek; second wave |"
   ]
  },
  {
   "url": "https://basedai.co/research/open-weight-inflection",
   "title": "The open-weight inflection",
   "published_at": "2026-07-17T00:00:00",
   "organization": "Moonshot",
   "model": null,
   "release_date": "2026-06-12",
   "access_type": null,
   "license": null,
   "significance": "Coding-agent specialist; Coinbase default",
   "organization_evidence": [
    "The open-weight inflectionJuly 2026 \u00b7 First editionBasedAI \u2014 Research",
    "June 12\u201313 brought Moonshot's Kimi K2.7 Code and Z.ai's GLM-5.2 (753B, MIT, 1M context; its model card advertises \"no regional limits\") \u2014 the latter recording the fastest adoption of any model Vercel tracked in 2026, with daily token volume up roughly 27x and customer count up roughly 80x in its first full week.",
    "| Jun 12 | Kimi K2.7 Code (Moonshot) | Modified MIT | Coding-agent specialist; Coinbase default |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The open-weight inflectionJuly 2026 \u00b7 First editionBasedAI \u2014 Research",
    "3. **The acceleration sits inside a release cadence**: Kimi K2.5 (January 27), GLM-5 (February 11 \u2014 inside the crossover week), Qwen3.5 (February 16), DeepSeek V4 (April 24, MIT, 1M context, agentic-grade), GLM-5.2 and Kimi K2.7 Code (June 12\u201313), then Kimi K3 (July 16, weights promised by July 27)."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The open-weight inflectionJuly 2026 \u00b7 First editionBasedAI \u2014 Research",
    "| Jun 12 | Kimi K2.7 Code (Moonshot) | Modified MIT | Coding-agent specialist; Coinbase default |"
   ]
  },
  {
   "url": "https://basedai.co/research/open-weight-inflection",
   "title": "The open-weight inflection",
   "published_at": "2026-07-17T00:00:00",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "Fastest 2026 adoption on Vercel's gateway",
   "organization_evidence": [
    "The open-weight inflectionJuly 2026 \u00b7 First editionBasedAI \u2014 Research",
    "June 12\u201313 brought Moonshot's Kimi K2.7 Code and Z.ai's GLM-5.2 (753B, MIT, 1M context; its model card advertises \"no regional limits\") \u2014 the latter recording the fastest adoption of any model Vercel tracked in 2026, with daily token volume up roughly 27x and customer count up roughly 80x in its first full week.",
    "| Jun 13 | GLM-5.2 (Z.ai) | MIT | Fastest 2026 adoption on Vercel's gateway |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The open-weight inflectionJuly 2026 \u00b7 First editionBasedAI \u2014 Research",
    "3. **The acceleration sits inside a release cadence**: Kimi K2.5 (January 27), GLM-5 (February 11 \u2014 inside the crossover week), Qwen3.5 (February 16), DeepSeek V4 (April 24, MIT, 1M context, agentic-grade), GLM-5.2 and Kimi K2.7 Code (June 12\u201313), then Kimi K3 (July 16, weights promised by July 27)."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The open-weight inflectionJuly 2026 \u00b7 First editionBasedAI \u2014 Research",
    "| Jun 13 | GLM-5.2 (Z.ai) | MIT | Fastest 2026 adoption on Vercel's gateway |"
   ]
  },
  {
   "url": "https://basedai.co/research/open-weight-inflection",
   "title": "The open-weight inflection",
   "published_at": "2026-07-17T00:00:00",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "First Western frontier-scale answer",
   "organization_evidence": [
    "The open-weight inflectionJuly 2026 \u00b7 First editionBasedAI \u2014 Research",
    "The first Western frontier-scale answer arrived only this week: Thinking Machines Lab's open-weight, 975B-parameter Inkling, released July 15.",
    "Mistral was the West's principal open publisher until Thinking Machines released the 975B open-weight Inkling on July 15."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The open-weight inflectionJuly 2026 \u00b7 First editionBasedAI \u2014 Research",
    "The first Western frontier-scale answer arrived only this week: Thinking Machines Lab's open-weight, 975B-parameter Inkling, released July 15."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The open-weight inflectionJuly 2026 \u00b7 First editionBasedAI \u2014 Research",
    "The first Western frontier-scale answer arrived only this week: Thinking Machines Lab's open-weight, 975B-parameter Inkling, released July 15."
   ]
  },
  {
   "url": "https://nerdleveltech.com/nex-n2-pro-open-weight-coding-benchmarks",
   "title": "Nex-N2-Pro: Open-Weight Coder vs GPT-5.5 (2026)",
   "published_at": "2026-06-10T00:00:00",
   "organization": "Nex AGI",
   "model": null,
   "release_date": "2026-06-02",
   "access_type": null,
   "license": null,
   "significance": "a 397-billion-parameter coding model featuring an Agentic Thinking framework and Adaptive Thinking that cuts thinking tokens by 30-50%",
   "organization_evidence": [
    "ai-ml",
    "# Nex-N2-Pro: Open-Weight Coder vs GPT-5.5 (2026)",
    "Nex-N2-Pro is a free, open-weight 397-billion-parameter coding model from Nex AGI, post-trained on Alibaba's Qwen3.5.12 On its own benchmarks it leads the open-weight pack and edges GPT-5.5 on SWE-Bench Pro, but it trails the closed frontier on the hardest agentic and reasoning tasks.13"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "ai-ml",
    "# Nex-N2-Pro: Open-Weight Coder vs GPT-5.5 (2026)",
    "Released June 2, 2026 and now spreading across free endpoints, it is the latest entry in an early-June surge of open-weight coding models \u2014 arriving a day after [MiniMax M3](minimax-m3-open-weight-coding-sparse-attention).45"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "ai-ml",
    "# Nex-N2-Pro: Open-Weight Coder vs GPT-5.5 (2026)",
    "Nex-N2-Pro is a free, open-weight 397-billion-parameter coding model from Nex AGI, post-trained on Alibaba's Qwen3.5.12 On its own benchmarks it leads the open-weight pack and edges GPT-5.5 on SWE-Bench Pro, but it trails the closed frontier on the hardest agentic and reasoning tasks.13",
    "Two things set it apart: its headline feature is an \"Agentic Thinking\" framework that decides on its own how hard to reason, and \u2014 unlike M3, whose weights were still pending at launch \u2014 you can actually download Nex-N2-Pro right now.12"
   ]
  },
  {
   "url": "https://the-agent-report.com/2026/07/kimi-k3-moonshot-28t-open-model-july-2026",
   "title": "Kimi K3: Moonshot AI Drops a 2.8-Trillion-Parameter Open Model \u2014 and It Codes, Designs Chips, and Edits Video",
   "published_at": "2026-07-17T08:00:00",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-16",
   "access_type": null,
   "license": null,
   "significance": "a 2.8-trillion-parameter Mixture-of-Experts model with a 1-million-token context window and native vision, the largest open-weight model ever announced",
   "organization_evidence": [
    "# Kimi K3: Moonshot AI Drops a 2.8-Trillion-Parameter Open Model \u2014 and It Codes, Designs Chips, and Edits Video",
    "Late on July 16, Chinese AI lab Moonshot AI published a blog post announcing Kimi K3 \u2014 a 2.8-trillion-parameter sparse Mixture-of-Experts model that instantly became the largest open-weight model in existence, roughly 1.75\u00d7 the size of DeepSeek V4 Pro\u2019s 1.6T parameters."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "**TL;DR** \u2014 Moonshot AI released Kimi K3 on July 16, a 2.8-trillion-parameter Mixture-of-Experts model with a 1-million-token context window and native vision.",
    "# Kimi K3: Moonshot AI Drops a 2.8-Trillion-Parameter Open Model \u2014 and It Codes, Designs Chips, and Edits Video"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "**TL;DR** \u2014 Moonshot AI released Kimi K3 on July 16, a 2.8-trillion-parameter Mixture-of-Experts model with a 1-million-token context window and native vision.",
    "# Kimi K3: Moonshot AI Drops a 2.8-Trillion-Parameter Open Model \u2014 and It Codes, Designs Chips, and Edits Video"
   ]
  },
  {
   "url": "https://theagenttimes.com/agents/article/thinking-machines-lab-releases-inkling-open-weight-multimoda-ba3abfe1",
   "title": "Thinking Machines Lab Releases Inkling, Open-Weight Multimodal Model\u2026",
   "published_at": null,
   "organization": "Thinking Machines Lab",
   "model": "Inkling",
   "release_date": "2026-07-15",
   "access_type": "open-weight",
   "license": null,
   "significance": "multimodal model trained from scratch with open weights",
   "organization_evidence": [
    "Thinking Machines Lab Releases Inkling, Open-Weight Multimodal Model Fine-Tunable on Tinker",
    "Thinking Machines Lab, led by former OpenAI CTO Mira Murati, released Inkling \u2014 an open-weight multimodal model trained from scratch, available for fine-tuning on the company's Tinker platform as of July 15, 2026."
   ],
   "model_evidence": [
    "Thinking Machines Lab Releases Inkling, Open-Weight Multimodal Model Fine-Tunable on Tinker",
    "Thinking Machines Lab, led by former OpenAI CTO Mira Murati, released Inkling \u2014 an open-weight multimodal model trained from scratch, available for fine-tuning on the company's Tinker platform as of July 15, 2026."
   ],
   "release_date_evidence": [
    "Thinking Machines Lab Releases Inkling, Open-Weight Multimodal Model Fine-Tunable on Tinker",
    "Thinking Machines Lab, led by former OpenAI CTO Mira Murati, released Inkling \u2014 an open-weight multimodal model trained from scratch, available for fine-tuning on the company's Tinker platform as of July 15, 2026."
   ],
   "access_type_evidence": [
    "Thinking Machines Lab Releases Inkling, Open-Weight Multimodal Model Fine-Tunable on Tinker",
    "Thinking Machines Lab, led by former OpenAI CTO Mira Murati, released Inkling \u2014 an open-weight multimodal model trained from scratch, available for fine-tuning on the company's Tinker platform as of July 15, 2026."
   ],
   "license_evidence": null,
   "significance_evidence": [
    "Thinking Machines Lab Releases Inkling, Open-Weight Multimodal Model Fine-Tunable on Tinker",
    "Thinking Machines Lab, led by former OpenAI CTO Mira Murati, released Inkling \u2014 an open-weight multimodal model trained from scratch, available for fine-tuning on the company's Tinker platform as of July 15, 2026."
   ]
  },
  {
   "url": "https://aiinsiders.net/article/thinking-machines-ships-its-first-open-model-a-975b-moe",
   "title": "Thinking Machines Ships Its First Open Model, a 975B MoE",
   "published_at": "2026-07-16T14:37:25",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "a 975 billion parameter mixture of experts model featuring a one million token context window and 41 billion active parameters",
   "organization_evidence": [
    "MODELS",
    "Thinking Machines Lab released **Inkling** on July 15, 2026, a 975 billion parameter **mixture of experts** model that marks the company\u2019s first open-weights release."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "MODELS",
    "Thinking Machines Lab released **Inkling** on July 15, 2026, a 975 billion parameter **mixture of experts** model that marks the company\u2019s first open-weights release."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "MODELS",
    "Thinking Machines Lab released **Inkling** on July 15, 2026, a 975 billion parameter **mixture of experts** model that marks the company\u2019s first open-weights release.",
    "Inkling supports multimodal reasoning and a context window spanning one million tokens, roughly enough to hold a large codebase or several full novels in a single prompt."
   ]
  },
  {
   "url": "https://blog.prompt20.com/posts/open-weights-ultimate-guide",
   "title": "Open Weights: The Ultimate Guide (2026 Edition)",
   "published_at": "2026-05-22T00:00:00",
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "397B sparse + dense variants (35B-A3B, 27B). Best multilingual coverage; tool-use parity with closed.",
   "organization_evidence": [
    "Open Weights: The Ultimate Guide (2026 Edition)",
    "In 2026 the gap has closed further: **DeepSeek V4, Qwen 3.6 Plus, GLM-5.1, Kimi K2.6, MiniMax M2.7** are all open-weight and benchmark within a few points of Claude Opus 4.7 and GPT-5.5 on most public evals \u2014 at 1/10th to 1/30th the per-token cost when self-hosted.",
    "**Qwen 3.6 Plus** (Alibaba, China)."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Open Weights: The Ultimate Guide (2026 Edition)",
    "The 2026 open-weight frontier (the roster)",
    "**Qwen 3.6 Plus** (Alibaba, China). 397B sparse + dense variants (35B-A3B, 27B). April 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Open Weights: The Ultimate Guide (2026 Edition)",
    "The 2026 open-weight frontier (the roster)",
    "**Qwen 3.6 Plus** (Alibaba, China). 397B sparse + dense variants (35B-A3B, 27B). April 2026. Apache 2.0 on base, API-only on Plus tier. Best multilingual coverage; tool-use parity with closed."
   ]
  },
  {
   "url": "https://blog.prompt20.com/posts/open-weights-ultimate-guide",
   "title": "Open Weights: The Ultimate Guide (2026 Edition)",
   "published_at": "2026-05-22T00:00:00",
   "organization": "Z.ai / Zhipu",
   "model": null,
   "release_date": "2026-04-03",
   "access_type": null,
   "license": null,
   "significance": "~750B sparse (256 routed experts + 1 shared, 8 active per token; hidden 6144, 78 layers). The GLM-5V Turbo variant adds native vision. Tops several coding benchmarks; OpenClaw-compatible agent harness scoring.",
   "organization_evidence": [
    "Open Weights: The Ultimate Guide (2026 Edition)",
    "In 2026 the gap has closed further: **DeepSeek V4, Qwen 3.6 Plus, GLM-5.1, Kimi K2.6, MiniMax M2.7** are all open-weight and benchmark within a few points of Claude Opus 4.7 and GPT-5.5 on most public evals \u2014 at 1/10th to 1/30th the per-token cost when self-hosted.",
    "**GLM-5.1** (Z.ai / Zhipu, China)."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Open Weights: The Ultimate Guide (2026 Edition)",
    "The 2026 open-weight frontier (the roster)",
    "**GLM-5.1** (Z.ai / Zhipu, China). ~750B sparse (256 routed experts + 1 shared, 8 active per token; hidden 6144, 78 layers; HF `createdAt: 2026-04-03`)."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Open Weights: The Ultimate Guide (2026 Edition)",
    "The 2026 open-weight frontier (the roster)",
    "**GLM-5.1** (Z.ai / Zhipu, China). ~750B sparse (256 routed experts + 1 shared, 8 active per token; hidden 6144, 78 layers; HF `createdAt: 2026-04-03`). MIT. The GLM-5V Turbo variant adds native vision. Tops several coding benchmarks; OpenClaw-compatible agent harness scoring."
   ]
  },
  {
   "url": "https://blog.prompt20.com/posts/open-weights-ultimate-guide",
   "title": "Open Weights: The Ultimate Guide (2026 Edition)",
   "published_at": "2026-05-22T00:00:00",
   "organization": "Moonshot",
   "model": null,
   "release_date": "2026-04-14",
   "access_type": null,
   "license": null,
   "significance": "~1T sparse, ~32B active. Long-context flagship (256K). Natively multimodal. Reasoning variant K2.5-Thinking ships as a separate release.",
   "organization_evidence": [
    "Open Weights: The Ultimate Guide (2026 Edition)",
    "In 2026 the gap has closed further: **DeepSeek V4, Qwen 3.6 Plus, GLM-5.1, Kimi K2.6, MiniMax M2.7** are all open-weight and benchmark within a few points of Claude Opus 4.7 and GPT-5.5 on most public evals \u2014 at 1/10th to 1/30th the per-token cost when self-hosted.",
    "**Kimi K2.6** (Moonshot, China)."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Open Weights: The Ultimate Guide (2026 Edition)",
    "The 2026 open-weight frontier (the roster)",
    "**Kimi K2.6** (Moonshot, China). ~1T sparse, ~32B active (384 routed experts, 8 active per token; HF `createdAt: 2026-04-14`)."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Open Weights: The Ultimate Guide (2026 Edition)",
    "The 2026 open-weight frontier (the roster)",
    "**Kimi K2.6** (Moonshot, China). ~1T sparse, ~32B active (384 routed experts, 8 active per token; HF `createdAt: 2026-04-14`). Modified MIT. Long-context flagship (256K). Natively multimodal (vision + text via `KimiK25ForConditionalGeneration`). Reasoning variant K2.5-Thinking ships as a separate release."
   ]
  },
  {
   "url": "https://blog.prompt20.com/posts/open-weights-ultimate-guide",
   "title": "Open Weights: The Ultimate Guide (2026 Edition)",
   "published_at": "2026-05-22T00:00:00",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-03",
   "access_type": null,
   "license": null,
   "significance": "230B sparse, 10B active. Non-commercial weights. Open weights but not commercial-OK by default.",
   "organization_evidence": [
    "Open Weights: The Ultimate Guide (2026 Edition)",
    "In 2026 the gap has closed further: **DeepSeek V4, Qwen 3.6 Plus, GLM-5.1, Kimi K2.6, MiniMax M2.7** are all open-weight and benchmark within a few points of Claude Opus 4.7 and GPT-5.5 on most public evals \u2014 at 1/10th to 1/30th the per-token cost when self-hosted.",
    "**MiniMax M2.7** (MiniMax, China)."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Open Weights: The Ultimate Guide (2026 Edition)",
    "The 2026 open-weight frontier (the roster)",
    "**MiniMax M2.7** (MiniMax, China). 230B sparse, 10B active. March 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Open Weights: The Ultimate Guide (2026 Edition)",
    "The 2026 open-weight frontier (the roster)",
    "**MiniMax M2.7** (MiniMax, China). 230B sparse, 10B active. March 2026. Non-commercial weights. Open weights but not commercial-OK by default."
   ]
  },
  {
   "url": "https://kie.ai/blog/what-is-inkling",
   "title": "What Is Inkling? The 975B Open-Weights MoE From Mira Murati's Thinking Machines",
   "published_at": "2026-07-16T11:05:20",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "975B/41B-active open-weights MoE with native text, image, audio and a 1M context window",
   "organization_evidence": [
    "# What Is Inkling? 975B Open-Weights MoE Explained",
    "Inkling is an open-weights Mixture-of-Experts foundation model released by Thinking Machines Lab on July 15, 2026, with 975B total parameters, 41B active per token, a 1M-token context window, and native reasoning across text, images, and audio."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# What Is Inkling? 975B Open-Weights MoE Explained",
    "Published: July 16, 2026",
    "Inkling is an open-weights Mixture-of-Experts foundation model released by Thinking Machines Lab on July 15, 2026, with 975B total parameters, 41B active per token, a 1M-token context window, and native reasoning across text, images, and audio."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# What Is Inkling? 975B Open-Weights MoE Explained",
    "TLDRInkling is Thinking Machines Lab's 975B/41B-active open-weights MoE with native text, image, audio and a 1M context window. Apache 2.0."
   ]
  },
  {
   "url": "https://aitechconnect.in/news/inkling-975b-apache-2-open-weight-thinking-machines-2026",
   "title": "Inkling: America's Largest Open-Weight Model Ships Apache 2.0",
   "published_at": "2026-07-27T06:00:00",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "975 billion total parameters in a mixture-of-experts architecture with 1 million tokens of context",
   "organization_evidence": [
    "Thinking Machines Lab \u2014 the research company founded by former OpenAI chief technology officer Mira Murati \u2014 released Inkling on 15 July 2026.",
    "## What changed on 15 July"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Thinking Machines Lab \u2014 the research company founded by former OpenAI chief technology officer Mira Murati \u2014 released Inkling on 15 July 2026.",
    "## What changed on 15 July"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "- **975 billion total parameters** in a mixture-of-experts architecture, with 41 billion active per token.",
    "- **One million tokens of context**, multimodal, with native reasoning across text, images and audio and a controllable \"thinking effort\" setting.",
    "## What changed on 15 July"
   ]
  },
  {
   "url": "https://aitechconnect.in/news/inkling-975b-apache-2-open-weight-thinking-machines-2026",
   "title": "Inkling: America's Largest Open-Weight Model Ships Apache 2.0",
   "published_at": "2026-07-27T06:00:00",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "276 billion total parameters matching or exceeding its larger sibling on many benchmarks",
   "organization_evidence": [
    "Thinking Machines Lab reports that Inkling-Small \u2014 276 billion total parameters, 12 billion active \u2014 matches or exceeds its far larger sibling on many benchmarks.",
    "## What changed on 15 July"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "| Inkling-Small (Thinking Machines Lab) | 276B (MoE) | 12B | 1M tokens | Apache 2.0 | 15 Jul 2026 |",
    "| Model | Total parameters | Active per token | Context | Licence | Weights released |",
    "## How Inkling sits against the current open-weight field"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Thinking Machines Lab reports that Inkling-Small \u2014 276 billion total parameters, 12 billion active \u2014 matches or exceeds its far larger sibling on many benchmarks.",
    "## What changed on 15 July"
   ]
  },
  {
   "url": "https://aitechconnect.in/news/inkling-975b-apache-2-open-weight-thinking-machines-2026",
   "title": "Inkling: America's Largest Open-Weight Model Ships Apache 2.0",
   "published_at": "2026-07-27T06:00:00",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-27",
   "access_type": null,
   "license": null,
   "significance": "2.8T mixture-of-experts model with 1M tokens of context",
   "organization_evidence": [
    "| [Kimi K3 (Moonshot AI)](/news/kimi-k3-open-moe-2-8t-frontend-arena-2026) | 2.8T (MoE) | 16 of 896 experts \u2014 roughly 50B of live compute per step | 1M tokens | Per model card at release | 27 Jul 2026, 00:00 UTC |",
    "| Model | Total parameters | Active per token | Context | Licence | Weights released |",
    "## How Inkling sits against the current open-weight field"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "| [Kimi K3 (Moonshot AI)](/news/kimi-k3-open-moe-2-8t-frontend-arena-2026) | 2.8T (MoE) | 16 of 896 experts \u2014 roughly 50B of live compute per step | 1M tokens | Per model card at release | 27 Jul 2026, 00:00 UTC |",
    "| Model | Total parameters | Active per token | Context | Licence | Weights released |",
    "## How Inkling sits against the current open-weight field"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "| [Kimi K3 (Moonshot AI)](/news/kimi-k3-open-moe-2-8t-frontend-arena-2026) | 2.8T (MoE) | 16 of 896 experts \u2014 roughly 50B of live compute per step | 1M tokens | Per model card at release | 27 Jul 2026, 00:00 UTC |",
    "| Model | Total parameters | Active per token | Context | Licence | Weights released |",
    "## How Inkling sits against the current open-weight field"
   ]
  },
  {
   "url": "https://aniketkarneai.com/blog/2026-07-26-inkling-thinking-machines-open-weights",
   "title": "Inkling: What Thinking Machines Lab's First Open-Weight Release Tells Us About the Frontier \u2014 aniketkarneai.com",
   "published_at": "2026-07-26T00:00:00",
   "organization": "Thinking Machines Lab",
   "model": "Inkling",
   "release_date": "2026-07-15",
   "access_type": "open-weights",
   "license": null,
   "significance": "a 975B MoE model that scores 97.1% on AIME 2026 and 87.2% on GPQA Diamond",
   "organization_evidence": [
    "Inkling: What Thinking Machines Lab's First Open-Weight Release Tells Us About the Frontier",
    "On July 15, 2026, Thinking Machines Lab \u2014 the company founded by former OpenAI CTO Mira Murati \u2014 released its first open-weights model."
   ],
   "model_evidence": [
    "Inkling: What Thinking Machines Lab's First Open-Weight Release Tells Us About the Frontier",
    "Thinking Machines Lab dropped Inkling on July 15 \u2014 a 975B MoE model that scores 97.1% on AIME 2026 and 87.2% on GPQA Diamond."
   ],
   "release_date_evidence": [
    "Inkling: What Thinking Machines Lab's First Open-Weight Release Tells Us About the Frontier",
    "Thinking Machines Lab dropped Inkling on July 15 \u2014 a 975B MoE model that scores 97.1% on AIME 2026 and 87.2% on GPQA Diamond.",
    "On July 15, 2026, Thinking Machines Lab \u2014 the company founded by former OpenAI CTO Mira Murati \u2014 released its first open-weights model."
   ],
   "access_type_evidence": [
    "Inkling: What Thinking Machines Lab's First Open-Weight Release Tells Us About the Frontier",
    "On July 15, 2026, Thinking Machines Lab \u2014 the company founded by former OpenAI CTO Mira Murati \u2014 released its first open-weights model."
   ],
   "license_evidence": null,
   "significance_evidence": [
    "Inkling: What Thinking Machines Lab's First Open-Weight Release Tells Us About the Frontier",
    "Thinking Machines Lab dropped Inkling on July 15 \u2014 a 975B MoE model that scores 97.1% on AIME 2026 and 87.2% on GPQA Diamond."
   ]
  },
  {
   "url": "https://techpillow.co/blog/thinking-machines-inkling-open-weight-multimodal-ai-2026",
   "title": "Thinking Machines Releases Inkling: A 975B Open-Weight Multimodal AI",
   "published_at": "2026-07-19T00:00:00",
   "organization": "Thinking Machines Lab",
   "model": "Inkling",
   "release_date": "2026-07-15",
   "access_type": "open-weight",
   "license": "Apache 2.0 licence",
   "significance": "A 975-billion-parameter sparse Mixture-of-Experts model with 41 billion active parameters, trained on 45 trillion tokens.",
   "organization_evidence": [
    "Thinking Machines Releases Inkling: A 975B Open-Weight Multimodal AI",
    "Thinking Machines Lab Releases Inkling on 15 July 2026",
    "On 15 July 2026, Thinking Machines Lab \u2014 the AI company founded by Mira Murati, former Chief Technology Officer of OpenAI \u2014 released Inkling, its first in-house AI model."
   ],
   "model_evidence": [
    "Thinking Machines Releases Inkling: A 975B Open-Weight Multimodal AI",
    "Thinking Machines Lab Releases Inkling on 15 July 2026",
    "On 15 July 2026, Thinking Machines Lab \u2014 the AI company founded by Mira Murati, former Chief Technology Officer of OpenAI \u2014 released Inkling, its first in-house AI model."
   ],
   "release_date_evidence": [
    "Thinking Machines Lab Releases Inkling on 15 July 2026",
    "On 15 July 2026, Thinking Machines Lab \u2014 the AI company founded by Mira Murati, former Chief Technology Officer of OpenAI \u2014 released Inkling, its first in-house AI model."
   ],
   "access_type_evidence": [
    "Thinking Machines Releases Inkling: A 975B Open-Weight Multimodal AI",
    "Inkling is a 975-billion-parameter sparse Mixture-of-Experts model with 41 billion active parameters, trained on 45 trillion tokens spanning text, images, audio, and video."
   ],
   "license_evidence": [
    "Thinking Machines Lab Releases Inkling on 15 July 2026",
    "The weights are available on Hugging Face under an Apache 2.0 licence, alongside an NVFP4 quantised checkpoint tuned for NVIDIA Blackwell hardware."
   ],
   "significance_evidence": [
    "Thinking Machines Releases Inkling: A 975B Open-Weight Multimodal AI",
    "Thinking Machines Lab Releases Inkling on 15 July 2026",
    "Inkling is a 975-billion-parameter sparse Mixture-of-Experts model with 41 billion active parameters, trained on 45 trillion tokens spanning text, images, audio, and video."
   ]
  },
  {
   "url": "https://rits.shanghai.nyu.edu/ai/thinking-machines-releases-inkling-its-first-open-weight-model",
   "title": "Thinking Machines Releases Inkling, Its First Open-Weight Model",
   "published_at": null,
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "A 975-billion-parameter Mixture-of-Experts model supporting a context window of up to 1 million tokens.",
   "organization_evidence": [
    "# Thinking Machines Releases Inkling, Its First Open-Weight Model",
    "**On July 15, 2026, Thinking Machines Lab released Inkling** \u2014 its first model trained from scratch and, notably, its first *open-weight* release."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "**On July 15, 2026, Thinking Machines Lab released Inkling** \u2014 its first model trained from scratch and, notably, its first *open-weight* release."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Thinking Machines Releases Inkling, Its First Open-Weight Model",
    "Inkling is a 975-billion-parameter Mixture-of-Experts model (41B active) published under the permissive Apache 2.0 license, with full weights on Hugging Face."
   ]
  },
  {
   "url": "https://buildfastwithai.com/blogs/best-open-source-ai-models-2026",
   "title": "Best Open Source AI Models July 2026: Full Collection",
   "published_at": "2026-07-17T11:27:10",
   "organization": "Thinking Machines",
   "model": "Inkling",
   "release_date": "2026-07-15",
   "access_type": "open-weight",
   "license": "Apache 2.0",
   "significance": "975B Apache 2.0 multimodal base built for fine-tuning",
   "organization_evidence": [
    "Best Open Source AI Models July 2026: Full Collection",
    "## Best for Customization: Inkling",
    "Inkling from Thinking Machines Lab is the best open model to fine-tune into your own, released July 15, 2026 under a clean Apache 2.0 license with 975B total parameters, 41B active, native text, image, and audio reasoning, and a thinking-effort dial from 0.2 to 0.99."
   ],
   "model_evidence": [
    "Best Open Source AI Models July 2026: Full Collection",
    "## Best for Customization: Inkling",
    "Inkling from Thinking Machines Lab is the best open model to fine-tune into your own, released July 15, 2026 under a clean Apache 2.0 license with 975B total parameters, 41B active, native text, image, and audio reasoning, and a thinking-effort dial from 0.2 to 0.99."
   ],
   "release_date_evidence": [
    "Best Open Source AI Models July 2026: Full Collection",
    "## Best for Customization: Inkling",
    "Inkling from Thinking Machines Lab is the best open model to fine-tune into your own, released July 15, 2026 under a clean Apache 2.0 license with 975B total parameters, 41B active, native text, image, and audio reasoning, and a thinking-effort dial from 0.2 to 0.99."
   ],
   "access_type_evidence": [
    "Best Open Source AI Models July 2026: Full Collection",
    "## Best Open Source AI Models in 2026: The Complete Collection",
    "In the space of 31 days, open-weight models matched the closed frontier on terminal coding, set the all-time record on web browsing agents, and shipped the largest model ever released to the public."
   ],
   "license_evidence": [
    "Best Open Source AI Models July 2026: Full Collection",
    "## Best for Customization: Inkling",
    "Inkling from Thinking Machines Lab is the best open model to fine-tune into your own, released July 15, 2026 under a clean Apache 2.0 license with 975B total parameters, 41B active, native text, image, and audio reasoning, and a thinking-effort dial from 0.2 to 0.99."
   ],
   "significance_evidence": [
    "Best Open Source AI Models July 2026: Full Collection",
    "## The State of Open Source AI in July 2026",
    "Thinking Machines dropped Inkling on July 15, a 975B Apache 2.0 multimodal base built for fine-tuning."
   ]
  },
  {
   "url": "https://magazine.sebastianraschka.com/p/a-dream-of-spring-for-open-weight",
   "title": "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
   "published_at": "2026-02-25T13:26:56",
   "organization": "Arcee AI",
   "model": null,
   "release_date": "2026-01-27",
   "access_type": null,
   "license": null,
   "significance": "400B parameter Mixture-of-Experts (MoE) flagship model with 13B active parameters matching the modeling performance of older models while introducing alternating local-global attention layers, QK-Norm, and gated attention",
   "organization_evidence": [
    "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
    "1. Arcee AI\u2019s Trinity Large (Jan 27, 2026)",
    "On January 27, Arcee AI (a company I hadn\u2019t had on my radar up to then) [began releasing](https://www.arcee.ai/blog/trinity-large) versions of their open-weight 400B Trinity Large LLMs on the [model hub](https://huggingface.co/arcee-ai/Trinity-Large-Preview), along with two smaller variants:"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
    "1. Arcee AI\u2019s Trinity Large (Jan 27, 2026)",
    "On January 27, Arcee AI (a company I hadn\u2019t had on my radar up to then) [began releasing](https://www.arcee.ai/blog/trinity-large) versions of their open-weight 400B Trinity Large LLMs on the [model hub](https://huggingface.co/arcee-ai/Trinity-Large-Preview), along with two smaller variants:"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
    "# 1. Arcee AI\u2019s Trinity Large: A New US-Based Start-Up Sharing Open-Weight Models",
    "On January 27, Arcee AI (a company I hadn\u2019t had on my radar up to then) [began releasing](https://www.arcee.ai/blog/trinity-large) versions of their open-weight 400B Trinity Large LLMs on the [model hub](https://huggingface.co/arcee-ai/Trinity-Large-Preview), along with two smaller variants:",
    "Their flagship large model is a 400B param [Mixture-of-Experts (MoE)](https://magazine.sebastianraschka.com/i/168650848/12-mixture-of-experts-moe) with 13B active parameters."
   ]
  },
  {
   "url": "https://magazine.sebastianraschka.com/p/a-dream-of-spring-for-open-weight",
   "title": "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
   "published_at": "2026-02-25T13:26:56",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-01-27",
   "access_type": null,
   "license": null,
   "significance": "1-trillion-parameter native multimodal model with vision support that set a new open-weight performance ceiling matching leading proprietary models",
   "organization_evidence": [
    "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
    "2. Moonshot AI\u2019s Kimi K2.5 (Jan 27, 2026)",
    "While Arcee Trinity essentially matched the modeling performance of the older GLM-4.5 model,[ Kimi K2.5 is an open-weight model](https://huggingface.co/moonshotai/Kimi-K2.5) that set a new open-weight performance ceiling at the time of its release on Jan 27."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
    "2. Moonshot AI\u2019s Kimi K2.5 (Jan 27, 2026)",
    "While Arcee Trinity essentially matched the modeling performance of the older GLM-4.5 model,[ Kimi K2.5 is an open-weight model](https://huggingface.co/moonshotai/Kimi-K2.5) that set a new open-weight performance ceiling at the time of its release on Jan 27."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
    "# 2. Moonshot AI\u2019s Kimi K2.5: A DeepSeek-Like Model at a 1-Trillion-Parameter Scale",
    "While Arcee Trinity essentially matched the modeling performance of the older GLM-4.5 model,[ Kimi K2.5 is an open-weight model](https://huggingface.co/moonshotai/Kimi-K2.5) that set a new open-weight performance ceiling at the time of its release on Jan 27.",
    "Overall, the Kimi K2.5 architecture is similar to Kimi K2, which, in turn, is a scaled-up version of the DeepSeek V3 architecture."
   ]
  },
  {
   "url": "https://magazine.sebastianraschka.com/p/a-dream-of-spring-for-open-weight",
   "title": "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
   "published_at": "2026-02-25T13:26:56",
   "organization": "StepFun",
   "model": null,
   "release_date": "2026-02-01",
   "access_type": null,
   "license": null,
   "significance": "196B parameter model delivering high throughput with multi-token prediction (MTP-3)",
   "organization_evidence": [
    "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
    "3. StepFun Step 3.5 Flash (Feb 1, 2026)",
    "Step 3.5 Flash is a 196B parameter model that is more than 3x smaller than the recent DeepSeek V3.2 model (671B) while being slightly ahead in modeling performance benchmarks."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
    "3. StepFun Step 3.5 Flash (Feb 1, 2026)"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
    "# 3. StepFun\u2019s Step 3.5 Flash: Good Performance at Great Tokens/Sec Throughput",
    "Step 3.5 Flash is a 196B parameter model that is more than 3x smaller than the recent DeepSeek V3.2 model (671B) while being slightly ahead in modeling performance benchmarks."
   ]
  },
  {
   "url": "https://magazine.sebastianraschka.com/p/a-dream-of-spring-for-open-weight",
   "title": "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
   "published_at": "2026-02-25T13:26:56",
   "organization": "Qwen",
   "model": null,
   "release_date": "2026-02-03",
   "access_type": null,
   "license": null,
   "significance": "80B coding model utilizing a Gated DeltaNet and Gated Attention hybrid that matches Claude Sonnet 4.5 on SWE-Bench Pro",
   "organization_evidence": [
    "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
    "4. Qwen3-Coder-Next (Feb 3, 2026)",
    "In early February 2026, the Qwen3 team shared the 80B Qwen3-Coder-Next model (3B parameters active), which made big headlines for outperforming much larger models like DeepSeek V3.2 (37B active) and Kimi K2.5 and GLM-4.7 (both 32B active) on coding tasks."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
    "4. Qwen3-Coder-Next (Feb 3, 2026)"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
    "# 4. Qwen3-Coder-Next: An Attention-Hybrid for Coding",
    "In early February 2026, the Qwen3 team shared the 80B Qwen3-Coder-Next model (3B parameters active), which made big headlines for outperforming much larger models like DeepSeek V3.2 (37B active) and Kimi K2.5 and GLM-4.7 (both 32B active) on coding tasks."
   ]
  },
  {
   "url": "https://magazine.sebastianraschka.com/p/a-dream-of-spring-for-open-weight",
   "title": "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
   "published_at": "2026-02-25T13:26:56",
   "organization": "z.AI",
   "model": null,
   "release_date": "2026-02-12",
   "access_type": null,
   "license": null,
   "significance": "744B parameter flagship Mixture-of-Experts model performing on par with major proprietary flagship LLMs",
   "organization_evidence": [
    "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
    "5. z.AI\u2019s GLM-5 (Feb 12, 2026)",
    "The [GLM-5 release](https://z.ai/blog/glm-5) on February 12th was a big deal, because at the time of its release it appeared to be on par with the major flagship LLM offerings, including GPT-5.2 extra-high, Gemini Pro 3, and Claude 4.6 Opus."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
    "5. z.AI\u2019s GLM-5 (Feb 12, 2026)",
    "The [GLM-5 release](https://z.ai/blog/glm-5) on February 12th was a big deal, because at the time of its release it appeared to be on par with the major flagship LLM offerings, including GPT-5.2 extra-high, Gemini Pro 3, and Claude 4.6 Opus."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
    "# 5. z.AI\u2019s GLM-5: A New Flagship Open-Weight Model",
    "The [GLM-5 release](https://z.ai/blog/glm-5) on February 12th was a big deal, because at the time of its release it appeared to be on par with the major flagship LLM offerings, including GPT-5.2 extra-high, Gemini Pro 3, and Claude 4.6 Opus."
   ]
  },
  {
   "url": "https://magazine.sebastianraschka.com/p/a-dream-of-spring-for-open-weight",
   "title": "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
   "published_at": "2026-02-25T13:26:56",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-02-12",
   "access_type": null,
   "license": null,
   "significance": "230B parameter model with classic Grouped Query Attention achieving strong coding performance and massive popularity",
   "organization_evidence": [
    "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
    "6. MiniMax M2.5 (Feb 12, 2026)",
    "Now, back to MiniMax M2.5. Pulling together the GLM-5 data from the SWE-Bench Verified coding benchmark and combining it with the reported MiniMax M2.5, the latter appears to be a slightly stronger model (at least when it comes to coding)."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
    "6. MiniMax M2.5 (Feb 12, 2026)",
    "The aforementioned GLM-5 and Kimi K2.5 are popular open-weight models, but according to[ OpenRouter statistics](https://openrouter.ai/rankings), they pale in comparison to[ MiniMax M2.5](https://huggingface.co/MiniMaxAI/MiniMax-M2.5), which was released on February 12 as well."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
    "# 6. MiniMax M2.5: A Strong Coder with \u201cOnly\u201d 230B Parameters",
    "Architecture-wise, MiniMax M2.5 is a 230B model with a fairly classic design: just plain Grouped Query Attention, no sliding window attention or other efficiency improvements."
   ]
  },
  {
   "url": "https://magazine.sebastianraschka.com/p/a-dream-of-spring-for-open-weight",
   "title": "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
   "published_at": "2026-02-25T13:26:56",
   "organization": "Nanbeige",
   "model": null,
   "release_date": "2026-02-13",
   "access_type": null,
   "license": null,
   "significance": "3B parameter small model designed for on-device use with strong performance gains from post-training",
   "organization_evidence": [
    "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
    "7. Nanbeige 4.1 3B (Feb 13, 2026)"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
    "7. Nanbeige 4.1 3B (Feb 13, 2026)"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
    "# 7. Nanbeige 4.1 3B: A Strong Llama 3 Successor",
    "Why I am mentioning all this is that Nanbeige 4.1 3B seems to target the \u201csmall\u201d LLM on-device use case that Qwen3 is so popular for."
   ]
  },
  {
   "url": "https://magazine.sebastianraschka.com/p/a-dream-of-spring-for-open-weight",
   "title": "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
   "published_at": "2026-02-25T13:26:56",
   "organization": "Qwen",
   "model": null,
   "release_date": "2026-02-15",
   "access_type": null,
   "license": null,
   "significance": "397B parameter Mixture-of-Experts model adopting hybrid attention and native multimodal support",
   "organization_evidence": [
    "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
    "8. Qwen 3.5 (Feb 15, 2026)",
    "However, the Qwen team just released a new Qwen3.5 model variant on February 15."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
    "8. Qwen 3.5 (Feb 15, 2026)",
    "However, the Qwen team just released a new Qwen3.5 model variant on February 15."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
    "# 8. Qwen3.5 and the Continuation of Hybrid Attention",
    "Qwen3.5 397B-A17B, a Mixture-of-Experts (MoE) with 397B parameters (17B active per token), is a step up from the largest Qwen3 model, which is 235B parameters in size.",
    "Architecture-wise, Qwen3.5 adopts the hybrid attention model (featuring Gated DeltaNet) that Qwen3-Next and Qwen3-Coder-Next (section 4) used."
   ]
  },
  {
   "url": "https://magazine.sebastianraschka.com/p/a-dream-of-spring-for-open-weight",
   "title": "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
   "published_at": "2026-02-25T13:26:56",
   "organization": "Ant Group",
   "model": null,
   "release_date": "2026-02-16",
   "access_type": null,
   "license": null,
   "significance": "1-trillion-parameter hybrid attention models with Lightning Attention and Multi-Head Latent Attention for high long-context efficiency",
   "organization_evidence": [
    "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
    "9. Ant Group\u2019s Ling 2.5 1T & Ring 2.5 1T (Feb 16, 2026)",
    "Ant Group\u2019s Ling 2.5 1T & Ring 2.5 1T (Feb 16, 2026)"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
    "9. Ant Group\u2019s Ling 2.5 1T & Ring 2.5 1T (Feb 16, 2026)"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
    "# 9. Ant Group\u2019s Ling 2.5 1T with Lightning Attention"
   ]
  },
  {
   "url": "https://magazine.sebastianraschka.com/p/a-dream-of-spring-for-open-weight",
   "title": "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
   "published_at": "2026-02-25T13:26:56",
   "organization": "Cohere",
   "model": null,
   "release_date": "2026-02-17",
   "access_type": null,
   "license": null,
   "significance": "3.35B parameter multilingual open-weight model featuring parallel transformer blocks",
   "organization_evidence": [
    "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
    "10. Cohere\u2019s Tiny Aya (Feb 17, 2026)",
    "Released on February 17, Tiny Aya is a new, \u201csmall\u201d LLM by Cohere that is said to be the \u201cmost capable multilingual open-weight model\u201d at the 3B parameter size class."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
    "10. Cohere\u2019s Tiny Aya (Feb 17, 2026)",
    "Released on February 17, Tiny Aya is a new, \u201csmall\u201d LLM by Cohere that is said to be the \u201cmost capable multilingual open-weight model\u201d at the 3B parameter size class."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
    "# 10. Tiny Aya: A 3.35B Model with Strong Multilingual Support",
    "Released on February 17, Tiny Aya is a new, \u201csmall\u201d LLM by Cohere that is said to be the \u201cmost capable multilingual open-weight model\u201d at the 3B parameter size class."
   ]
  },
  {
   "url": "https://magazine.sebastianraschka.com/p/a-dream-of-spring-for-open-weight",
   "title": "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
   "published_at": "2026-02-25T13:26:56",
   "organization": "Sarvam",
   "model": null,
   "release_date": "2026-03-06",
   "access_type": null,
   "license": null,
   "significance": "Strong reasoning models in 30B and 105B sizes offering superior performance and token efficiency for Indian languages",
   "organization_evidence": [
    "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
    "11. Update 1: Sarvam 30B and 105B (Mar 6, 2026)",
    "There are two size flavors, [Sarvam 30B](https://huggingface.co/sarvamai/sarvam-30b) and [Sarvam 105B](https://huggingface.co/sarvamai/sarvam-105b) model (both reasoning models), which were released as open-weight models on March 6th alongside a fairly detailed [announcement blog](https://www.sarvam.ai/blogs/sarvam-30b-105b)."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
    "11. Update 1: Sarvam 30B and 105B (Mar 6, 2026)",
    "There are two size flavors, [Sarvam 30B](https://huggingface.co/sarvamai/sarvam-30b) and [Sarvam 105B](https://huggingface.co/sarvamai/sarvam-105b) model (both reasoning models), which were released as open-weight models on March 6th alongside a fairly detailed [announcement blog](https://www.sarvam.ai/blogs/sarvam-30b-105b)."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
    "# Update 1: Sarvam 30B and 105B (Mar 6, 2026)",
    "While waiting for DeepSeek V4 we got two very strong open-weight LLMs from India.",
    "There are two size flavors, [Sarvam 30B](https://huggingface.co/sarvamai/sarvam-30b) and [Sarvam 105B](https://huggingface.co/sarvamai/sarvam-105b) model (both reasoning models), which were released as open-weight models on March 6th alongside a fairly detailed [announcement blog](https://www.sarvam.ai/blogs/sarvam-30b-105b)."
   ]
  },
  {
   "url": "https://edgen.tech/news/post/mira-muratis-inkling-debuts-as-best-open-source-model-in-the-west",
   "title": "Mira Murati's Inkling debuts as best open-source model in the West",
   "published_at": "2026-07-26T14:42:40",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-26",
   "access_type": null,
   "license": null,
   "significance": "A 975-billion-parameter MoE model with 41 billion active parameters at inference supporting text, image, and audio inputs with a 1-million-token context window.",
   "organization_evidence": [
    "Mira Murati's Inkling debuts as best open-source model in the West",
    "Mira Murati's Thinking Machines Lab released Inkling, a 975-billion-parameter open-source model that is the best Western-trained open-weight AI model \u2014 but its narrow strengths and heavy censorship limit its appeal to most developers."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Mira Murati's Inkling debuts as best open-source model in the West",
    "[![Rachel Kim](https://static.edgen.tech/authors/rachel-kim.png?image_process=resize,w_32/quality,q_75/format,webp/sharpen,100)\n## Rachel Kim](https://www.edgen.tech/author/news/rachel-kim)\u00b7Jul 26 2026, 14:42"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Mira Murati's Inkling debuts as best open-source model in the West",
    "Inkling is a 975-billion-parameter MoE model with 41 billion active parameters at inference",
    "Inkling uses a mixture-of-experts architecture with 41 billion parameters active at inference, supports text, image and audio inputs, and processes a 1-million-token context window."
   ]
  },
  {
   "url": "https://yfarmx.com/ai/llms",
   "title": "Large Language Models Explained: Every Major LLM in 2026 | YFarmX",
   "published_at": "2026-08-14T00:00:00",
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-08-03",
   "access_type": null,
   "license": null,
   "significance": "the first Qwen Max released with open weights",
   "organization_evidence": [
    "# Large Language Models",
    "## Latest LLMs",
    "- [![](/media/header-images/qwen3-8-max.webp)\nAlibaba\n## Qwen3.8-Max\nReleased 3 Aug 2026\nthe first Qwen Max released with open weights](/ai/llms/qwen3-8-max/)"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Large Language Models",
    "## Latest LLMs",
    "- [![](/media/header-images/qwen3-8-max.webp)\nAlibaba\n## Qwen3.8-Max\nReleased 3 Aug 2026\nthe first Qwen Max released with open weights](/ai/llms/qwen3-8-max/)"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Large Language Models",
    "## Latest LLMs",
    "- [![](/media/header-images/qwen3-8-max.webp)\nAlibaba\n## Qwen3.8-Max\nReleased 3 Aug 2026\nthe first Qwen Max released with open weights](/ai/llms/qwen3-8-max/)"
   ]
  },
  {
   "url": "https://machineherald.io/article/2026-07/19-mira-muratis-thinking-machines-lab-releases-inkling-its-first-ai-model-a-975-billion-parameter-open-weight-system",
   "title": "Mira Murati's Thinking Machines Lab Releases Inkling, Its First AI Model, a 975-Billion-Parameter Open-Weight System",
   "published_at": "2026-07-19T09:56:08",
   "organization": "Thinking Machines Lab",
   "model": "Inkling",
   "release_date": "2026-07-15",
   "access_type": "open-weight",
   "license": null,
   "significance": "a 975-billion-parameter mixture-of-experts model that tops US open-weight benchmarks",
   "organization_evidence": [
    "# Mira Murati's Thinking Machines Lab Releases Inkling, Its First AI Model, a 975-Billion-Parameter Open-Weight System",
    "Thinking Machines Lab shipped its first model, Inkling, an open-weight 975-billion-parameter system that tops US open-weight benchmarks but still trails China's leading models.",
    "Thinking Machines Lab, the AI startup founded by former OpenAI chief technology officer Mira Murati, released its first in-house model on July 15, 2026, calling it [Inkling](https://thinkingmachines.ai/news/introducing-inkling/) in the company\u2019s own announcement."
   ],
   "model_evidence": [
    "# Mira Murati's Thinking Machines Lab Releases Inkling, Its First AI Model, a 975-Billion-Parameter Open-Weight System",
    "Thinking Machines Lab shipped its first model, Inkling, an open-weight 975-billion-parameter system that tops US open-weight benchmarks but still trails China's leading models.",
    "Thinking Machines Lab, the AI startup founded by former OpenAI chief technology officer Mira Murati, released its first in-house model on July 15, 2026, calling it [Inkling](https://thinkingmachines.ai/news/introducing-inkling/) in the company\u2019s own announcement."
   ],
   "release_date_evidence": [
    "# Mira Murati's Thinking Machines Lab Releases Inkling, Its First AI Model, a 975-Billion-Parameter Open-Weight System",
    "Thinking Machines Lab, the AI startup founded by former OpenAI chief technology officer Mira Murati, released its first in-house model on July 15, 2026, calling it [Inkling](https://thinkingmachines.ai/news/introducing-inkling/) in the company\u2019s own announcement."
   ],
   "access_type_evidence": [
    "# Mira Murati's Thinking Machines Lab Releases Inkling, Its First AI Model, a 975-Billion-Parameter Open-Weight System",
    "Thinking Machines Lab shipped its first model, Inkling, an open-weight 975-billion-parameter system that tops US open-weight benchmarks but still trails China's leading models."
   ],
   "license_evidence": null,
   "significance_evidence": [
    "# Mira Murati's Thinking Machines Lab Releases Inkling, Its First AI Model, a 975-Billion-Parameter Open-Weight System",
    "Thinking Machines Lab shipped its first model, Inkling, an open-weight 975-billion-parameter system that tops US open-weight benchmarks but still trails China's leading models."
   ]
  },
  {
   "url": "https://arxiv.org/pdf/2604.07190v1.pdf",
   "title": "The ATOM Report: Measuring the Open Language Model Ecosystem",
   "published_at": null,
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-02",
   "access_type": null,
   "license": null,
   "significance": "nearly doubling the downloads of the Llama models and demonstrating continuing ecosystem dominance",
   "organization_evidence": [
    "The ATOM Report:\nMeasuring the Open Language Model Ecosystem",
    "1\nIntroduction",
    "The growth of the Chinese\nmodel ecosystem can largely be attributed to Alibaba\u2019s Qwen, which is responsible for almost\na billion cumulative downloads by March 2026."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The ATOM Report:\nMeasuring the Open Language Model Ecosystem",
    "5\nModel Adoption by Organization",
    "The early adoption numbers of Qwen3.5 (Qwen Team, 2026) since its release in February 2026 are\nan indication that the dominance of Qwen relative to its peers will continue (see Section 6 and\nspecifically Fig. 15 for early data on Qwen 3.5\u2019s adoption)."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The ATOM Report:\nMeasuring the Open Language Model Ecosystem",
    "5\nModel Adoption by Organization",
    "5.1\nEcosystem Leaders (Qwen, Llama, DeepSeek, Mistral, and OpenAI)",
    "Qwen surpassed Llama in cumulative downloads in\nSeptember 2025 (325.4M vs. 323.7M) and by March 2026 reached 942.1M, thus nearly doubling the\ndownloads of the Llama models (476.0M)."
   ]
  },
  {
   "url": "https://dev.to/docdavkitty/kimi-k3-moonshot-ai-drops-a-28-trillion-parameter-open-model-and-it-codes-designs-chips-and-48c9",
   "title": "Kimi K3: Moonshot AI Drops a 2.8-Trillion-Parameter Open Model \u2014 and It Codes, Designs Chips, and Edits Video",
   "published_at": "2026-07-17T12:05:01",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-16",
   "access_type": null,
   "license": null,
   "significance": "a 2.8-trillion-parameter Mixture-of-Experts model with a 1-million-token context window and native vision",
   "organization_evidence": [
    "# Kimi K3: Moonshot AI Drops a 2.8-Trillion-Parameter Open Model \u2014 and It Codes, Designs Chips, and Edits Video",
    "Late on July 16, Chinese AI lab Moonshot AI published a blog post announcing Kimi K3 \u2014 a 2.8-trillion-parameter sparse Mixture-of-Experts model that instantly became the largest open-weight model in existence, roughly 1.75\u00d7 the size of DeepSeek V4 Pro's 1.6T parameters."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Kimi K3: Moonshot AI Drops a 2.8-Trillion-Parameter Open Model \u2014 and It Codes, Designs Chips, and Edits Video",
    "**TL;DR** \u2014 Moonshot AI released Kimi K3 on July 16, a 2.8-trillion-parameter Mixture-of-Experts model with a 1-million-token context window and native vision."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Kimi K3: Moonshot AI Drops a 2.8-Trillion-Parameter Open Model \u2014 and It Codes, Designs Chips, and Edits Video",
    "**TL;DR** \u2014 Moonshot AI released Kimi K3 on July 16, a 2.8-trillion-parameter Mixture-of-Experts model with a 1-million-token context window and native vision."
   ]
  },
  {
   "url": "https://dev.to/rishi_kora/inkling-americas-largest-open-weight-model-ships-apache-20-2cm3",
   "title": "Inkling: America's Largest Open-Weight Model Ships Apache 2.0",
   "published_at": "2026-07-27T07:30:23",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "975 billion total parameters in a mixture-of-experts architecture, with 41 billion active per token.",
   "organization_evidence": [
    "What changed on 15 July Thinking Machines Lab \u2014 the research company founded by former OpenAI chief technology officer Mira Murati \u2014 released Inkling on 15 July 2026."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "What changed on 15 July Thinking Machines Lab \u2014 the research company founded by former OpenAI chief technology officer Mira Murati \u2014 released Inkling on 15 July 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "975 billion total parameters in a mixture-of-experts architecture, with 41 billion active per token."
   ]
  },
  {
   "url": "https://webpronews.com/thinking-machines-inkling-challenges-dominance-of-chinese-open-weight-models",
   "title": "Thinking Machines\u2019 Inkling Challenges Dominance of Chinese Open-Weight Models",
   "published_at": "2026-07-15T19:07:00",
   "organization": "Thinking Machines",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "975B-parameter open-weight multimodal model trained on 45T tokens with 1 million context window",
   "organization_evidence": [
    "Thinking Machines\u2019 Inkling Challenges Dominance of Chinese Open-Weight Models",
    "Thinking Machines launched Inkling on July 15."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Thinking Machines launched Inkling on July 15.",
    "Wednesday, July 15, 2026"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Thinking Machines, founded by Mira Murati, released Inkling, a 975B-parameter open-weight multimodal model trained on 45T tokens.",
    "Thinking Machines\u2019 Inkling Challenges Dominance of Chinese Open-Weight Models"
   ]
  },
  {
   "url": "https://cirra.ai/articles/kimi-k3-release-open-weight-models",
   "title": "Kimi K3 Release: Impact on Frontier Open-Weight Models",
   "published_at": "2026-07-26T09:17:04",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-16",
   "access_type": null,
   "license": null,
   "significance": "a 2.8 trillion parameter mixture of experts language model",
   "organization_evidence": [
    "Cirra AI Article",
    "Kimi K3 Release: Impact on Frontier Open-Weight Models",
    "On July 16, 2026, Beijing based startup **Moonshot AI** introduced **Kimi K3**, a **2.8 trillion parameter** mixture of experts language model that VentureBeat described as \"a frontier-class large language model with 2.8 trillion total parameters\""
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Cirra AI Article",
    "Kimi K3 Release: Impact on Frontier Open-Weight Models",
    "On July 16, 2026, Beijing based startup **Moonshot AI** introduced **Kimi K3**, a **2.8 trillion parameter** mixture of experts language model that VentureBeat described as \"a frontier-class large language model with 2.8 trillion total parameters\""
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Cirra AI Article",
    "Kimi K3 Release: Impact on Frontier Open-Weight Models",
    "On July 16, 2026, Beijing based startup **Moonshot AI** introduced **Kimi K3**, a **2.8 trillion parameter** mixture of experts language model that VentureBeat described as \"a frontier-class large language model with 2.8 trillion total parameters\""
   ]
  },
  {
   "url": "https://computerworld.com/article/4197755/thinking-machines-lab-offers-enterprises-a-us-alternative-in-open-weight-ai.html",
   "title": "Thinking Machines Lab offers enterprises a US alternative in open-weight AI",
   "published_at": "2026-07-16T05:36:21",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-16",
   "access_type": null,
   "license": null,
   "significance": "A mixture-of-experts model with 975 billion total parameters and a 1-million-token context window",
   "organization_evidence": [
    "# Thinking Machines Lab offers enterprises a US alternative in open-weight AI",
    "Thinking Machines Lab, the San Francisco startup founded by former OpenAI [CTO Mira Murati](https://www.computerworld.com/article/3829004/ex-openai-cto-mira-murati-launches-ai-startup-recruits-top-talent-from-rivals.html), has released Inkling, its first general-purpose AI model."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Jul 16, 20265 mins",
    "# Thinking Machines Lab offers enterprises a US alternative in open-weight AI",
    "Thinking Machines Lab, the San Francisco startup founded by former OpenAI [CTO Mira Murati](https://www.computerworld.com/article/3829004/ex-openai-cto-mira-murati-launches-ai-startup-recruits-top-talent-from-rivals.html), has released Inkling, its first general-purpose AI model."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## Inkling, a 975-billion-parameter model, can be customized through Tinker and supports a 1-million-token context window, but it enters a market where Chinese models lead several coding and reasoning benchmarks.",
    "Inkling uses a mixture-of-experts architecture with 975 billion total parameters, of which 41 billion are active during processing.",
    "# Thinking Machines Lab offers enterprises a US alternative in open-weight AI"
   ]
  },
  {
   "url": "https://fazm.ai/blog/open-source-llm-releases-2026",
   "title": "Fazm - AI Computer Agent for macOS | Voice-Controlled Desktop Automation",
   "published_at": "2026-04-11T01:54:54",
   "organization": "DeepSeek",
   "model": "DeepSeek V3-0324",
   "release_date": "2026-01-01",
   "access_type": "open-weight",
   "license": "MIT",
   "significance": "improved the already strong DeepSeek V3 with better instruction following and coding performance",
   "organization_evidence": [
    "## 2026 Open Source LLM Release Timeline",
    "| DeepSeek V3-0324 | DeepSeek | Jan 2026 | 685B (37B active) | MoE | MIT | 128K |"
   ],
   "model_evidence": [
    "## 2026 Open Source LLM Release Timeline",
    "| DeepSeek V3-0324 | DeepSeek | Jan 2026 | 685B (37B active) | MoE | MIT | 128K |"
   ],
   "release_date_evidence": [
    "## 2026 Open Source LLM Release Timeline",
    "| DeepSeek V3-0324 | DeepSeek | Jan 2026 | 685B (37B active) | MoE | MIT | 128K |"
   ],
   "access_type_evidence": [
    "## 2026 Open Source LLM Release Timeline",
    "\"Open source\" here includes open-weight models with permissive or semi-permissive licenses."
   ],
   "license_evidence": [
    "## 2026 Open Source LLM Release Timeline",
    "| DeepSeek V3-0324 | DeepSeek | Jan 2026 | 685B (37B active) | MoE | MIT | 128K |"
   ],
   "significance_evidence": [
    "# Open Source LLM Releases 2026: Every Major Model So Far",
    "The V3-0324 update in January improved the already strong DeepSeek V3 with better instruction following and coding performance."
   ]
  },
  {
   "url": "https://fazm.ai/blog/open-source-llm-releases-2026",
   "title": "Fazm - AI Computer Agent for macOS | Voice-Controlled Desktop Automation",
   "published_at": "2026-04-11T01:54:54",
   "organization": "Alibaba",
   "model": "Qwen 2.5-Max",
   "release_date": "2026-01-01",
   "access_type": "open-weight",
   "license": "Qwen License",
   "significance": "large dense model that competes with GPT-4o on benchmarks like Arena-Hard and LiveBench",
   "organization_evidence": [
    "## 2026 Open Source LLM Release Timeline",
    "| Qwen 2.5-Max | Alibaba | Jan 2026 | Undisclosed | Dense | Qwen License | 128K |"
   ],
   "model_evidence": [
    "## 2026 Open Source LLM Release Timeline",
    "| Qwen 2.5-Max | Alibaba | Jan 2026 | Undisclosed | Dense | Qwen License | 128K |"
   ],
   "release_date_evidence": [
    "## 2026 Open Source LLM Release Timeline",
    "| Qwen 2.5-Max | Alibaba | Jan 2026 | Undisclosed | Dense | Qwen License | 128K |"
   ],
   "access_type_evidence": [
    "## 2026 Open Source LLM Release Timeline",
    "\"Open source\" here includes open-weight models with permissive or semi-permissive licenses."
   ],
   "license_evidence": [
    "## 2026 Open Source LLM Release Timeline",
    "| Qwen 2.5-Max | Alibaba | Jan 2026 | Undisclosed | Dense | Qwen License | 128K |"
   ],
   "significance_evidence": [
    "# Open Source LLM Releases 2026: Every Major Model So Far",
    "Qwen 2.5-Max, released in January, is a large dense model that competes with GPT-4o on benchmarks like Arena-Hard and LiveBench."
   ]
  },
  {
   "url": "https://fazm.ai/blog/open-source-llm-releases-2026",
   "title": "Fazm - AI Computer Agent for macOS | Voice-Controlled Desktop Automation",
   "published_at": "2026-04-11T01:54:54",
   "organization": "Google",
   "model": "Gemma 3",
   "release_date": "2026-01-01",
   "access_type": "open-weight",
   "license": "Gemma License",
   "significance": "notable for its efficiency at smaller scales",
   "organization_evidence": [
    "## 2026 Open Source LLM Release Timeline",
    "| Gemma 3 | Google | Jan 2026 | 1B / 4B / 12B / 27B | Dense | Gemma License | 128K |"
   ],
   "model_evidence": [
    "## 2026 Open Source LLM Release Timeline",
    "| Gemma 3 | Google | Jan 2026 | 1B / 4B / 12B / 27B | Dense | Gemma License | 128K |"
   ],
   "release_date_evidence": [
    "## 2026 Open Source LLM Release Timeline",
    "| Gemma 3 | Google | Jan 2026 | 1B / 4B / 12B / 27B | Dense | Gemma License | 128K |"
   ],
   "access_type_evidence": [
    "## 2026 Open Source LLM Release Timeline",
    "\"Open source\" here includes open-weight models with permissive or semi-permissive licenses."
   ],
   "license_evidence": [
    "## 2026 Open Source LLM Release Timeline",
    "| Gemma 3 | Google | Jan 2026 | 1B / 4B / 12B / 27B | Dense | Gemma License | 128K |"
   ],
   "significance_evidence": [
    "# Open Source LLM Releases 2026: Every Major Model So Far",
    "The Gemma 3 family is notable for its efficiency at smaller scales."
   ]
  },
  {
   "url": "https://fazm.ai/blog/open-source-llm-releases-2026",
   "title": "Fazm - AI Computer Agent for macOS | Voice-Controlled Desktop Automation",
   "published_at": "2026-04-11T01:54:54",
   "organization": "Mistral",
   "model": "Mistral Small 3.1",
   "release_date": "2026-02-01",
   "access_type": "open-weight",
   "license": "Apache 2.0",
   "significance": "adds vision capabilities and extends the context window to 128K tokens",
   "organization_evidence": [
    "## 2026 Open Source LLM Release Timeline",
    "| Mistral Small 3.1 | Mistral | Feb 2026 | 24B | Dense | Apache 2.0 | 128K |"
   ],
   "model_evidence": [
    "## 2026 Open Source LLM Release Timeline",
    "| Mistral Small 3.1 | Mistral | Feb 2026 | 24B | Dense | Apache 2.0 | 128K |"
   ],
   "release_date_evidence": [
    "## 2026 Open Source LLM Release Timeline",
    "| Mistral Small 3.1 | Mistral | Feb 2026 | 24B | Dense | Apache 2.0 | 128K |"
   ],
   "access_type_evidence": [
    "## 2026 Open Source LLM Release Timeline",
    "\"Open source\" here includes open-weight models with permissive or semi-permissive licenses."
   ],
   "license_evidence": [
    "## 2026 Open Source LLM Release Timeline",
    "| Mistral Small 3.1 | Mistral | Feb 2026 | 24B | Dense | Apache 2.0 | 128K |"
   ],
   "significance_evidence": [
    "## Mistral Small 3.1 and Mistral Large 3",
    "**Mistral Small 3.1** (February, 24B parameters, Apache 2.0) adds vision capabilities and extends the context window to 128K tokens."
   ]
  },
  {
   "url": "https://fazm.ai/blog/open-source-llm-releases-2026",
   "title": "Fazm - AI Computer Agent for macOS | Voice-Controlled Desktop Automation",
   "published_at": "2026-04-11T01:54:54",
   "organization": "Cohere",
   "model": "Command A",
   "release_date": "2026-03-01",
   "access_type": "open-weight",
   "license": "CC-BY-NC",
   "significance": "optimized for retrieval-augmented generation (RAG) workflows, with strong performance on tasks that require synthesizing information from long documents",
   "organization_evidence": [
    "## 2026 Open Source LLM Release Timeline",
    "| Command A | Cohere | Mar 2026 | 111B (23B active) | MoE | CC-BY-NC | 256K |"
   ],
   "model_evidence": [
    "## 2026 Open Source LLM Release Timeline",
    "| Command A | Cohere | Mar 2026 | 111B (23B active) | MoE | CC-BY-NC | 256K |"
   ],
   "release_date_evidence": [
    "## 2026 Open Source LLM Release Timeline",
    "| Command A | Cohere | Mar 2026 | 111B (23B active) | MoE | CC-BY-NC | 256K |"
   ],
   "access_type_evidence": [
    "## 2026 Open Source LLM Release Timeline",
    "\"Open source\" here includes open-weight models with permissive or semi-permissive licenses."
   ],
   "license_evidence": [
    "## 2026 Open Source LLM Release Timeline",
    "| Command A | Cohere | Mar 2026 | 111B (23B active) | MoE | CC-BY-NC | 256K |"
   ],
   "significance_evidence": [
    "## Command A (Cohere)",
    "Command A is optimized for retrieval-augmented generation (RAG) workflows, with strong performance on tasks that require synthesizing information from long documents."
   ]
  },
  {
   "url": "https://fazm.ai/blog/open-source-llm-releases-2026",
   "title": "Fazm - AI Computer Agent for macOS | Voice-Controlled Desktop Automation",
   "published_at": "2026-04-11T01:54:54",
   "organization": "Meta",
   "model": "Llama 4 Scout",
   "release_date": "2026-04-01",
   "access_type": "open-weight",
   "license": "Llama 4 License",
   "significance": "features a 10 million token context window, the longest of any open-weight model",
   "organization_evidence": [
    "## 2026 Open Source LLM Release Timeline",
    "| Llama 4 Scout | Meta | Apr 2026 | 109B (17B active) | MoE | Llama 4 License | 10M |"
   ],
   "model_evidence": [
    "## 2026 Open Source LLM Release Timeline",
    "| Llama 4 Scout | Meta | Apr 2026 | 109B (17B active) | MoE | Llama 4 License | 10M |"
   ],
   "release_date_evidence": [
    "## 2026 Open Source LLM Release Timeline",
    "| Llama 4 Scout | Meta | Apr 2026 | 109B (17B active) | MoE | Llama 4 License | 10M |"
   ],
   "access_type_evidence": [
    "## 2026 Open Source LLM Release Timeline",
    "\"Open source\" here includes open-weight models with permissive or semi-permissive licenses."
   ],
   "license_evidence": [
    "## 2026 Open Source LLM Release Timeline",
    "| Llama 4 Scout | Meta | Apr 2026 | 109B (17B active) | MoE | Llama 4 License | 10M |"
   ],
   "significance_evidence": [
    "## Llama 4: Scout and Maverick",
    "The headline feature is a 10 million token context window, the longest of any open-weight model."
   ]
  },
  {
   "url": "https://fazm.ai/blog/open-source-llm-releases-2026",
   "title": "Fazm - AI Computer Agent for macOS | Voice-Controlled Desktop Automation",
   "published_at": "2026-04-11T01:54:54",
   "organization": "Meta",
   "model": "Llama 4 Maverick",
   "release_date": "2026-04-01",
   "access_type": "open-weight",
   "license": "Llama 4 License",
   "significance": "achieves performance competitive with GPT-4o and Gemini 2.0 Pro on standard benchmarks",
   "organization_evidence": [
    "## 2026 Open Source LLM Release Timeline",
    "| Llama 4 Maverick | Meta | Apr 2026 | 400B (17B active) | MoE | Llama 4 License | 1M |"
   ],
   "model_evidence": [
    "## 2026 Open Source LLM Release Timeline",
    "| Llama 4 Maverick | Meta | Apr 2026 | 400B (17B active) | MoE | Llama 4 License | 1M |"
   ],
   "release_date_evidence": [
    "## 2026 Open Source LLM Release Timeline",
    "| Llama 4 Maverick | Meta | Apr 2026 | 400B (17B active) | MoE | Llama 4 License | 1M |"
   ],
   "access_type_evidence": [
    "## 2026 Open Source LLM Release Timeline",
    "\"Open source\" here includes open-weight models with permissive or semi-permissive licenses."
   ],
   "license_evidence": [
    "## 2026 Open Source LLM Release Timeline",
    "| Llama 4 Maverick | Meta | Apr 2026 | 400B (17B active) | MoE | Llama 4 License | 1M |"
   ],
   "significance_evidence": [
    "## Llama 4: Scout and Maverick",
    "It targets a 1 million token context window and achieves performance competitive with GPT-4o and Gemini 2.0 Pro on standard benchmarks."
   ]
  },
  {
   "url": "https://fazm.ai/blog/open-source-llm-releases-2026",
   "title": "Fazm - AI Computer Agent for macOS | Voice-Controlled Desktop Automation",
   "published_at": "2026-04-11T01:54:54",
   "organization": "Alibaba",
   "model": "QwQ-32B",
   "release_date": "2026-03-01",
   "access_type": "open-weight",
   "license": "Apache 2.0",
   "significance": "matches or exceeds much larger models on reasoning benchmarks at a fraction of the compute cost",
   "organization_evidence": [
    "## 2026 Open Source LLM Release Timeline",
    "| QwQ-32B | Alibaba | Mar 2026 | 32B | Dense | Apache 2.0 | 128K |"
   ],
   "model_evidence": [
    "## 2026 Open Source LLM Release Timeline",
    "| QwQ-32B | Alibaba | Mar 2026 | 32B | Dense | Apache 2.0 | 128K |"
   ],
   "release_date_evidence": [
    "## 2026 Open Source LLM Release Timeline",
    "| QwQ-32B | Alibaba | Mar 2026 | 32B | Dense | Apache 2.0 | 128K |"
   ],
   "access_type_evidence": [
    "## 2026 Open Source LLM Release Timeline",
    "\"Open source\" here includes open-weight models with permissive or semi-permissive licenses."
   ],
   "license_evidence": [
    "## 2026 Open Source LLM Release Timeline",
    "| QwQ-32B | Alibaba | Mar 2026 | 32B | Dense | Apache 2.0 | 128K |"
   ],
   "significance_evidence": [
    "## Qwen 2.5-Max and QwQ-32B",
    "Despite its relatively small size, it matches or exceeds much larger models on reasoning benchmarks."
   ]
  },
  {
   "url": "https://fazm.ai/blog/open-source-llm-releases-2026",
   "title": "Fazm - AI Computer Agent for macOS | Voice-Controlled Desktop Automation",
   "published_at": "2026-04-11T01:54:54",
   "organization": "DeepSeek",
   "model": "DeepSeek R2",
   "release_date": "2026-03-01",
   "access_type": "open-weight",
   "license": "MIT",
   "significance": "is a reasoning-focused model built on the V3 architecture",
   "organization_evidence": [
    "## 2026 Open Source LLM Release Timeline",
    "| DeepSeek R2 | DeepSeek | Mar 2026 | 685B (37B active) | MoE | MIT | 128K |"
   ],
   "model_evidence": [
    "## 2026 Open Source LLM Release Timeline",
    "| DeepSeek R2 | DeepSeek | Mar 2026 | 685B (37B active) | MoE | MIT | 128K |"
   ],
   "release_date_evidence": [
    "## 2026 Open Source LLM Release Timeline",
    "| DeepSeek R2 | DeepSeek | Mar 2026 | 685B (37B active) | MoE | MIT | 128K |"
   ],
   "access_type_evidence": [
    "## 2026 Open Source LLM Release Timeline",
    "\"Open source\" here includes open-weight models with permissive or semi-permissive licenses."
   ],
   "license_evidence": [
    "## 2026 Open Source LLM Release Timeline",
    "| DeepSeek R2 | DeepSeek | Mar 2026 | 685B (37B active) | MoE | MIT | 128K |"
   ],
   "significance_evidence": [
    "## DeepSeek V3 and R2",
    "DeepSeek R2, released in March, is a reasoning-focused model built on the V3 architecture."
   ]
  },
  {
   "url": "https://fazm.ai/blog/open-source-llm-releases-2026",
   "title": "Fazm - AI Computer Agent for macOS | Voice-Controlled Desktop Automation",
   "published_at": "2026-04-11T01:54:54",
   "organization": "Mistral",
   "model": "Mistral Large 3",
   "release_date": "2026-04-01",
   "access_type": "open-weight",
   "license": "Mistral Research",
   "significance": "is the company's flagship open-weight model competing directly with Llama 4 Maverick and GPT-4o",
   "organization_evidence": [
    "## 2026 Open Source LLM Release Timeline",
    "| Mistral Large 3 | Mistral | Apr 2026 | 123B | Dense | Mistral Research | 128K |"
   ],
   "model_evidence": [
    "## 2026 Open Source LLM Release Timeline",
    "| Mistral Large 3 | Mistral | Apr 2026 | 123B | Dense | Mistral Research | 128K |"
   ],
   "release_date_evidence": [
    "## 2026 Open Source LLM Release Timeline",
    "| Mistral Large 3 | Mistral | Apr 2026 | 123B | Dense | Mistral Research | 128K |"
   ],
   "access_type_evidence": [
    "## 2026 Open Source LLM Release Timeline",
    "\"Open source\" here includes open-weight models with permissive or semi-permissive licenses."
   ],
   "license_evidence": [
    "## 2026 Open Source LLM Release Timeline",
    "| Mistral Large 3 | Mistral | Apr 2026 | 123B | Dense | Mistral Research | 128K |"
   ],
   "significance_evidence": [
    "## Mistral Small 3.1 and Mistral Large 3",
    "**Mistral Large 3** (April, 123B parameters, Mistral Research License) is the company's flagship open-weight model."
   ]
  },
  {
   "url": "https://fazm.ai/t/large-language-model-releases-may-2026",
   "title": "Large language model releases in May 2026: the full calendar, and how a Mac agent actually runs each one",
   "published_at": "2026-05-29T00:00:00",
   "organization": "OpenBMB",
   "model": null,
   "release_date": "2026-05-11",
   "access_type": null,
   "license": null,
   "significance": "1.3B multimodal model optimized for on-device inference with 262K context",
   "organization_evidence": [
    "# Large language model releases in May 2026",
    "## The large language models released in May 2026",
    "| Model | Date | Type | Context |",
    "| MiniCPM-V 4.6 1.3B | May 11, 2026 | Apache 2.0 | 262K |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Large language model releases in May 2026",
    "## The large language models released in May 2026",
    "| Model | Date | Type | Context |",
    "| MiniCPM-V 4.6 1.3B | May 11, 2026 | Apache 2.0 | 262K |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Large language model releases in May 2026",
    "## The large language models released in May 2026",
    "| Model | Date | Type | Context |",
    "| MiniCPM-V 4.6 1.3B | May 11, 2026 | Apache 2.0 | 262K |",
    "MiniCPM-V 4.6 1.3B is a 1.3B multimodal model optimized for on-device inference; on a Mac it is a fit for an always-on screen-understanding pass, not the multi-step planning loop."
   ]
  },
  {
   "url": "https://fazm.ai/t/large-language-model-releases-may-2026",
   "title": "Large language model releases in May 2026: the full calendar, and how a Mac agent actually runs each one",
   "published_at": "2026-05-29T00:00:00",
   "organization": "Cohere",
   "model": null,
   "release_date": "2026-05-20",
   "access_type": null,
   "license": null,
   "significance": "218B total / 25B active Sparse MoE with 128K input context under Apache 2.0",
   "organization_evidence": [
    "# Large language model releases in May 2026",
    "## The large language models released in May 2026",
    "| Model | Date | Type | Context |",
    "| Command A+ | May 20, 2026 | Apache 2.0 | 128K |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Large language model releases in May 2026",
    "## The large language models released in May 2026",
    "| Model | Date | Type | Context |",
    "| Command A+ | May 20, 2026 | Apache 2.0 | 128K |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Large language model releases in May 2026",
    "## The large language models released in May 2026",
    "| Model | Date | Type | Context |",
    "| Command A+ | May 20, 2026 | Apache 2.0 | 128K |",
    "Command A+ is a 218B / 25B-active Sparse MoE with 128K input context under Apache 2.0, shipped on Hugging Face in BF16, FP8, and W4A4 quantizations on day one, so it can run on as few as two H100s."
   ]
  },
  {
   "url": "https://fazm.ai/t/large-language-model-releases-may-2026",
   "title": "Large language model releases in May 2026: the full calendar, and how a Mac agent actually runs each one",
   "published_at": "2026-05-29T00:00:00",
   "organization": "Xiaomi",
   "model": null,
   "release_date": "2026-04-28",
   "access_type": null,
   "license": null,
   "significance": "Largest by total parameters (1.02T MoE, 42B active) and the most permissive frontier-class license (MIT)",
   "organization_evidence": [
    "# Large language model releases in May 2026",
    "## The large language models released in May 2026",
    "| Model | Date | Type | Context |",
    "| MiMo-V2.5-Pro | Apr 28, 2026 | MIT | 256K |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Large language model releases in May 2026",
    "## The large language models released in May 2026",
    "| Model | Date | Type | Context |",
    "| MiMo-V2.5-Pro | Apr 28, 2026 | MIT | 256K |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Large language model releases in May 2026",
    "## The late-April cluster still setting the bar",
    "The shortest honest read: Xiaomi MiMo-V2.5-Pro is the largest by total parameters (1.02T MoE, 42B active) and the most permissive frontier-class license (MIT)."
   ]
  },
  {
   "url": "https://fazm.ai/t/large-language-model-releases-may-2026",
   "title": "Large language model releases in May 2026: the full calendar, and how a Mac agent actually runs each one",
   "published_at": "2026-05-29T00:00:00",
   "organization": "NVIDIA",
   "model": null,
   "release_date": "2026-04-28",
   "access_type": null,
   "license": null,
   "significance": "Unifies vision, speech, and language, activates only 3B parameters per token, and runs in 25GB of RAM",
   "organization_evidence": [
    "# Large language model releases in May 2026",
    "## The large language models released in May 2026",
    "| Model | Date | Type | Context |",
    "| Nemotron 3 Nano Omni | Apr 28, 2026 | NVIDIA Open Model | 128K |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Large language model releases in May 2026",
    "## The large language models released in May 2026",
    "| Model | Date | Type | Context |",
    "| Nemotron 3 Nano Omni | Apr 28, 2026 | NVIDIA Open Model | 128K |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Large language model releases in May 2026",
    "## The late-April cluster still setting the bar",
    "NVIDIA Nemotron 3 Nano Omni unifies vision, speech, and language, activates only 3B parameters per token, and runs in 25GB of RAM."
   ]
  },
  {
   "url": "https://fazm.ai/t/large-language-model-releases-may-2026",
   "title": "Large language model releases in May 2026: the full calendar, and how a Mac agent actually runs each one",
   "published_at": "2026-05-29T00:00:00",
   "organization": "IBM",
   "model": null,
   "release_date": "2026-04-29",
   "access_type": null,
   "license": null,
   "significance": "Clean Apache 2.0 dense pick (3B / 8B / 30B, up to 512K context, twelve languages)",
   "organization_evidence": [
    "# Large language model releases in May 2026",
    "## The large language models released in May 2026",
    "| Model | Date | Type | Context |",
    "| IBM Granite 4.1 | Apr 29, 2026 | Apache 2.0 | up to 512K |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Large language model releases in May 2026",
    "## The large language models released in May 2026",
    "| Model | Date | Type | Context |",
    "| IBM Granite 4.1 | Apr 29, 2026 | Apache 2.0 | up to 512K |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Large language model releases in May 2026",
    "## The late-April cluster still setting the bar",
    "IBM Granite 4.1 is the clean Apache 2.0 dense pick (3B / 8B / 30B, up to 512K context, twelve languages)."
   ]
  },
  {
   "url": "https://fazm.ai/t/large-language-model-releases-may-2026",
   "title": "Large language model releases in May 2026: the full calendar, and how a Mac agent actually runs each one",
   "published_at": "2026-05-29T00:00:00",
   "organization": "Mistral AI",
   "model": null,
   "release_date": "2026-04-29",
   "access_type": null,
   "license": null,
   "significance": "128B dense coder with 256K context under a Modified MIT license",
   "organization_evidence": [
    "# Large language model releases in May 2026",
    "## The large language models released in May 2026",
    "| Model | Date | Type | Context |",
    "| Mistral Medium 3.5 | Apr 29, 2026 | Modified MIT | 256K |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Large language model releases in May 2026",
    "## The large language models released in May 2026",
    "| Model | Date | Type | Context |",
    "| Mistral Medium 3.5 | Apr 29, 2026 | Modified MIT | 256K |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Large language model releases in May 2026",
    "## The late-April cluster still setting the bar",
    "Mistral Medium 3.5 is the 128B dense coder (256K context, 77.6% on SWE-Bench Verified) under a Modified MIT license you must read before deploying inside a large company."
   ]
  },
  {
   "url": "https://aimagazine.blog/posts/975b-parameters-full-weights-free-to-fine-tune-thinking-machines-just-changed-the-open-model-race",
   "title": "975B Open-Weight Model Inkling Drops \u2014 and It Fine-Tunes Itself",
   "published_at": "2026-07-20T04:54:03",
   "organization": "Thinking Machines",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "975B-parameter Mixture-of-Experts model with 41B active parameters per token, 1 million token context window, native multimodal capabilities, and a tunable effort parameter",
   "organization_evidence": [
    "# 975B Parameters, Full Weights, Free to Fine-Tune \u2014 Thinking Machines Just Changed the Open Model Race",
    "Thinking Machines, the AI lab behind the Tinker customization platform, is not playing by those rules."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "On July 15, Thinking Machines released **Inkling** \u2014 a 975B-parameter Mixture-of-Experts model with 41B active parameters per token, trained from scratch on 45 trillion tokens spanning text, images, audio, and video.",
    "# 975B Parameters, Full Weights, Free to Fine-Tune \u2014 Thinking Machines Just Changed the Open Model Race",
    "Published Jul 20, 2026 \u2022 6 min read"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "On July 15, Thinking Machines released **Inkling** \u2014 a 975B-parameter Mixture-of-Experts model with 41B active parameters per token, trained from scratch on 45 trillion tokens spanning text, images, audio, and video.",
    "The context window extends to **1 million tokens**.",
    "# 975B Parameters, Full Weights, Free to Fine-Tune \u2014 Thinking Machines Just Changed the Open Model Race",
    "Published Jul 20, 2026 \u2022 6 min read"
   ]
  },
  {
   "url": "https://recatools.com/news/mira-muratis-thinking-machines-releases-inkling-its-first-open-weight-model",
   "title": "Mira Murati's Thinking Machines Releases Inkling, Its First Open-Weight Model",
   "published_at": "2026-07-21T09:41:30",
   "organization": "Thinking Machines Lab",
   "model": "Inkling",
   "release_date": "2026-07-15",
   "access_type": "open-weight",
   "license": "Apache 2.0",
   "significance": "An open-weight, multimodal Mixture-of-Experts system with 975 billion total parameters and up to 1 million tokens of context.",
   "organization_evidence": [
    "# Mira Murati's Thinking Machines Releases Inkling, Its First Open-Weight Model",
    "Thinking Machines Lab, the company founded by former OpenAI chief technology officer Mira Murati, released its first in-house model on 15 July 2026."
   ],
   "model_evidence": [
    "# Mira Murati's Thinking Machines Releases Inkling, Its First Open-Weight Model",
    "Mira Murati's Thinking Machines has released Inkling, its first model: an open-weight, multimodal Mixture-of-Experts system with 975 billion parameters that the lab pitches as a base for customisation rather than the strongest model on the leaderboard."
   ],
   "release_date_evidence": [
    "# Mira Murati's Thinking Machines Releases Inkling, Its First Open-Weight Model",
    "Thinking Machines Lab, the company founded by former OpenAI chief technology officer Mira Murati, released its first in-house model on 15 July 2026."
   ],
   "access_type_evidence": [
    "# Mira Murati's Thinking Machines Releases Inkling, Its First Open-Weight Model",
    "Mira Murati's Thinking Machines has released Inkling, its first model: an open-weight, multimodal Mixture-of-Experts system with 975 billion parameters that the lab pitches as a base for customisation rather than the strongest model on the leaderboard."
   ],
   "license_evidence": [
    "# Mira Murati's Thinking Machines Releases Inkling, Its First Open-Weight Model",
    "The full weights are published on Hugging Face under an Apache 2.0 licence, and the company is explicit that Inkling is not built to top the leaderboards."
   ],
   "significance_evidence": [
    "# Mira Murati's Thinking Machines Releases Inkling, Its First Open-Weight Model",
    "Mira Murati's Thinking Machines has released Inkling, its first model: an open-weight, multimodal Mixture-of-Experts system with 975 billion parameters that the lab pitches as a base for customisation rather than the strongest model on the leaderboard."
   ]
  },
  {
   "url": "https://venturebeat.com/technology/ai-joins-the-8-hour-work-day-as-glm-ships-5-1-open-source-llm-beating-opus-4",
   "title": "AI joins the 8-hour work day as GLM ships 5.1 open source LLM, beating Opus 4.6 and GPT-5.4 on SWE-Bench Pro",
   "published_at": "2026-04-07T21:27:07",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-04-07",
   "access_type": null,
   "license": null,
   "significance": "A 754-billion parameter Mixture-of-Experts model engineered to maintain goal alignment over extended execution traces that span thousands of tool calls.",
   "organization_evidence": [
    "Is China picking back up the open source AI baton?",
    "Z.ai, also known as Zhupai AI, a Chinese AI startup best known for its powerful, open source GLM family of models, has [unveiled GLM-5.1 today](https://z.ai/blog/glm-5.1) under a [permissive MIT License](https://huggingface.co/datasets/choosealicense/licenses/blob/main/markdown/mit.md), allowing for enterprises to download, customize and use it for commercial purposes."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Is China picking back up the open source AI baton?",
    "Z.ai, also known as Zhupai AI, a Chinese AI startup best known for its powerful, open source GLM family of models, has [unveiled GLM-5.1 today](https://z.ai/blog/glm-5.1) under a [permissive MIT License](https://huggingface.co/datasets/choosealicense/licenses/blob/main/markdown/mit.md), allowing for enterprises to download, customize and use it for commercial purposes.",
    "![VectorDBBench graphic from z.ai for GLM-5.1](/_next/image?url=https%3A%2F%2Fimages.ctfassets.net%2Fjdtwqhzvc2n1%2F6m6jseAmPAH28vQGVESQTG%2F8f603174d0b0a4ec95eaf6cd28ebc62c%2FScreenshot_2026-04-07_at_1.41.40%C3%A2__PM.png%3Fw%3D1000%26q%3D100&w=3840&q=75)"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Is China picking back up the open source AI baton?",
    "GLM-5.1 is a 754-billion parameter Mixture-of-Experts model engineered to maintain goal alignment over extended execution traces that span thousands of tool calls."
   ]
  },
  {
   "url": "https://fireworks.ai/blog/best-open-source-llms",
   "title": "Best Open Source LLMs in 2026",
   "published_at": "2026-07-08T00:00:00",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-06",
   "access_type": null,
   "license": null,
   "significance": "Strongest benchmark profile among currently open models in this set. First eval for broad reasoning, coding, and long-context agents.",
   "organization_evidence": [
    "Best Open Source Llms",
    "## GLM 5.2",
    "[GLM-5.2](/models/fireworks/glm-5p2) is [Z.ai's](https://z.ai/blog/glm-5.2) flagship coding model, a 743B-parameter MoE with a 1,040k-token context window on Fireworks."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Best Open Source Llms",
    "## GLM 5.2",
    "- \u2022**Release date:** June 2026"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Best Open Source Llms",
    "## The best open source LLMs at a glance",
    "| [GLM 5.2](/models/fireworks/glm-5p2) | June 2026 | 743B total | 1,040k tokens | Strongest benchmark profile among currently open models in this set. First eval for broad reasoning, coding, and long-context agents. | [Try in playground](/models/fireworks/glm-5p2) |"
   ]
  },
  {
   "url": "https://fireworks.ai/blog/best-open-source-llms",
   "title": "Best Open Source LLMs in 2026",
   "published_at": "2026-07-08T00:00:00",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-16",
   "access_type": null,
   "license": null,
   "significance": "Benchmark leader awaiting full weights and a Fireworks listing. License details remain unpublished.",
   "organization_evidence": [
    "Best Open Source Llms",
    "## Kimi K3",
    "[Kimi K3](https://www.kimi.com/blog/kimi-k3) is Moonshot's 2.8-trillion-parameter mixture-of-experts model."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Best Open Source Llms",
    "## Kimi K3",
    "- \u2022**Release date:** July 16, 2026"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Best Open Source Llms",
    "## The best open source LLMs at a glance",
    "| [Kimi K3](https://www.kimi.com/blog/kimi-k3) | July 2026 | 2.8T total, 16 of 896 experts active | 1M tokens | Benchmark leader awaiting full weights and a Fireworks listing. License details remain unpublished. | Not yet listed |"
   ]
  },
  {
   "url": "https://fireworks.ai/blog/best-open-source-llms",
   "title": "Best Open Source LLMs in 2026",
   "published_at": "2026-07-08T00:00:00",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-06",
   "access_type": null,
   "license": null,
   "significance": "Coding agents, repository work, patch planning, and multimodal developer tools. Thinking mode is mandatory.",
   "organization_evidence": [
    "Best Open Source Llms",
    "## Kimi K2.7 Code",
    "[Kimi K2.7 Code](/models/fireworks/kimi-k2p7-code) is a coding-specialized 1.02T-parameter MoE model from [Moonshot AI](https://www.moonshot.ai/) with a 262k-token context window."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Best Open Source Llms",
    "## The best open source LLMs at a glance",
    "| [Kimi K2.7 Code](/models/fireworks/kimi-k2p7-code) | June 2026 | 1.02T total | 262k tokens | Coding agents, repository work, patch planning, and multimodal developer tools. Thinking mode is mandatory. | [Try in playground](/models/fireworks/kimi-k2p7-code) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Best Open Source Llms",
    "## The best open source LLMs at a glance",
    "| [Kimi K2.7 Code](/models/fireworks/kimi-k2p7-code) | June 2026 | 1.02T total | 262k tokens | Coding agents, repository work, patch planning, and multimodal developer tools. Thinking mode is mandatory. | [Try in playground](/models/fireworks/kimi-k2p7-code) |"
   ]
  },
  {
   "url": "https://fireworks.ai/blog/best-open-source-llms",
   "title": "Best Open Source LLMs in 2026",
   "published_at": "2026-07-08T00:00:00",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "Long-context reasoning and coding from a separate open-source family. Second eval when GLM 5.2 misses on your repo.",
   "organization_evidence": [
    "Best Open Source Llms",
    "## DeepSeek-V4-Pro",
    "[DeepSeek-V4-Pro](/models/fireworks/deepseek-v4-pro) is a large MoE model for reasoning, coding, and long-context work from [DeepSeek](https://www.deepseek.com/)."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Best Open Source Llms",
    "## The best open source LLMs at a glance",
    "| [DeepSeek-V4-Pro](/models/fireworks/deepseek-v4-pro) | April 2026 | 1.6T total | 1,040k tokens | Long-context reasoning and coding from a separate open-source family. Second eval when GLM 5.2 misses on your repo. | [Try in playground](/models/fireworks/deepseek-v4-pro) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Best Open Source Llms",
    "## The best open source LLMs at a glance",
    "| [DeepSeek-V4-Pro](/models/fireworks/deepseek-v4-pro) | April 2026 | 1.6T total | 1,040k tokens | Long-context reasoning and coding from a separate open-source family. Second eval when GLM 5.2 misses on your repo. | [Try in playground](/models/fireworks/deepseek-v4-pro) |"
   ]
  },
  {
   "url": "https://fireworks.ai/blog/best-open-source-llms",
   "title": "Best Open Source LLMs in 2026",
   "published_at": "2026-07-08T00:00:00",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "Same 1,040k context class as Pro at higher throughput and lower cost. Default DeepSeek route for high-volume workloads.",
   "organization_evidence": [
    "Best Open Source Llms",
    "## DeepSeek-V4-Flash",
    "[DeepSeek-V4-Flash](/models/fireworks/deepseek-v4-flash) is the faster V4 route for long-context reasoning and agent work."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Best Open Source Llms",
    "## The best open source LLMs at a glance",
    "| [DeepSeek-V4-Flash](/models/fireworks/deepseek-v4-flash) | April 2026 | 284B total | 1,040k tokens | Same 1,040k context class as Pro at higher throughput and lower cost. Default DeepSeek route for high-volume workloads. | [Try in playground](/models/fireworks/deepseek-v4-flash) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Best Open Source Llms",
    "## The best open source LLMs at a glance",
    "| [DeepSeek-V4-Flash](/models/fireworks/deepseek-v4-flash) | April 2026 | 284B total | 1,040k tokens | Same 1,040k context class as Pro at higher throughput and lower cost. Default DeepSeek route for high-volume workloads. | [Try in playground](/models/fireworks/deepseek-v4-flash) |"
   ]
  },
  {
   "url": "https://fireworks.ai/blog/best-open-source-llms",
   "title": "Best Open Source LLMs in 2026",
   "published_at": "2026-07-08T00:00:00",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-06",
   "access_type": null,
   "license": null,
   "significance": "Native image and video input, second-highest GPQA score in the table. First eval when multimodality sets the constraint.",
   "organization_evidence": [
    "Best Open Source Llms",
    "## MiniMax M3",
    "[MiniMax M3](/models/fireworks/minimax-m3) is a native multimodal MoE model built for long-context, agentic, and multimodal workflows."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Best Open Source Llms",
    "## The best open source LLMs at a glance",
    "| [MiniMax M3](/models/fireworks/minimax-m3) | June 202622, 2025 | 428B total, about 23B activated | 512k tokens | Native image and video input, second-highest GPQA score in the table. First eval when multimodality sets the constraint. | [Try in playground](/models/fireworks/minimax-m3) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Best Open Source Llms",
    "## The best open source LLMs at a glance",
    "| [MiniMax M3](/models/fireworks/minimax-m3) | June 202622, 2025 | 428B total, about 23B activated | 512k tokens | Native image and video input, second-highest GPQA score in the table. First eval when multimodality sets the constraint. | [Try in playground](/models/fireworks/minimax-m3) |"
   ]
  },
  {
   "url": "https://fireworks.ai/blog/best-open-source-llms",
   "title": "Best Open Source LLMs in 2026",
   "published_at": "2026-07-08T00:00:00",
   "organization": "Google",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "Smaller dense multimodal model with Apache-2.0 licensing and an On-Demand deployment path. First eval when adaptation and deployment control matter more than frontier benchmark rank.",
   "organization_evidence": [
    "Best Open Source Llms",
    "## Gemma 4 31B IT",
    "[Gemma 4 31B IT](/models/fireworks/gemma-4-31b-it) is a smaller multimodal instruction-tuned model from Google."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Best Open Source Llms",
    "## The best open source LLMs at a glance",
    "| [Gemma 4 31B IT](/models/fireworks/gemma-4-31b-it) | April 2026 | 32.2B dense | 262k tokens | Smaller dense multimodal model with Apache-2.0 licensing and an On-Demand deployment path. First eval when adaptation and deployment control matter more than frontier benchmark rank. | [View On-Demand model](/models/fireworks/gemma-4-31b-it) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Best Open Source Llms",
    "## The best open source LLMs at a glance",
    "| [Gemma 4 31B IT](/models/fireworks/gemma-4-31b-it) | April 2026 | 32.2B dense | 262k tokens | Smaller dense multimodal model with Apache-2.0 licensing and an On-Demand deployment path. First eval when adaptation and deployment control matter more than frontier benchmark rank. | [View On-Demand model](/models/fireworks/gemma-4-31b-it) |"
   ]
  },
  {
   "url": "https://venturebeat.com/technology/chinas-moonshot-ai-releases-kimi-k3-the-largest-open-source-model-ever-rivaling-top-u-s-systems",
   "title": "China\u2019s Moonshot AI releases Kimi K3, the largest open-source model ever, rivaling top U.S. systems",
   "published_at": "2026-07-16T19:42:09",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-27",
   "access_type": null,
   "license": null,
   "significance": "A 2.8-trillion-parameter frontier-class large language model that benchmarks show performs neck-and-neck with the most powerful proprietary systems from Anthropic and OpenAI.",
   "organization_evidence": null,
   "model_evidence": null,
   "release_date_evidence": [
    "Full model weights are scheduled to be released on July 27, according to details shared by researchers who reviewed the company's technical documentation."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": null
  },
  {
   "url": "https://turingpost.com/p/llms-in-china",
   "title": "Chinese LLMs in 2026: DeepSeek, Qwen3, Kimi K2 and More",
   "published_at": "2026-08-16T21:00:00",
   "organization": "Baichuan",
   "model": null,
   "release_date": "2026",
   "access_type": null,
   "license": null,
   "significance": "medical-focused Baichuan-M3-235B released in 2026",
   "organization_evidence": [
    "## The Original Five Chinese LLM Families\u2014Updated for 2026",
    "Baichuan\u2019s open work has expanded beyond the original general-purpose text models."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## The Original Five Chinese LLM Families\u2014Updated for 2026",
    "Its official repositories now include Baichuan-Omni-1.5 for text, image, audio, and video understanding, plus the medical-focused Baichuan-M3-235B released in 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## The Original Five Chinese LLM Families\u2014Updated for 2026",
    "Its official repositories now include Baichuan-Omni-1.5 for text, image, audio, and video understanding, plus the medical-focused Baichuan-M3-235B released in 2026."
   ]
  },
  {
   "url": "https://secondtalent.com/resources/chinese-open-source-llms-ai-leaders",
   "title": "Top 5 Chinese Open-Source LLMs Dominating 2026",
   "published_at": "2026-07-28T03:00:00",
   "organization": "Moonshot AI",
   "model": "Kimi K3",
   "release_date": "2026-07-27",
   "access_type": "open-weight",
   "license": "Kimi K3 (custom)",
   "significance": "2.8 trillion total parameters, 1M context length, and 88.3 on Terminal-Bench 2.1",
   "organization_evidence": [
    "Top 5 Chinese Open-Source LLMs Dominating 2026",
    "| **1** | Kimi K3 | Moonshot AI | 2.8T (104B active) | 1M | Kimi K3 (custom) | 88.3 Terminal-Bench 2.1 |",
    "Moonshot AI released Kimi K3 on 16 July 2026 and published the full weights eleven days later, on 27 July."
   ],
   "model_evidence": [
    "Top 5 Chinese Open-Source LLMs Dominating 2026",
    "Moonshot AI released Kimi K3 on 16 July 2026 and published the full weights eleven days later, on 27 July."
   ],
   "release_date_evidence": [
    "Top 5 Chinese Open-Source LLMs Dominating 2026",
    "Moonshot AI released Kimi K3 on 16 July 2026 and published the full weights eleven days later, on 27 July."
   ],
   "access_type_evidence": [
    "Top 5 Chinese Open-Source LLMs Dominating 2026",
    "Moonshot AI released Kimi K3 on 16 July 2026 and published the full weights eleven days later, on 27 July."
   ],
   "license_evidence": [
    "Top 5 Chinese Open-Source LLMs Dominating 2026",
    "| **1** | Kimi K3 | Moonshot AI | 2.8T (104B active) | 1M | Kimi K3 (custom) | 88.3 Terminal-Bench 2.1 |",
    "The K3 licence is not MIT \u2014 read it first"
   ],
   "significance_evidence": [
    "Top 5 Chinese Open-Source LLMs Dominating 2026",
    "| **1** | Kimi K3 | Moonshot AI | 2.8T (104B active) | 1M | Kimi K3 (custom) | 88.3 Terminal-Bench 2.1 |",
    "Moonshot AI released Kimi K3 on 16 July 2026 and published the full weights eleven days later, on 27 July."
   ]
  },
  {
   "url": "https://secondtalent.com/resources/chinese-open-source-llms-ai-leaders",
   "title": "Top 5 Chinese Open-Source LLMs Dominating 2026",
   "published_at": "2026-07-28T03:00:00",
   "organization": "DeepSeek",
   "model": "DeepSeek V4",
   "release_date": "2026-04-24",
   "access_type": "open-weight",
   "license": "MIT",
   "significance": "1.6T parameters, 1M context, and 80.6 on SWE-bench Verified",
   "organization_evidence": [
    "Top 5 Chinese Open-Source LLMs Dominating 2026",
    "| **2** | DeepSeek V4-Pro | DeepSeek | 1.6T (49B active) | 1M | MIT | 80.6 SWE-bench Verified |",
    "DeepSeek V4 shipped on 24 April 2026 in two sizes: **V4-Pro** at 1.6 trillion total parameters with 49 billion active, and **V4-Flash** at 284 billion total with 13 billion active."
   ],
   "model_evidence": [
    "Top 5 Chinese Open-Source LLMs Dominating 2026",
    "DeepSeek V4 shipped on 24 April 2026 in two sizes: **V4-Pro** at 1.6 trillion total parameters with 49 billion active, and **V4-Flash** at 284 billion total with 13 billion active."
   ],
   "release_date_evidence": [
    "Top 5 Chinese Open-Source LLMs Dominating 2026",
    "DeepSeek V4 shipped on 24 April 2026 in two sizes: **V4-Pro** at 1.6 trillion total parameters with 49 billion active, and **V4-Flash** at 284 billion total with 13 billion active."
   ],
   "access_type_evidence": [
    "Top 5 Chinese Open-Source LLMs Dominating 2026",
    "DeepSeek V4 shipped on 24 April 2026 in two sizes: **V4-Pro** at 1.6 trillion total parameters with 49 billion active, and **V4-Flash** at 284 billion total with 13 billion active."
   ],
   "license_evidence": [
    "Top 5 Chinese Open-Source LLMs Dominating 2026",
    "| **2** | DeepSeek V4-Pro | DeepSeek | 1.6T (49B active) | 1M | MIT | 80.6 SWE-bench Verified |",
    "Both are MIT-licensed with no revenue thresholds, no attribution clauses and no regional restrictions \u2014 the most permissive terms of any model in this guide."
   ],
   "significance_evidence": [
    "Top 5 Chinese Open-Source LLMs Dominating 2026",
    "| **2** | DeepSeek V4-Pro | DeepSeek | 1.6T (49B active) | 1M | MIT | 80.6 SWE-bench Verified |",
    "DeepSeek V4 shipped on 24 April 2026 in two sizes: **V4-Pro** at 1.6 trillion total parameters with 49 billion active, and **V4-Flash** at 284 billion total with 13 billion active."
   ]
  },
  {
   "url": "https://secondtalent.com/resources/chinese-open-source-llms-ai-leaders",
   "title": "Top 5 Chinese Open-Source LLMs Dominating 2026",
   "published_at": "2026-07-28T03:00:00",
   "organization": "Z.ai (ex-Zhipu)",
   "model": "GLM-5.2",
   "release_date": "2026-06-13",
   "access_type": "open-weight",
   "license": "MIT",
   "significance": "753B parameter mixture-of-experts model with 1M context and 62.1 on SWE-bench Pro",
   "organization_evidence": [
    "Top 5 Chinese Open-Source LLMs Dominating 2026",
    "| **3** | GLM-5.2 | Z.ai (ex-Zhipu) | 753B (~40B active) | 1M | MIT | 62.1 SWE-bench Pro |",
    "Zhipu AI now trades as Z.ai, and GLM-5.2 \u2014 released 13 June 2026 \u2014 is its flagship."
   ],
   "model_evidence": [
    "Top 5 Chinese Open-Source LLMs Dominating 2026",
    "Zhipu AI now trades as Z.ai, and GLM-5.2 \u2014 released 13 June 2026 \u2014 is its flagship."
   ],
   "release_date_evidence": [
    "Top 5 Chinese Open-Source LLMs Dominating 2026",
    "Zhipu AI now trades as Z.ai, and GLM-5.2 \u2014 released 13 June 2026 \u2014 is its flagship."
   ],
   "access_type_evidence": [
    "Top 5 Chinese Open-Source LLMs Dominating 2026",
    "Zhipu AI now trades as Z.ai, and GLM-5.2 \u2014 released 13 June 2026 \u2014 is its flagship."
   ],
   "license_evidence": [
    "Top 5 Chinese Open-Source LLMs Dominating 2026",
    "| **3** | GLM-5.2 | Z.ai (ex-Zhipu) | 753B (~40B active) | 1M | MIT | 62.1 SWE-bench Pro |",
    "Z.ai describes the licensing position bluntly on the model card: *\u201cPure Open: An MIT open-source licence \u2014 no regional limits, technical access without borders.\u201d*"
   ],
   "significance_evidence": [
    "Top 5 Chinese Open-Source LLMs Dominating 2026",
    "| **3** | GLM-5.2 | Z.ai (ex-Zhipu) | 753B (~40B active) | 1M | MIT | 62.1 SWE-bench Pro |",
    "Zhipu AI now trades as Z.ai, and GLM-5.2 \u2014 released 13 June 2026 \u2014 is its flagship."
   ]
  },
  {
   "url": "https://secondtalent.com/resources/chinese-open-source-llms-ai-leaders",
   "title": "Top 5 Chinese Open-Source LLMs Dominating 2026",
   "published_at": "2026-07-28T03:00:00",
   "organization": "MiniMax",
   "model": "MiniMax M3",
   "release_date": "2026-06-01",
   "access_type": "open-weight",
   "license": "minimax-community",
   "significance": "427B parameter MoE model with native multimodality and 80.5 on SWE-bench Verified",
   "organization_evidence": [
    "Top 5 Chinese Open-Source LLMs Dominating 2026",
    "| **4** | MiniMax M3 | MiniMax | 427B (~23B active) | 1M | minimax-community | 80.5 SWE-bench Verified |",
    "Released 1 June 2026, MiniMax M3 is a 427-billion-parameter mixture-of-experts model with roughly 23 billion active per token."
   ],
   "model_evidence": [
    "Top 5 Chinese Open-Source LLMs Dominating 2026",
    "Released 1 June 2026, MiniMax M3 is a 427-billion-parameter mixture-of-experts model with roughly 23 billion active per token."
   ],
   "release_date_evidence": [
    "Top 5 Chinese Open-Source LLMs Dominating 2026",
    "Released 1 June 2026, MiniMax M3 is a 427-billion-parameter mixture-of-experts model with roughly 23 billion active per token."
   ],
   "access_type_evidence": [
    "Top 5 Chinese Open-Source LLMs Dominating 2026",
    "Released 1 June 2026, MiniMax M3 is a 427-billion-parameter mixture-of-experts model with roughly 23 billion active per token."
   ],
   "license_evidence": [
    "Top 5 Chinese Open-Source LLMs Dominating 2026",
    "| **4** | MiniMax M3 | MiniMax | 427B (~23B active) | 1M | minimax-community | 80.5 SWE-bench Verified |",
    "The catch is licensing. M3\u2019s weights ship under a custom **minimax-community** licence, and commercial use of M3 or its derivatives requires a separate agreement with MiniMax."
   ],
   "significance_evidence": [
    "Top 5 Chinese Open-Source LLMs Dominating 2026",
    "| **4** | MiniMax M3 | MiniMax | 427B (~23B active) | 1M | minimax-community | 80.5 SWE-bench Verified |",
    "Released 1 June 2026, MiniMax M3 is a 427-billion-parameter mixture-of-experts model with roughly 23 billion active per token."
   ]
  },
  {
   "url": "https://secondtalent.com/resources/chinese-open-source-llms-ai-leaders",
   "title": "Top 5 Chinese Open-Source LLMs Dominating 2026",
   "published_at": "2026-07-28T03:00:00",
   "organization": "Alibaba",
   "model": "Qwen3.6-35B-A3B",
   "release_date": "2026-04-16",
   "access_type": "open-weight",
   "license": "Apache 2.0",
   "significance": "35B parameter MoE model that runs on a consumer laptop and scores 73.4 on SWE-bench Verified",
   "organization_evidence": [
    "Top 5 Chinese Open-Source LLMs Dominating 2026",
    "| **5** | Qwen3.6-35B-A3B | Alibaba | 35B (3B active) | 256K | Apache 2.0 | 73.4 SWE-bench Verified |",
    "Qwen3.6-35B-A3B is a 35-billion-parameter mixture-of-experts model with roughly 3 billion active per token, released 16 April 2026 as the first open-weight variant of the Qwen3.6 series."
   ],
   "model_evidence": [
    "Top 5 Chinese Open-Source LLMs Dominating 2026",
    "Qwen3.6-35B-A3B is a 35-billion-parameter mixture-of-experts model with roughly 3 billion active per token, released 16 April 2026 as the first open-weight variant of the Qwen3.6 series."
   ],
   "release_date_evidence": [
    "Top 5 Chinese Open-Source LLMs Dominating 2026",
    "Qwen3.6-35B-A3B is a 35-billion-parameter mixture-of-experts model with roughly 3 billion active per token, released 16 April 2026 as the first open-weight variant of the Qwen3.6 series."
   ],
   "access_type_evidence": [
    "Top 5 Chinese Open-Source LLMs Dominating 2026",
    "Qwen3.6-35B-A3B is a 35-billion-parameter mixture-of-experts model with roughly 3 billion active per token, released 16 April 2026 as the first open-weight variant of the Qwen3.6 series."
   ],
   "license_evidence": [
    "Top 5 Chinese Open-Source LLMs Dominating 2026",
    "| **5** | Qwen3.6-35B-A3B | Alibaba | 35B (3B active) | 256K | Apache 2.0 | 73.4 SWE-bench Verified |",
    "Qwen3.6-35B-A3B is a 35-billion-parameter mixture-of-experts model with roughly 3 billion active per token, released 16 April 2026 as the first open-weight variant of the Qwen3.6 series."
   ],
   "significance_evidence": [
    "Top 5 Chinese Open-Source LLMs Dominating 2026",
    "| **5** | Qwen3.6-35B-A3B | Alibaba | 35B (3B active) | 256K | Apache 2.0 | 73.4 SWE-bench Verified |",
    "Qwen3.6-35B-A3B is a 35-billion-parameter mixture-of-experts model with roughly 3 billion active per token, released 16 April 2026 as the first open-weight variant of the Qwen3.6 series."
   ]
  },
  {
   "url": "https://tech-insider.org/meta-muse-glimmer-open-weight-ai-model-2026",
   "title": "Meta Muse Glimmer: 30B Open AI Model for Consumer GPUs",
   "published_at": "2026-08-28T09:47:54",
   "organization": "Meta Superintelligence Labs",
   "model": null,
   "release_date": "2026-08-10",
   "access_type": null,
   "license": null,
   "significance": "A dense 30-billion-parameter open-weight model optimized for always-on local agent workflows that runs on a single consumer GPU.",
   "organization_evidence": [
    "# Meta Muse Glimmer: 30B AI Model Fits in 20GB VRAM [2026]",
    "That is the parameter count of **Muse Glimmer**, an open-weight AI model that Meta Superintelligence Labs says is small enough to run on a single consumer GPU, no data center required.",
    "Muse Glimmer is a dense, 30-billion-parameter model built by Meta Superintelligence Labs and released as open weights under an Apache 2.0 license."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Meta Muse Glimmer: 30B AI Model Fits in 20GB VRAM [2026]",
    "Meta put a number on the table on August 10, 2026, and the number is 30 billion."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Meta Muse Glimmer: 30B AI Model Fits in 20GB VRAM [2026]",
    "That is the parameter count of **Muse Glimmer**, an open-weight AI model that Meta Superintelligence Labs says is small enough to run on a single consumer GPU, no data center required.",
    "Muse Glimmer is a dense, 30-billion-parameter model built by Meta Superintelligence Labs and released as open weights under an Apache 2.0 license."
   ]
  },
  {
   "url": "https://techjacksolutions.com/ai-brief/ibm-granite-4-2-open-source-reasoning-enterprise",
   "title": "IBM Granite 4.2 Brings Chain-of-Thought Reasoning to Open-Source Enterprise AI",
   "published_at": "2026-08-26T00:00:00",
   "organization": "IBM",
   "model": "Granite 4.2",
   "release_date": "2026-08-25",
   "access_type": "open-weight",
   "license": "Apache 2.0",
   "significance": "first family of dense, decoder-only reasoning models with a thinking/non-thinking switch and low-effort thinking mode",
   "organization_evidence": [
    "#### What happened",
    "IBM (International Business Machines) released Granite 4.2, its first family of dense, decoder-only reasoning models, on August 25, 2026."
   ],
   "model_evidence": [
    "#### What happened",
    "IBM (International Business Machines) released Granite 4.2, its first family of dense, decoder-only reasoning models, on August 25, 2026."
   ],
   "release_date_evidence": [
    "#### What happened",
    "IBM (International Business Machines) released Granite 4.2, its first family of dense, decoder-only reasoning models, on August 25, 2026."
   ],
   "access_type_evidence": [
    "#### What happened",
    "Enterprise developers evaluating open-weight alternatives to frontier API costs now have a reasoning-capable option with a permissive license."
   ],
   "license_evidence": [
    "#### What happened",
    "Three sizes, 3B, 8B, and 30B parameters, are available now on [Hugging Face under an Apache 2.0 license](https://huggingface.co/blog/ibm-granite/granite-4-2), per IBM\u2019s technical blog."
   ],
   "significance_evidence": [
    "#### What happened",
    "IBM (International Business Machines) released Granite 4.2, its first family of dense, decoder-only reasoning models, on August 25, 2026."
   ]
  },
  {
   "url": "https://www.reddit.com/r/LocalLLaMA/comments/1rccsjg/i_made_an_interactive_timeline_of_171_llms/",
   "title": "I made an interactive timeline of 171 LLMs (2017\u20132026)",
   "published_at": "2026-02-23T09:18:16",
   "organization": "StepFun",
   "model": null,
   "release_date": "2026-02",
   "access_type": null,
   "license": null,
   "significance": "196B MoE open",
   "organization_evidence": [
    "Built a visual timeline tracking every major Large Language Model \u2014 from the original Transformer paper to GPT-5.3 Codex.",
    "Great list! Just added the notable base models: Falcon H1 (TII, May 2025), Step-3.5-Flash (StepFun, Feb 2026, 196B MoE open), MiniCPM-o 4.5 (OpenBMB, Feb 2026, 9B multimodal), and INTELLECT-3.1 (Prime Intellect, 106B MoE)."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Built a visual timeline tracking every major Large Language Model \u2014 from the original Transformer paper to GPT-5.3 Codex.",
    "Great list! Just added the notable base models: Falcon H1 (TII, May 2025), Step-3.5-Flash (StepFun, Feb 2026, 196B MoE open), MiniCPM-o 4.5 (OpenBMB, Feb 2026, 9B multimodal), and INTELLECT-3.1 (Prime Intellect, 106B MoE)."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Built a visual timeline tracking every major Large Language Model \u2014 from the original Transformer paper to GPT-5.3 Codex.",
    "Great list! Just added the notable base models: Falcon H1 (TII, May 2025), Step-3.5-Flash (StepFun, Feb 2026, 196B MoE open), MiniCPM-o 4.5 (OpenBMB, Feb 2026, 9B multimodal), and INTELLECT-3.1 (Prime Intellect, 106B MoE)."
   ]
  },
  {
   "url": "https://www.reddit.com/r/LocalLLaMA/comments/1rccsjg/i_made_an_interactive_timeline_of_171_llms/",
   "title": "I made an interactive timeline of 171 LLMs (2017\u20132026)",
   "published_at": "2026-02-23T09:18:16",
   "organization": "OpenBMB",
   "model": "MiniCPM-o 4.5",
   "release_date": "2026-02",
   "access_type": "open multimodal",
   "license": null,
   "significance": "9B multimodal",
   "organization_evidence": [
    "Built a visual timeline tracking every major Large Language Model \u2014 from the original Transformer paper to GPT-5.3 Codex.",
    "Great list! Just added the notable base models: Falcon H1 (TII, May 2025), Step-3.5-Flash (StepFun, Feb 2026, 196B MoE open), MiniCPM-o 4.5 (OpenBMB, Feb 2026, 9B multimodal), and INTELLECT-3.1 (Prime Intellect, 106B MoE)."
   ],
   "model_evidence": [
    "Built a visual timeline tracking every major Large Language Model \u2014 from the original Transformer paper to GPT-5.3 Codex.",
    "Great list! Just added the notable base models: Falcon H1 (TII, May 2025), Step-3.5-Flash (StepFun, Feb 2026, 196B MoE open), MiniCPM-o 4.5 (OpenBMB, Feb 2026, 9B multimodal), and INTELLECT-3.1 (Prime Intellect, 106B MoE)."
   ],
   "release_date_evidence": [
    "Built a visual timeline tracking every major Large Language Model \u2014 from the original Transformer paper to GPT-5.3 Codex.",
    "Great list! Just added the notable base models: Falcon H1 (TII, May 2025), Step-3.5-Flash (StepFun, Feb 2026, 196B MoE open), MiniCPM-o 4.5 (OpenBMB, Feb 2026, 9B multimodal), and INTELLECT-3.1 (Prime Intellect, 106B MoE)."
   ],
   "access_type_evidence": [
    "Built a visual timeline tracking every major Large Language Model \u2014 from the original Transformer paper to GPT-5.3 Codex.",
    "Great list! Just added the notable base models: Falcon H1 (TII, May 2025), Step-3.5-Flash (StepFun, Feb 2026, 196B MoE open), MiniCPM-o 4.5 (OpenBMB, Feb 2026, 9B multimodal), and INTELLECT-3.1 (Prime Intellect, 106B MoE)."
   ],
   "license_evidence": null,
   "significance_evidence": [
    "Built a visual timeline tracking every major Large Language Model \u2014 from the original Transformer paper to GPT-5.3 Codex.",
    "Great list! Just added the notable base models: Falcon H1 (TII, May 2025), Step-3.5-Flash (StepFun, Feb 2026, 196B MoE open), MiniCPM-o 4.5 (OpenBMB, Feb 2026, 9B multimodal), and INTELLECT-3.1 (Prime Intellect, 106B MoE)."
   ]
  },
  {
   "url": "https://www.reddit.com/r/LocalLLaMA/comments/1rccsjg/i_made_an_interactive_timeline_of_171_llms/",
   "title": "I made an interactive timeline of 171 LLMs (2017\u20132026)",
   "published_at": "2026-02-23T09:18:16",
   "organization": "zAI",
   "model": "GLM-5",
   "release_date": "2026-02-11",
   "access_type": "open weights",
   "license": null,
   "significance": null,
   "organization_evidence": [
    "GLM-5 by zAI also just released (open weights)"
   ],
   "model_evidence": [
    "GLM-5 by zAI also just released (open weights)"
   ],
   "release_date_evidence": [
    "Already there! GLM-5 (Feb 11, open) is live on the timeline"
   ],
   "access_type_evidence": [
    "GLM-5 by zAI also just released (open weights)"
   ],
   "license_evidence": null,
   "significance_evidence": null
  },
  {
   "url": "https://www.reddit.com/r/LocalLLaMA/comments/1rccsjg/i_made_an_interactive_timeline_of_171_llms/",
   "title": "I made an interactive timeline of 171 LLMs (2017\u20132026)",
   "published_at": "2026-02-23T09:18:16",
   "organization": null,
   "model": "Qwen 3.5",
   "release_date": "2026-02-17",
   "access_type": "open",
   "license": null,
   "significance": "agentic, 1M context",
   "organization_evidence": null,
   "model_evidence": [
    "Just added both! Qwen 3.5 (Feb 17, agentic, 1M context) and Ministral 3 (Dec 2025, 3B/8B/14B edge family). Refresh to see them."
   ],
   "release_date_evidence": [
    "Just added both! Qwen 3.5 (Feb 17, agentic, 1M context) and Ministral 3 (Dec 2025, 3B/8B/14B edge family). Refresh to see them."
   ],
   "access_type_evidence": [
    "Already there! GLM-5 (Feb 11, open) is live on the timeline"
   ],
   "license_evidence": null,
   "significance_evidence": [
    "Just added both! Qwen 3.5 (Feb 17, agentic, 1M context) and Ministral 3 (Dec 2025, 3B/8B/14B edge family). Refresh to see them."
   ]
  },
  {
   "url": "https://codersera.com/blog/kimi-k3-complete-guide-2026",
   "title": "Kimi K3: Specs, Pricing & Release (2026)",
   "published_at": "2026-07-17T18:05:59",
   "organization": "Moonshot AI",
   "model": "Kimi K3",
   "release_date": "2026-07-16",
   "access_type": "open-weight",
   "license": null,
   "significance": "largest open-weight model shipped to date with 2.8 trillion parameters and a 1-million-token context window",
   "organization_evidence": [
    "# Kimi K3: Moonshot AI\u2019s 2.8T Open-Weight Model \u2014 Release, Specs & Pricing (2026)",
    "Kimi K3 is Moonshot AI\u2019s 2.8-trillion-parameter open-weight model, released July 2026.",
    "Moonshot AI shipped **Kimi K3** on July 16, 2026, and it immediately reset expectations for what an open-weight model can do."
   ],
   "model_evidence": [
    "# Kimi K3: Moonshot AI\u2019s 2.8T Open-Weight Model \u2014 Release, Specs & Pricing (2026)",
    "Kimi K3 is Moonshot AI\u2019s 2.8-trillion-parameter open-weight model, released July 2026.",
    "Moonshot AI shipped **Kimi K3** on July 16, 2026, and it immediately reset expectations for what an open-weight model can do."
   ],
   "release_date_evidence": [
    "# Kimi K3: Moonshot AI\u2019s 2.8T Open-Weight Model \u2014 Release, Specs & Pricing (2026)",
    "**Quick answer.** Kimi K3 is Moonshot AI's frontier model, released July 16, 2026.",
    "Moonshot AI shipped **Kimi K3** on July 16, 2026, and it immediately reset expectations for what an open-weight model can do."
   ],
   "access_type_evidence": [
    "# Kimi K3: Moonshot AI\u2019s 2.8T Open-Weight Model \u2014 Release, Specs & Pricing (2026)",
    "Kimi K3 is Moonshot AI\u2019s 2.8-trillion-parameter open-weight model, released July 2026.",
    "It's a 2.8-trillion-parameter open-weight, multimodal reasoning model \u2014 the largest open-weight model shipped to date \u2014 with a 1-million-token context window, an always-on \"thinking mode,\" and pricing of $3 per million input tokens and $15 per million output tokens.",
    "At 2.8 trillion parameters it is the largest open-weight model released so far, and on independent testing it lands fourth among all frontier models \u2014 trailing only Claude Fable 5 and GPT-5.6 Sol, and edging past Claude Opus 4.8."
   ],
   "license_evidence": null,
   "significance_evidence": [
    "# Kimi K3: Moonshot AI\u2019s 2.8T Open-Weight Model \u2014 Release, Specs & Pricing (2026)",
    "It's a 2.8-trillion-parameter open-weight, multimodal reasoning model \u2014 the largest open-weight model shipped to date \u2014 with a 1-million-token context window, an always-on \"thinking mode,\" and pricing of $3 per million input tokens and $15 per million output tokens.",
    "At 2.8 trillion parameters it is the largest open-weight model released so far, and on independent testing it lands fourth among all frontier models \u2014 trailing only Claude Fable 5 and GPT-5.6 Sol, and edging past Claude Opus 4.8."
   ]
  },
  {
   "url": "https://huggingface.co/blog/gemma4",
   "title": "Welcome Gemma 4: Frontier multimodal intelligence on device",
   "published_at": "2026-04-02T00:00:00",
   "organization": "Google DeepMind",
   "model": null,
   "release_date": "2026-04-02",
   "access_type": null,
   "license": null,
   "significance": "A major new multimodal model family supporting image, video, text, and audio inputs with advanced reasoning, long context, and multiple sizes.",
   "organization_evidence": [
    "Published April 2, 2026",
    "The Gemma 4 family of multimodal models by Google DeepMind is out on Hugging Face, with support for your favorite agents, inference engines, and fine-tuning libraries \ud83e\udd17"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Published April 2, 2026",
    "The Gemma 4 family of multimodal models by Google DeepMind is out on Hugging Face, with support for your favorite agents, inference engines, and fine-tuning libraries \ud83e\udd17"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Published April 2, 2026",
    "The Gemma 4 family of multimodal models by Google DeepMind is out on Hugging Face, with support for your favorite agents, inference engines, and fine-tuning libraries \ud83e\udd17",
    "These models are the real deal: truly open with Apache 2 licenses, high quality with pareto frontier arena scores, multimodal including audio, and sizes you can use *everywhere* including on-device."
   ]
  },
  {
   "url": "https://memeburn.com/chinese-ai-models-three-launches-one-month",
   "title": "Three Chinese AI Models Launched in One Month \u2014 and the West Should Be Worried",
   "published_at": "2026-08-05T22:34:05",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-16",
   "access_type": null,
   "license": null,
   "significance": "2.8 trillion parameters, open weights, and the largest open model ever published",
   "organization_evidence": [
    "# Three Chinese AI Models Launched in One Month \u2014 and the West Should Be Worried",
    "- Moonshot AI released Kimi K3 on July 16 \u2014 2.8 trillion parameters, open weights, and the largest open model ever published."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Three Chinese AI Models Launched in One Month \u2014 and the West Should Be Worried",
    "- Moonshot AI released Kimi K3 on July 16 \u2014 2.8 trillion parameters, open weights, and the largest open model ever published."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Three Chinese AI Models Launched in One Month \u2014 and the West Should Be Worried",
    "- Moonshot AI released Kimi K3 on July 16 \u2014 2.8 trillion parameters, open weights, and the largest open model ever published."
   ]
  },
  {
   "url": "https://memeburn.com/chinese-ai-models-three-launches-one-month",
   "title": "Three Chinese AI Models Launched in One Month \u2014 and the West Should Be Worried",
   "published_at": "2026-08-05T22:34:05",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-07-31",
   "access_type": null,
   "license": null,
   "significance": "under the MIT license at a fraction of US model pricing",
   "organization_evidence": [
    "# Three Chinese AI Models Launched in One Month \u2014 and the West Should Be Worried",
    "- DeepSeek officially launched V4-Flash on July 31 under the MIT license at a fraction of US model pricing."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Three Chinese AI Models Launched in One Month \u2014 and the West Should Be Worried",
    "- DeepSeek officially launched V4-Flash on July 31 under the MIT license at a fraction of US model pricing."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Three Chinese AI Models Launched in One Month \u2014 and the West Should Be Worried",
    "- DeepSeek officially launched V4-Flash on July 31 under the MIT license at a fraction of US model pricing."
   ]
  },
  {
   "url": "https://presenc.ai/research/best-open-weight-reasoning-models-2026",
   "title": "Best Open-Weight Reasoning Models 2026",
   "published_at": "2026-05-23T00:00:00",
   "organization": null,
   "model": "Qwen3-Thinking",
   "release_date": "2026-04",
   "access_type": "open-weight",
   "license": "Apache 2.0",
   "significance": "lead the mid-size open-weight reasoning leaderboard with strong AIME, GPQA-Diamond, and LiveCodeBench scores",
   "organization_evidence": null,
   "model_evidence": [
    "Qwen3-Thinking variants (4B, 8B, 14B, 32B, 235B-A22B MoE) released April 2026 lead the mid-size open-weight reasoning leaderboard with strong AIME, GPQA-Diamond, and LiveCodeBench scores.",
    "Reasoning models with test-time compute scaling became the most important AI capability frontier in 2025-2026.",
    "## Key Findings"
   ],
   "release_date_evidence": [
    "Qwen3-Thinking variants (4B, 8B, 14B, 32B, 235B-A22B MoE) released April 2026 lead the mid-size open-weight reasoning leaderboard with strong AIME, GPQA-Diamond, and LiveCodeBench scores.",
    "Reasoning models with test-time compute scaling became the most important AI capability frontier in 2025-2026.",
    "## Key Findings"
   ],
   "access_type_evidence": [
    "Qwen3-Thinking variants (4B, 8B, 14B, 32B, 235B-A22B MoE) released April 2026 lead the mid-size open-weight reasoning leaderboard with strong AIME, GPQA-Diamond, and LiveCodeBench scores.",
    "Reasoning models with test-time compute scaling became the most important AI capability frontier in 2025-2026.",
    "## Key Findings"
   ],
   "license_evidence": [
    "| Qwen3-235B-A22B (Thinking) | ~235B MoE (~22B active) | ~83.0 | ~71.1 | Apache 2.0 |",
    "Open-Weight Reasoning Leaderboard (May 2026)",
    "| Model | Parameters | AIME 2024 | GPQA-Diamond | License |"
   ],
   "significance_evidence": [
    "Qwen3-Thinking variants (4B, 8B, 14B, 32B, 235B-A22B MoE) released April 2026 lead the mid-size open-weight reasoning leaderboard with strong AIME, GPQA-Diamond, and LiveCodeBench scores.",
    "Reasoning models with test-time compute scaling became the most important AI capability frontier in 2025-2026.",
    "## Key Findings"
   ]
  },
  {
   "url": "https://inblix.com/article/openai-ships-open-weight-reasoning-models-first-since-gpt-2-0e8773",
   "title": "OpenAI Ships Open-Weight Reasoning Models, First Since GPT-2",
   "published_at": "2026-07-13T08:46:39",
   "organization": "OpenAI",
   "model": null,
   "release_date": "2026-07-13",
   "access_type": null,
   "license": null,
   "significance": "The gpt-oss family lands with two sizes\u2014a 120-billion-parameter beast and a lean 20-billion-parameter sibling\u2014that nearly match or trade blows with reasoning benchmarks like o4-mini and o3-mini while outperforming GPT-4o and o1 on HealthBench.",
   "organization_evidence": [
    "OpenAI Ships Open-Weight Reasoning Models, First Since GPT-2",
    "OpenAI Blog \u00b7 Jul 13, 2026 \u00b7 2 min read \u00b7 [ Read original article \u2192 ](https://openai.com/index/introducing-gpt-oss?ref=inblix)",
    "OpenAI just did something it hasn\u2019t done since 2019: release an open-weight language model."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "OpenAI Ships Open-Weight Reasoning Models, First Since GPT-2",
    "OpenAI Blog \u00b7 Jul 13, 2026 \u00b7 2 min read \u00b7 [ Read original article \u2192 ](https://openai.com/index/introducing-gpt-oss?ref=inblix)"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "OpenAI Ships Open-Weight Reasoning Models, First Since GPT-2",
    "The gpt-oss family lands with two sizes\u2014a 120-billion-parameter beast and a lean 20-billion-parameter sibling\u2014both under the Apache 2.0 license."
   ]
  },
  {
   "url": "https://nerdleveltech.com/china-open-weight-coding-models-cost-war",
   "title": "China's Open-Weight Coding Wave: 4 Models, 18 Days",
   "published_at": "2026-05-17T00:00:00",
   "organization": "Z.ai",
   "model": "GLM-5.1",
   "release_date": "2026-04-07",
   "access_type": "open-weight",
   "license": "MIT license",
   "significance": "Mixture-of-experts model with roughly 744\u2013754 billion total parameters, 40 billion active parameters per token, and a 200K context window.",
   "organization_evidence": [
    "Between April 7 and April 24, 2026, four Chinese AI labs shipped open-weight coding models in close succession: Z.ai's **GLM-5.1**, MiniMax's **M2.7** (open-sourced April 12 after a March announcement), Moonshot's **Kimi K2.6**, and DeepSeek's **V4** (Pro and Flash variants).",
    "### GLM-5.1 (Z.ai) \u2014 April 7"
   ],
   "model_evidence": [
    "Between April 7 and April 24, 2026, four Chinese AI labs shipped open-weight coding models in close succession: Z.ai's **GLM-5.1**, MiniMax's **M2.7** (open-sourced April 12 after a March announcement), Moonshot's **Kimi K2.6**, and DeepSeek's **V4** (Pro and Flash variants).",
    "### GLM-5.1 (Z.ai) \u2014 April 7"
   ],
   "release_date_evidence": [
    "### GLM-5.1 (Z.ai) \u2014 April 7",
    "# China's Open-Weight Coding Wave: 4 Models, 18 Days"
   ],
   "access_type_evidence": [
    "Between April 7 and April 24, 2026, four Chinese AI labs shipped open-weight coding models in close succession: Z.ai's **GLM-5.1**, MiniMax's **M2.7** (open-sourced April 12 after a March announcement), Moonshot's **Kimi K2.6**, and DeepSeek's **V4** (Pro and Flash variants).",
    "# China's Open-Weight Coding Wave: 4 Models, 18 Days"
   ],
   "license_evidence": [
    "GLM-5.1 is a mixture-of-experts model with roughly **744\u2013754 billion total parameters** (sources disagree on the exact figure, with both widely cited), 40 billion active parameters per token, a 200K context window, and an MIT license2.",
    "### GLM-5.1 (Z.ai) \u2014 April 7"
   ],
   "significance_evidence": [
    "GLM-5.1 is a mixture-of-experts model with roughly **744\u2013754 billion total parameters** (sources disagree on the exact figure, with both widely cited), 40 billion active parameters per token, a 200K context window, and an MIT license2.",
    "### GLM-5.1 (Z.ai) \u2014 April 7"
   ]
  },
  {
   "url": "https://nerdleveltech.com/china-open-weight-coding-models-cost-war",
   "title": "China's Open-Weight Coding Wave: 4 Models, 18 Days",
   "published_at": "2026-05-17T00:00:00",
   "organization": "MiniMax",
   "model": "MiniMax M2.7",
   "release_date": "2026-04-12",
   "access_type": "open weights",
   "license": "Modified-MIT",
   "significance": "A 230-billion-parameter MoE with 10 billion active per token and a 200K context window.",
   "organization_evidence": [
    "Between April 7 and April 24, 2026, four Chinese AI labs shipped open-weight coding models in close succession: Z.ai's **GLM-5.1**, MiniMax's **M2.7** (open-sourced April 12 after a March announcement), Moonshot's **Kimi K2.6**, and DeepSeek's **V4** (Pro and Flash variants).",
    "### MiniMax M2.7 \u2014 April 12 (open)"
   ],
   "model_evidence": [
    "Between April 7 and April 24, 2026, four Chinese AI labs shipped open-weight coding models in close succession: Z.ai's **GLM-5.1**, MiniMax's **M2.7** (open-sourced April 12 after a March announcement), Moonshot's **Kimi K2.6**, and DeepSeek's **V4** (Pro and Flash variants).",
    "### MiniMax M2.7 \u2014 April 12 (open)"
   ],
   "release_date_evidence": [
    "MiniMax M2.7 was originally announced March 18, 2026 as a closed model and then released with open weights on Hugging Face on April 123.",
    "### MiniMax M2.7 \u2014 April 12 (open)"
   ],
   "access_type_evidence": [
    "MiniMax M2.7 was originally announced March 18, 2026 as a closed model and then released with open weights on Hugging Face on April 123.",
    "# China's Open-Weight Coding Wave: 4 Models, 18 Days"
   ],
   "license_evidence": [
    "MiniMax M2.7 is a 230-billion-parameter MoE with 10 billion active per token, a 200K context window (technically 204,800 tokens), and weights released on Hugging Face under a \"Modified-MIT\" license that \u2014 unlike the standard MIT license MiniMax used for M2 and M2.5 \u2014 restricts commercial use without prior written authorization3.",
    "### MiniMax M2.7 \u2014 April 12 (open)"
   ],
   "significance_evidence": [
    "MiniMax M2.7 is a 230-billion-parameter MoE with 10 billion active per token, a 200K context window (technically 204,800 tokens), and weights released on Hugging Face under a \"Modified-MIT\" license that \u2014 unlike the standard MIT license MiniMax used for M2 and M2.5 \u2014 restricts commercial use without prior written authorization3.",
    "### MiniMax M2.7 \u2014 April 12 (open)"
   ]
  },
  {
   "url": "https://nerdleveltech.com/china-open-weight-coding-models-cost-war",
   "title": "China's Open-Weight Coding Wave: 4 Models, 18 Days",
   "published_at": "2026-05-17T00:00:00",
   "organization": "Moonshot AI",
   "model": "Kimi K2.6",
   "release_date": "2026-04-20",
   "access_type": "open-weight",
   "license": "Modified MIT",
   "significance": "A 1-trillion-parameter MoE with 32 billion active per token, a 256K context window, and an Agent Swarm primitive.",
   "organization_evidence": [
    "Between April 7 and April 24, 2026, four Chinese AI labs shipped open-weight coding models in close succession: Z.ai's **GLM-5.1**, MiniMax's **M2.7** (open-sourced April 12 after a March announcement), Moonshot's **Kimi K2.6**, and DeepSeek's **V4** (Pro and Flash variants).",
    "### Kimi K2.6 (Moonshot AI) \u2014 April 20"
   ],
   "model_evidence": [
    "Between April 7 and April 24, 2026, four Chinese AI labs shipped open-weight coding models in close succession: Z.ai's **GLM-5.1**, MiniMax's **M2.7** (open-sourced April 12 after a March announcement), Moonshot's **Kimi K2.6**, and DeepSeek's **V4** (Pro and Flash variants).",
    "### Kimi K2.6 (Moonshot AI) \u2014 April 20"
   ],
   "release_date_evidence": [
    "### Kimi K2.6 (Moonshot AI) \u2014 April 20",
    "# China's Open-Weight Coding Wave: 4 Models, 18 Days"
   ],
   "access_type_evidence": [
    "Between April 7 and April 24, 2026, four Chinese AI labs shipped open-weight coding models in close succession: Z.ai's **GLM-5.1**, MiniMax's **M2.7** (open-sourced April 12 after a March announcement), Moonshot's **Kimi K2.6**, and DeepSeek's **V4** (Pro and Flash variants).",
    "# China's Open-Weight Coding Wave: 4 Models, 18 Days"
   ],
   "license_evidence": [
    "Kimi K2.6 is a 1-trillion-parameter MoE with 32 billion active per token, a 256K context window, and a Modified MIT license4.",
    "### Kimi K2.6 (Moonshot AI) \u2014 April 20"
   ],
   "significance_evidence": [
    "Kimi K2.6 is a 1-trillion-parameter MoE with 32 billion active per token, a 256K context window, and a Modified MIT license4.",
    "### Kimi K2.6 (Moonshot AI) \u2014 April 20"
   ]
  },
  {
   "url": "https://nerdleveltech.com/china-open-weight-coding-models-cost-war",
   "title": "China's Open-Weight Coding Wave: 4 Models, 18 Days",
   "published_at": "2026-05-17T00:00:00",
   "organization": "DeepSeek",
   "model": "V4",
   "release_date": "2026-04-24",
   "access_type": "open-weight",
   "license": "MIT",
   "significance": "Shipped Pro and Flash variants under MIT-licensed weights and a 1M-token context window.",
   "organization_evidence": [
    "Between April 7 and April 24, 2026, four Chinese AI labs shipped open-weight coding models in close succession: Z.ai's **GLM-5.1**, MiniMax's **M2.7** (open-sourced April 12 after a March announcement), Moonshot's **Kimi K2.6**, and DeepSeek's **V4** (Pro and Flash variants).",
    "### DeepSeek V4 \u2014 April 24"
   ],
   "model_evidence": [
    "Between April 7 and April 24, 2026, four Chinese AI labs shipped open-weight coding models in close succession: Z.ai's **GLM-5.1**, MiniMax's **M2.7** (open-sourced April 12 after a March announcement), Moonshot's **Kimi K2.6**, and DeepSeek's **V4** (Pro and Flash variants).",
    "### DeepSeek V4 \u2014 April 24"
   ],
   "release_date_evidence": [
    "### DeepSeek V4 \u2014 April 24",
    "# China's Open-Weight Coding Wave: 4 Models, 18 Days"
   ],
   "access_type_evidence": [
    "Between April 7 and April 24, 2026, four Chinese AI labs shipped open-weight coding models in close succession: Z.ai's **GLM-5.1**, MiniMax's **M2.7** (open-sourced April 12 after a March announcement), Moonshot's **Kimi K2.6**, and DeepSeek's **V4** (Pro and Flash variants).",
    "# China's Open-Weight Coding Wave: 4 Models, 18 Days"
   ],
   "license_evidence": [
    "DeepSeek shipped two variants on the same day under MIT-licensed weights and a 1M-token context window5:",
    "### DeepSeek V4 \u2014 April 24"
   ],
   "significance_evidence": [
    "DeepSeek shipped two variants on the same day under MIT-licensed weights and a 1M-token context window5:",
    "### DeepSeek V4 \u2014 April 24"
   ]
  },
  {
   "url": "https://startupfortune.com/ibm-releases-granite-42-open-reasoning-models-free-for-local-deployment",
   "title": "IBM Releases Granite 4.2 Open Reasoning Models Free for Local Deployment",
   "published_at": "2026-08-26T22:15:47",
   "organization": "IBM",
   "model": null,
   "release_date": "2026-08-25",
   "access_type": null,
   "license": null,
   "significance": "a family of open-weight reasoning models in 3B, 8B, and 30B sizes that enterprises can run on their own hardware under an Apache 2.0 license, no API meter running in the background",
   "organization_evidence": [
    "*IBM shipped Granite 4.2 on August 25, 2026, a family of open-weight reasoning models in 3B, 8B, and 30B sizes that enterprises can run on their own hardware under an Apache 2.0 license, no API meter running in the background.*"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "*IBM shipped Granite 4.2 on August 25, 2026, a family of open-weight reasoning models in 3B, 8B, and 30B sizes that enterprises can run on their own hardware under an Apache 2.0 license, no API meter running in the background.*"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "*IBM shipped Granite 4.2 on August 25, 2026, a family of open-weight reasoning models in 3B, 8B, and 30B sizes that enterprises can run on their own hardware under an Apache 2.0 license, no API meter running in the background.*"
   ]
  },
  {
   "url": "https://techtimes.com/articles/320717/20260716/inkling-ships-muratis-lab-puts-largest-us-open-weight-ai-hugging-face.htm",
   "title": "Inkling Ships: Murati\u2019s Lab Puts Largest US Open-Weight AI on Hugging Face",
   "published_at": "2026-07-16T15:09:41",
   "organization": "Thinking Machines Lab",
   "model": "Inkling",
   "release_date": "2026-07-15",
   "access_type": "open-weight",
   "license": "Apache 2.0",
   "significance": "975-billion-parameter multimodal model pretrained from scratch on 45 trillion tokens",
   "organization_evidence": [
    "Thinking Machines Lab released Inkling on Wednesday \u2014 the first AI model the company built from scratch \u2014 giving any developer or enterprise the right to download, inspect, and fine-tune a 975-billion-parameter multimodal model without routing a single query through a third-party API."
   ],
   "model_evidence": [
    "Thinking Machines Lab released Inkling on Wednesday \u2014 the first AI model the company built from scratch \u2014 giving any developer or enterprise the right to download, inspect, and fine-tune a 975-billion-parameter multimodal model without routing a single query through a third-party API."
   ],
   "release_date_evidence": [
    "Thinking Machines Lab released Inkling on Wednesday \u2014 the first AI model the company built from scratch \u2014 giving any developer or enterprise the right to download, inspect, and fine-tune a 975-billion-parameter multimodal model without routing a single query through a third-party API.",
    "One claim requires specific attribution: the 84.7 percent financial reasoning score cited in Thinking Machines' collaboration with Bridgewater Associates \u2014 the result of fine-tuning an open model on Bridgewater's proprietary financial data \u2014 came from an evaluation conducted by the two companies themselves, not by an independent third party, as [TechCrunch's launch coverage](https://techcrunch.com/2026/07/15/thinking-machines-amps-up-its-bet-against-one-size-fits-all-ai-with-its-first-open-model-inkling/) confirmed."
   ],
   "access_type_evidence": [
    "Thinking Machines Lab released Inkling on Wednesday \u2014 the first AI model the company built from scratch \u2014 giving any developer or enterprise the right to download, inspect, and fine-tune a 975-billion-parameter multimodal model without routing a single query through a third-party API.",
    "The full weights are live on Hugging Face today under Apache 2.0, making Inkling the largest US-built open-weight model publicly available."
   ],
   "license_evidence": [
    "The full weights are live on Hugging Face today under Apache 2.0, making Inkling the largest US-built open-weight model publicly available."
   ],
   "significance_evidence": [
    "Thinking Machines Lab released Inkling on Wednesday \u2014 the first AI model the company built from scratch \u2014 giving any developer or enterprise the right to download, inspect, and fine-tune a 975-billion-parameter multimodal model without routing a single query through a third-party API.",
    "The model was pretrained from scratch on 45 trillion tokens spanning text, images, audio, and video, and trained entirely on NVIDIA GB300 NVL72 systems under the company's March 2026 partnership with NVIDIA for one gigawatt of Vera Rubin computing capacity."
   ]
  },
  {
   "url": "https://venturebeat.com/ai/glm-5-3-flash-will-likely-handle-45-of-your-ai-workloads",
   "title": "GLM-5.3-Flash will likely handle 45% of your AI workloads",
   "published_at": "2026-08-27T00:42:20",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-08-26",
   "access_type": null,
   "license": null,
   "significance": "really good model served entirely on Chinese chips and infrastructure at a low cost",
   "organization_evidence": [
    "On August 26, [Z.ai put its name on it](https://z.ai/blog/glm-5.3-flash). Ox Alpha was GLM-5.3-Flash. They'd been running it"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "On August 26, [Z.ai put its name on it](https://z.ai/blog/glm-5.3-flash). Ox Alpha was GLM-5.3-Flash. They'd been running it"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "It was served entirely on Chinese chips and infrastructure."
   ]
  },
  {
   "url": "https://aitraining2u.com/chinese-open-source-ai-models-2026.html",
   "title": "Chinese Open-Source AI Models in 2026: Kimi, Qwen, DeepSeek, GLM, Doubao & Hunyuan Compared",
   "published_at": "2026-07-23T00:00:00",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-01",
   "access_type": null,
   "license": null,
   "significance": "reported ~2.8-trillion-parameter model with a 1M-token context, billed as the largest open-weight model yet and tuned for long-running autonomous coding agents",
   "organization_evidence": [
    "A cluster of Chinese labs \u2014 **Moonshot AI, Alibaba, DeepSeek, Zhipu (Z.ai), ByteDance and Tencent** \u2014 now ship open-weight models that trade blows with the best closed systems on coding, reasoning and agentic benchmarks, and they release the weights under permissive licences you can actually deploy.",
    "The 2026 landscape at a glance",
    "Moonshot AI's **Kimi** line is the agentic-coding specialist."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "in July 2026 Moonshot released **Kimi K3**, a reported ~2.8-trillion-parameter model with a 1M-token context, billed as the largest open-weight model yet and tuned for long-running autonomous coding agents.",
    "Moonshot \u2014 Kimi K2.6 and K3"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "in July 2026 Moonshot released **Kimi K3**, a reported ~2.8-trillion-parameter model with a 1M-token context, billed as the largest open-weight model yet and tuned for long-running autonomous coding agents.",
    "Moonshot \u2014 Kimi K2.6 and K3"
   ]
  },
  {
   "url": "https://aitraining2u.com/chinese-open-source-ai-models-2026.html",
   "title": "Chinese Open-Source AI Models in 2026: Kimi, Qwen, DeepSeek, GLM, Doubao & Hunyuan Compared",
   "published_at": "2026-07-23T00:00:00",
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026",
   "access_type": null,
   "license": null,
   "significance": "includes a dense 27B and a 35B-A3B MoE (only ~3B parameters active per token) that score at the frontier on agentic coding while fitting on modest hardware",
   "organization_evidence": [
    "A cluster of Chinese labs \u2014 **Moonshot AI, Alibaba, DeepSeek, Zhipu (Z.ai), ByteDance and Tencent** \u2014 now ship open-weight models that trade blows with the best closed systems on coding, reasoning and agentic benchmarks, and they release the weights under permissive licences you can actually deploy.",
    "The 2026 landscape at a glance",
    "Alibaba \u2014 Qwen 3.6"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The 2026 **Qwen 3.6** series includes a dense 27B and a 35B-A3B MoE (only ~3B parameters active per token) that score at the frontier on agentic coding while fitting on modest hardware.",
    "Alibaba \u2014 Qwen 3.6"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The 2026 **Qwen 3.6** series includes a dense 27B and a 35B-A3B MoE (only ~3B parameters active per token) that score at the frontier on agentic coding while fitting on modest hardware.",
    "Alibaba \u2014 Qwen 3.6"
   ]
  },
  {
   "url": "https://aitraining2u.com/chinese-open-source-ai-models-2026.html",
   "title": "Chinese Open-Source AI Models in 2026: Kimi, Qwen, DeepSeek, GLM, Doubao & Hunyuan Compared",
   "published_at": "2026-07-23T00:00:00",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-01",
   "access_type": null,
   "license": null,
   "significance": "merged its general (V) and reasoning (R) lineages into a single model that decides how hard to think per task \u2014 which is why there is no separate \u201cR2\u201d or \u201cR4\u201d product",
   "organization_evidence": [
    "A cluster of Chinese labs \u2014 **Moonshot AI, Alibaba, DeepSeek, Zhipu (Z.ai), ByteDance and Tencent** \u2014 now ship open-weight models that trade blows with the best closed systems on coding, reasoning and agentic benchmarks, and they release the weights under permissive licences you can actually deploy.",
    "The 2026 landscape at a glance",
    "DeepSeek \u2014 V4"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "In April 2026 it shipped **DeepSeek V4**, which *merged* its general (V) and reasoning (R) lineages into a single model that decides how hard to think per task \u2014 which is why there is no separate \u201cR2\u201d or \u201cR4\u201d product.",
    "DeepSeek \u2014 V4"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "In April 2026 it shipped **DeepSeek V4**, which *merged* its general (V) and reasoning (R) lineages into a single model that decides how hard to think per task \u2014 which is why there is no separate \u201cR2\u201d or \u201cR4\u201d product.",
    "DeepSeek \u2014 V4"
   ]
  },
  {
   "url": "https://aitraining2u.com/chinese-open-source-ai-models-2026.html",
   "title": "Chinese Open-Source AI Models in 2026: Kimi, Qwen, DeepSeek, GLM, Doubao & Hunyuan Compared",
   "published_at": "2026-07-23T00:00:00",
   "organization": "Zhipu / Z.ai",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "~753B-parameter MoE (~40B active) under the MIT licence, with a 1M-token context and a clear focus on long-horizon coding agents at roughly a sixth of the cost of closed rivals",
   "organization_evidence": [
    "A cluster of Chinese labs \u2014 **Moonshot AI, Alibaba, DeepSeek, Zhipu (Z.ai), ByteDance and Tencent** \u2014 now ship open-weight models that trade blows with the best closed systems on coding, reasoning and agentic benchmarks, and they release the weights under permissive licences you can actually deploy.",
    "The 2026 landscape at a glance",
    "Zhipu / Z.ai \u2014 GLM-5.2"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "**GLM-5.2** (June 2026) is a ~753B-parameter MoE (~40B active) under the MIT licence, with a 1M-token context and a clear focus on **long-horizon coding agents** at roughly a sixth of the cost of closed rivals.",
    "Zhipu / Z.ai \u2014 GLM-5.2"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "**GLM-5.2** (June 2026) is a ~753B-parameter MoE (~40B active) under the MIT licence, with a 1M-token context and a clear focus on **long-horizon coding agents** at roughly a sixth of the cost of closed rivals.",
    "Zhipu / Z.ai \u2014 GLM-5.2"
   ]
  },
  {
   "url": "https://aitraining2u.com/chinese-open-source-ai-models-2026.html",
   "title": "Chinese Open-Source AI Models in 2026: Kimi, Qwen, DeepSeek, GLM, Doubao & Hunyuan Compared",
   "published_at": "2026-07-23T00:00:00",
   "organization": "Tencent",
   "model": null,
   "release_date": "2026-07-01",
   "access_type": null,
   "license": null,
   "significance": "295B MoE that activates just 21B parameters per token, released under Apache-2.0",
   "organization_evidence": [
    "A cluster of Chinese labs \u2014 **Moonshot AI, Alibaba, DeepSeek, Zhipu (Z.ai), ByteDance and Tencent** \u2014 now ship open-weight models that trade blows with the best closed systems on coding, reasoning and agentic benchmarks, and they release the weights under permissive licences you can actually deploy.",
    "The 2026 landscape at a glance",
    "Tencent \u2014 Hunyuan Hy3"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Tencent's **Hunyuan Hy3** (HunYuan 3.0, July 2026) is a 295B MoE that activates just 21B parameters per token, released under Apache-2.0.",
    "Tencent \u2014 Hunyuan Hy3"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Tencent's **Hunyuan Hy3** (HunYuan 3.0, July 2026) is a 295B MoE that activates just 21B parameters per token, released under Apache-2.0.",
    "Tencent \u2014 Hunyuan Hy3"
   ]
  },
  {
   "url": "https://fazm.ai/blog/open-source-llm-release-2026",
   "title": "Fazm - AI Computer Agent for macOS | Voice-Controlled Desktop Automation",
   "published_at": "2026-04-08T05:15:32",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-01",
   "access_type": null,
   "license": null,
   "significance": "Reasoning, math",
   "organization_evidence": [
    "Open Source LLM Releases in 2026",
    "## Major Open Source LLM Releases in 2026",
    "| Model | Organization | Release Date | Parameters | License | Key Strength |",
    "| DeepSeek R1 | DeepSeek | Jan 2026 | 671B MoE | MIT | Reasoning, math |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Open Source LLM Releases in 2026",
    "## Major Open Source LLM Releases in 2026",
    "| Model | Organization | Release Date | Parameters | License | Key Strength |",
    "| DeepSeek R1 | DeepSeek | Jan 2026 | 671B MoE | MIT | Reasoning, math |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Open Source LLM Releases in 2026",
    "## Major Open Source LLM Releases in 2026",
    "| Model | Organization | Release Date | Parameters | License | Key Strength |",
    "| DeepSeek R1 | DeepSeek | Jan 2026 | 671B MoE | MIT | Reasoning, math |"
   ]
  },
  {
   "url": "https://fazm.ai/blog/open-source-llm-release-2026",
   "title": "Fazm - AI Computer Agent for macOS | Voice-Controlled Desktop Automation",
   "published_at": "2026-04-08T05:15:32",
   "organization": "Meta",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "10M token context",
   "organization_evidence": [
    "Open Source LLM Releases in 2026",
    "## Major Open Source LLM Releases in 2026",
    "| Model | Organization | Release Date | Parameters | License | Key Strength |",
    "| Llama 4 Scout | Meta | Apr 2026 | 109B (17B active) MoE | Llama 4 | 10M token context |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Open Source LLM Releases in 2026",
    "## Major Open Source LLM Releases in 2026",
    "| Model | Organization | Release Date | Parameters | License | Key Strength |",
    "| Llama 4 Scout | Meta | Apr 2026 | 109B (17B active) MoE | Llama 4 | 10M token context |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Open Source LLM Releases in 2026",
    "## Major Open Source LLM Releases in 2026",
    "| Model | Organization | Release Date | Parameters | License | Key Strength |",
    "| Llama 4 Scout | Meta | Apr 2026 | 109B (17B active) MoE | Llama 4 | 10M token context |"
   ]
  },
  {
   "url": "https://fazm.ai/blog/open-source-llm-release-2026",
   "title": "Fazm - AI Computer Agent for macOS | Voice-Controlled Desktop Automation",
   "published_at": "2026-04-08T05:15:32",
   "organization": "Meta",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "Multilingual, code",
   "organization_evidence": [
    "Open Source LLM Releases in 2026",
    "## Major Open Source LLM Releases in 2026",
    "| Model | Organization | Release Date | Parameters | License | Key Strength |",
    "| Llama 4 Maverick | Meta | Apr 2026 | 400B (17B active) MoE | Llama 4 | Multilingual, code |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Open Source LLM Releases in 2026",
    "## Major Open Source LLM Releases in 2026",
    "| Model | Organization | Release Date | Parameters | License | Key Strength |",
    "| Llama 4 Maverick | Meta | Apr 2026 | 400B (17B active) MoE | Llama 4 | Multilingual, code |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Open Source LLM Releases in 2026",
    "## Major Open Source LLM Releases in 2026",
    "| Model | Organization | Release Date | Parameters | License | Key Strength |",
    "| Llama 4 Maverick | Meta | Apr 2026 | 400B (17B active) MoE | Llama 4 | Multilingual, code |"
   ]
  },
  {
   "url": "https://fazm.ai/blog/open-source-llm-release-2026",
   "title": "Fazm - AI Computer Agent for macOS | Voice-Controlled Desktop Automation",
   "published_at": "2026-04-08T05:15:32",
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-02",
   "access_type": null,
   "license": null,
   "significance": "Code generation",
   "organization_evidence": [
    "Open Source LLM Releases in 2026",
    "## Major Open Source LLM Releases in 2026",
    "| Model | Organization | Release Date | Parameters | License | Key Strength |",
    "| Qwen 2.5 Coder | Alibaba | Feb 2026 | 32B | Apache 2.0 | Code generation |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Open Source LLM Releases in 2026",
    "## Major Open Source LLM Releases in 2026",
    "| Model | Organization | Release Date | Parameters | License | Key Strength |",
    "| Qwen 2.5 Coder | Alibaba | Feb 2026 | 32B | Apache 2.0 | Code generation |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Open Source LLM Releases in 2026",
    "## Major Open Source LLM Releases in 2026",
    "| Model | Organization | Release Date | Parameters | License | Key Strength |",
    "| Qwen 2.5 Coder | Alibaba | Feb 2026 | 32B | Apache 2.0 | Code generation |"
   ]
  },
  {
   "url": "https://fazm.ai/blog/open-source-llm-release-2026",
   "title": "Fazm - AI Computer Agent for macOS | Voice-Controlled Desktop Automation",
   "published_at": "2026-04-08T05:15:32",
   "organization": "Google",
   "model": null,
   "release_date": "2026-03",
   "access_type": null,
   "license": null,
   "significance": "Multimodal, on-device",
   "organization_evidence": [
    "Open Source LLM Releases in 2026",
    "## Major Open Source LLM Releases in 2026",
    "| Model | Organization | Release Date | Parameters | License | Key Strength |",
    "| Gemma 3 | Google | Mar 2026 | 1B/4B/12B/27B | Gemma | Multimodal, on-device |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Open Source LLM Releases in 2026",
    "## Major Open Source LLM Releases in 2026",
    "| Model | Organization | Release Date | Parameters | License | Key Strength |",
    "| Gemma 3 | Google | Mar 2026 | 1B/4B/12B/27B | Gemma | Multimodal, on-device |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Open Source LLM Releases in 2026",
    "## Major Open Source LLM Releases in 2026",
    "| Model | Organization | Release Date | Parameters | License | Key Strength |",
    "| Gemma 3 | Google | Mar 2026 | 1B/4B/12B/27B | Gemma | Multimodal, on-device |"
   ]
  },
  {
   "url": "https://fazm.ai/blog/open-source-llm-release-2026",
   "title": "Fazm - AI Computer Agent for macOS | Voice-Controlled Desktop Automation",
   "published_at": "2026-04-08T05:15:32",
   "organization": "Mistral",
   "model": null,
   "release_date": "2026-03",
   "access_type": null,
   "license": null,
   "significance": "Vision, function calling",
   "organization_evidence": [
    "Open Source LLM Releases in 2026",
    "## Major Open Source LLM Releases in 2026",
    "| Model | Organization | Release Date | Parameters | License | Key Strength |",
    "| Mistral Small 3.1 | Mistral | Mar 2026 | 24B | Apache 2.0 | Vision, function calling |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Open Source LLM Releases in 2026",
    "## Major Open Source LLM Releases in 2026",
    "| Model | Organization | Release Date | Parameters | License | Key Strength |",
    "| Mistral Small 3.1 | Mistral | Mar 2026 | 24B | Apache 2.0 | Vision, function calling |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Open Source LLM Releases in 2026",
    "## Major Open Source LLM Releases in 2026",
    "| Model | Organization | Release Date | Parameters | License | Key Strength |",
    "| Mistral Small 3.1 | Mistral | Mar 2026 | 24B | Apache 2.0 | Vision, function calling |"
   ]
  },
  {
   "url": "https://fazm.ai/blog/open-source-llm-release-2026",
   "title": "Fazm - AI Computer Agent for macOS | Voice-Controlled Desktop Automation",
   "published_at": "2026-04-08T05:15:32",
   "organization": "Cohere",
   "model": null,
   "release_date": "2026-03",
   "access_type": null,
   "license": null,
   "significance": "Agentic tool use",
   "organization_evidence": [
    "Open Source LLM Releases in 2026",
    "## Major Open Source LLM Releases in 2026",
    "| Model | Organization | Release Date | Parameters | License | Key Strength |",
    "| Command A | Cohere | Mar 2026 | 111B (active ~36B) MoE | CC-BY-NC | Agentic tool use |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Open Source LLM Releases in 2026",
    "## Major Open Source LLM Releases in 2026",
    "| Model | Organization | Release Date | Parameters | License | Key Strength |",
    "| Command A | Cohere | Mar 2026 | 111B (active ~36B) MoE | CC-BY-NC | Agentic tool use |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Open Source LLM Releases in 2026",
    "## Major Open Source LLM Releases in 2026",
    "| Model | Organization | Release Date | Parameters | License | Key Strength |",
    "| Command A | Cohere | Mar 2026 | 111B (active ~36B) MoE | CC-BY-NC | Agentic tool use |"
   ]
  },
  {
   "url": "https://fazm.ai/blog/open-source-llm-release-2026",
   "title": "Fazm - AI Computer Agent for macOS | Voice-Controlled Desktop Automation",
   "published_at": "2026-04-08T05:15:32",
   "organization": "Allen AI",
   "model": null,
   "release_date": "2026-02",
   "access_type": null,
   "license": null,
   "significance": "Fully open training data",
   "organization_evidence": [
    "Open Source LLM Releases in 2026",
    "## Major Open Source LLM Releases in 2026",
    "| Model | Organization | Release Date | Parameters | License | Key Strength |",
    "| OLMo 2 | Allen AI | Feb 2026 | 7B/13B/32B | Apache 2.0 | Fully open training data |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Open Source LLM Releases in 2026",
    "## Major Open Source LLM Releases in 2026",
    "| Model | Organization | Release Date | Parameters | License | Key Strength |",
    "| OLMo 2 | Allen AI | Feb 2026 | 7B/13B/32B | Apache 2.0 | Fully open training data |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Open Source LLM Releases in 2026",
    "## Major Open Source LLM Releases in 2026",
    "| Model | Organization | Release Date | Parameters | License | Key Strength |",
    "| OLMo 2 | Allen AI | Feb 2026 | 7B/13B/32B | Apache 2.0 | Fully open training data |"
   ]
  },
  {
   "url": "https://taskade.com/blog/open-source-llm-history",
   "title": "Open-Source LLM History: GPT-2 to Kimi K3 (2026)",
   "published_at": "2026-07-25T08:00:00",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-16",
   "access_type": null,
   "license": null,
   "significance": "A 2.8-trillion-parameter open-weight model with native vision and a 1-million-token context window",
   "organization_evidence": [
    "# Open-Source LLM History: GPT-2 to Kimi K3 (2026)",
    "The complete history of open-source LLMs, 2019 to 2026: from GPT-2's cautious release through BLOOM, LLaMA, Mistral, DeepSeek, Qwen, and the 2.8T open-weight Kimi K3.",
    "In July 2026, a Chinese lab shipped **Kimi K3, a 2.8-trillion-parameter open-weight model** at frontier level \u2014 released in the open, its full weights set to follow within days for anyone to download."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## A Visual Timeline of Open-Source LLMs (2019 \u2192 2026)",
    "**Last updated: July 2026** \u2014 refreshed for Kimi K3 (July 16, 2026) and the July 27 open-weight drop, plus DeepSeek V4 and the latest Qwen and GLM releases."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## 2025\u20132026: China's Open-Weight Surge \u2014 Qwen, Kimi, GLM, DeepSeek V4",
    "The capstone landed in July 2026: **Kimi K3**, a **2.8-trillion-parameter** open-weight model \u2014 the **largest open-weight model to date** \u2014 with native vision and a 1-million-token context window, released July 16 with full weights scheduled for July 27."
   ]
  },
  {
   "url": "https://huggingface.co/blog/liquidai/lfm25-dspark",
   "title": "Up to 3.2x Faster Inference with LFM2.5-DSpark",
   "published_at": "2026-08-20T16:52:57",
   "organization": "Liquid AI",
   "model": null,
   "release_date": "2026-08",
   "access_type": null,
   "license": null,
   "significance": "draft model checkpoints for the LFM2.5 family providing speculative decoding paths with up to 3.18x throughput improvement",
   "organization_evidence": [
    "Today, we release **DSpark draft model checkpoints** for three models from our LFM2.5 family: LFM2.5-1.2B-Instruct, LFM2.5-2.6B, and LFM2.5-8B-A1B.",
    "Liquid AI, \"LFM2.5-DSpark: Up to 3.2x Faster Inference from H100 to MacBook\", Liquid AI Blog, Aug 2026."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Liquid AI, \"LFM2.5-DSpark: Up to 3.2x Faster Inference from H100 to MacBook\", Liquid AI Blog, Aug 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Today, we release **DSpark draft model checkpoints** for three models from our LFM2.5 family: LFM2.5-1.2B-Instruct, LFM2.5-2.6B, and LFM2.5-8B-A1B.",
    "These add a speculative decoding path that trades a minimal memory increase for a large decoding speedup without changing output quality:",
    "- **Faster inference**: up to 3.18 throughput improvement on a GPU and up to 2.87x on-device."
   ]
  },
  {
   "url": "https://venturebeat.com/technology/meta-returns-to-open-source-with-muse-glimmer-an-apache-2-0-licensed-30b-parameter-ai-model-optimized-for-agents-available-now",
   "title": "Meta returns to open source with Muse Glimmer, an Apache 2.0 licensed 30B parameter LLM available now",
   "published_at": "2026-08-10T16:22:13",
   "organization": "Meta",
   "model": null,
   "release_date": "2026-08-31",
   "access_type": null,
   "license": null,
   "significance": "a 30-billion-parameter open-weight model designed to run autonomous AI agents directly on consumer hardware",
   "organization_evidence": [
    "Meta today[ released Muse Glimmer, a 30-billion-parameter open-weight model](https://research.meta.ai/blog/introducing-muse-glimmer-open-agentic-model) designed to run autonomous AI agents directly on consumer hardware \u2014 pushing agentic workloads that normally depend on cloud infrastructure onto high-end Macs and PCs."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Meta today[ released Muse Glimmer, a 30-billion-parameter open-weight model](https://research.meta.ai/blog/introducing-muse-glimmer-open-agentic-model) designed to run autonomous AI agents directly on consumer hardware \u2014 pushing agentic workloads that normally depend on cloud infrastructure onto high-end Macs and PCs."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Meta today[ released Muse Glimmer, a 30-billion-parameter open-weight model](https://research.meta.ai/blog/introducing-muse-glimmer-open-agentic-model) designed to run autonomous AI agents directly on consumer hardware \u2014 pushing agentic workloads that normally depend on cloud infrastructure onto high-end Macs and PCs."
   ]
  },
  {
   "url": "https://huggingface.co/blog/LiquidAI/lfm25-dspark",
   "title": "Up to 3.2x Faster Inference with LFM2.5-DSpark",
   "published_at": "2026-08-20T16:52:57",
   "organization": "Liquid AI",
   "model": "LFM2.5 (DSpark draft model checkpoints)",
   "release_date": "2026-08",
   "access_type": "open-sourced",
   "license": null,
   "significance": "up to 3.18 throughput improvement on a GPU and up to 2.87x on-device with speculative decoding",
   "organization_evidence": [
    "Liquid AI, \"LFM2.5-DSpark: Up to 3.2x Faster Inference from H100 to MacBook\", Liquid AI Blog, Aug 2026."
   ],
   "model_evidence": [
    "Liquid AI, \"LFM2.5-DSpark: Up to 3.2x Faster Inference from H100 to MacBook\", Liquid AI Blog, Aug 2026.",
    "Today, we release **DSpark draft model checkpoints** for three models from our LFM2.5 family: LFM2.5-1.2B-Instruct, LFM2.5-2.6B, and LFM2.5-8B-A1B."
   ],
   "release_date_evidence": [
    "Liquid AI, \"LFM2.5-DSpark: Up to 3.2x Faster Inference from H100 to MacBook\", Liquid AI Blog, Aug 2026."
   ],
   "access_type_evidence": [
    "Liquid AI, \"LFM2.5-DSpark: Up to 3.2x Faster Inference from H100 to MacBook\", Liquid AI Blog, Aug 2026.",
    "**Day-one support for llama.cpp and SGLang**: LFM-compatible DSpark integration is open-sourced upstream"
   ],
   "license_evidence": null,
   "significance_evidence": [
    "Liquid AI, \"LFM2.5-DSpark: Up to 3.2x Faster Inference from H100 to MacBook\", Liquid AI Blog, Aug 2026.",
    "These add a speculative decoding path that trades a minimal memory increase for a large decoding speedup without changing output quality:",
    "**Faster inference**: up to 3.18 throughput improvement on a GPU and up to 2.87x on-device."
   ]
  },
  {
   "url": "https://techcrunch.com/2026/07/15/thinking-machines-amps-up-its-bet-against-one-size-fits-all-ai-with-its-first-open-model-inkling",
   "title": "Thinking Machines amps up its bet against one-size-fits-all AI with its first open model, Inkling",
   "published_at": "2026-07-15T18:04:06",
   "organization": "Thinking Machines Lab",
   "model": "Inkling",
   "release_date": "2026-08-26",
   "access_type": "open-weight",
   "license": null,
   "significance": "a mixture-of-experts system with 975 billion total parameters that reasons natively across text, image, audio, and video",
   "organization_evidence": [
    "Thinking Machines Lab, the AI startup founded by former OpenAI CTO Mira Murati, released its first in-house AI model Wednesday morning, called [Inkling](https://thinkingmachines.ai/news/introducing-inkling/). And unlike the flagship models from OpenAI, Anthropic, or Google, it\u2019s open-weight, meaning outside developers and companies can download it and modify it directly.\n\nInkling"
   ],
   "model_evidence": [
    "Thinking Machines Lab, the AI startup founded by former OpenAI CTO Mira Murati, released its first in-house AI model Wednesday morning, called [Inkling](https://thinkingmachines.ai/news/introducing-inkling/). And unlike the flagship models from OpenAI, Anthropic, or Google, it\u2019s open-weight, meaning outside developers and companies can download it and modify it directly.\n\nInkling"
   ],
   "release_date_evidence": [
    "Thinking Machines Lab, the AI startup founded by former OpenAI CTO Mira Murati, released its first in-house AI model Wednesday morning, called [Inkling](https://thinkingmachines.ai/news/introducing-inkling/). And unlike the flagship models from OpenAI, Anthropic, or Google, it\u2019s open-weight, meaning outside developers and companies can download it and modify it directly.\n\nInkling"
   ],
   "access_type_evidence": [
    "Thinking Machines Lab, the AI startup founded by former OpenAI CTO Mira Murati, released its first in-house AI model Wednesday morning, called [Inkling](https://thinkingmachines.ai/news/introducing-inkling/). And unlike the flagship models from OpenAI, Anthropic, or Google, it\u2019s open-weight, meaning outside developers and companies can download it and modify it directly.\n\nInkling",
    "And unlike the flagship models from OpenAI, Anthropic, or Google, it\u2019s open-weight, meaning outside developers and companies can download it and modify it directly."
   ],
   "license_evidence": null,
   "significance_evidence": [
    "Inkling is a mixture-of-experts system with 975 billion total parameters, though it only draws on a fraction of that \u2014 about 41 billion \u2014 for any given task, a common design that keeps very large models faster and cheaper to run.",
    "It was trained on 45 trillion tokens of text, image, audio, and video, and reasons natively across all four, according to the company\u2019s own release materials."
   ]
  },
  {
   "url": "https://www.youtube.com/watch?v=V0RsocRqjIU",
   "title": "Is This the Biggest AI Release of 2026? (China\u2019s New DeepSeek Moment)",
   "published_at": "2026-07-17T22:09:57",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-27",
   "access_type": null,
   "license": null,
   "significance": "A 2.8 trillion parameter open-weight AI model competing with and outperforming top proprietary models in coding and reasoning.",
   "organization_evidence": [
    "On the same day that Moonshot AI unveiled a giant new model called Kimmy K3, Xiinene Ping stood on a stage in Shanghai and told the world that China doesn't plan to follow America's rules for artificial intelligence.",
    "Transcript"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Moonshot says it's the first open model to approach the 3 trillion mark and its full weights are expected to be released on July 27th.",
    "Transcript"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Now, K3 has 2.8 trillion parameters, making it the largest openweight AI model ever announced.",
    "Moonshot says K3 can compete [music] with Anthropics Fable 5 and outperform Claude Opus 4.8, GPT 5.5, and parts of GPT 5.6 in demanding coding work.",
    "Transcript"
   ]
  },
  {
   "url": "https://noqta.tn/en/news/moonshot-kimi-k3-largest-open-model-3t-class-2026",
   "title": "Moonshot AI Releases Kimi K3, the Largest Open Model Ever at 2.8 Trillion Parameters",
   "published_at": "2026-07-17T00:00:00",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-16",
   "access_type": null,
   "license": null,
   "significance": "a 2.8-trillion-parameter mixture-of-experts model the company calls \"the world's first open 3T-class model.\"",
   "organization_evidence": [
    "*Source: [Moonshot AI](https://www.kimi.com/blog/kimi-k3)*",
    "Beijing-based Moonshot AI released Kimi K3 on July 16, 2026, a 2.8-trillion-parameter mixture-of-experts model the company calls \"the world's first open 3T-class model.\""
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "*Source: [Moonshot AI](https://www.kimi.com/blog/kimi-k3)*",
    "Beijing-based Moonshot AI released Kimi K3 on July 16, 2026, a 2.8-trillion-parameter mixture-of-experts model the company calls \"the world's first open 3T-class model.\""
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "*Source: [Moonshot AI](https://www.kimi.com/blog/kimi-k3)*",
    "Beijing-based Moonshot AI released Kimi K3 on July 16, 2026, a 2.8-trillion-parameter mixture-of-experts model the company calls \"the world's first open 3T-class model.\""
   ]
  },
  {
   "url": "https://venturebeat.com/technology/american-ai-startup-poolside-launches-free-high-performing-open-model-laguna-xs-2-for-local-agentic-coding",
   "title": "American AI startup Poolside launches free, high-performing open model Laguna XS.2 for local agentic coding",
   "published_at": "2026-04-28T21:36:52",
   "organization": "Poolside",
   "model": null,
   "release_date": "2026-04-28",
   "access_type": null,
   "license": null,
   "significance": "an Apache 2.0 open licensed 33-billion parameter MoE with 3 billion active engineered for efficiency and community innovation, designed for local agentic coding tasks",
   "organization_evidence": [
    "Today, we got one from the smaller, lesser-known U.S. AI startup, [Poolside](https://poolside.ai/), founded in San Francisco in 2023.",
    "The company [launched its two new Laguna large language models](https://x.com/eisokant/status/2049142230397370537), both of which offer affordable intelligence optimized for agentic workflows (AI that does more than just chat or generate content, but can, in this case, write code, use third-party tools, and take actions autonomously), as well as a new coding agent harness called (fittingly) \"[pool](https://poolside.ai/get-started)\" and a new web-based,[ mobile optimized](https://x.com/almonk/status/2049160796601475508) agentic coding development and interactive preview environment, \"[shimmer](https://shimmer.poolside.ai/),\" which lets you write code with the Laguna models on the go."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Today, we got one from the smaller, lesser-known U.S. AI startup, [Poolside](https://poolside.ai/), founded in San Francisco in 2023.",
    "The new AI models that Poolside released today include:"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "- **Laguna XS.2**: an Apache 2.0 open licensed 33-billion parameter MoE with 3 billion active."
   ]
  },
  {
   "url": "https://techcrunch.com/2026/08/10/metas-new-glimmer-ai-model-offers-a-hint-at-zuckerbergs-personal-intelligence-vision",
   "title": "Meta\u2019s new Glimmer AI model offers a hint at Zuckerberg\u2019s personal intelligence vision",
   "published_at": "2026-08-10T16:20:13",
   "organization": "Meta",
   "model": null,
   "release_date": "2026-04-06",
   "access_type": null,
   "license": null,
   "significance": "30-billion parameter open-weight model designed to power AI agents locally on consumer hardware",
   "organization_evidence": [
    "Meta on Monday released [Muse Glimmer](https://research.meta.ai/blog/introducing-muse-glimmer-open-agentic-model), an open-weight model designed to power AI agents locally on consumer hardware, providing the clearest picture yet of what CEO Mark Zuckerberg\u2019s vision of \u201cpersonal superintelligence\u201d could look like in practice."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Meta on Monday released [Muse Glimmer](https://research.meta.ai/blog/introducing-muse-glimmer-open-agentic-model), an open-weight model designed to power AI agents locally on consumer hardware, providing the clearest picture yet of what CEO Mark Zuckerberg\u2019s vision of \u201cpersonal superintelligence\u201d could look like in practice."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Meta on Monday released [Muse Glimmer](https://research.meta.ai/blog/introducing-muse-glimmer-open-agentic-model), an open-weight model designed to power AI agents locally on consumer hardware, providing the clearest picture yet of what CEO Mark Zuckerberg\u2019s vision of \u201cpersonal superintelligence\u201d could look like in practice.",
    "The 30-billion parameter model is essentially an open version of Meta\u2019s most powerful closed model, Muse Spark, which [the company debuted in April](https://techcrunch.com/2026/04/08/meta-debuts-the-muse-spark-model-in-a-ground-up-overhaul-of-its-ai/)."
   ]
  },
  {
   "url": "https://huggingface.co/blog/mishig/local-moores-law",
   "title": "Two Years of Local AI on a Laptop: When Open Models Outpaced Moore's Law",
   "published_at": "2026-05-11T10:13:08",
   "organization": "Qwen",
   "model": null,
   "release_date": "2026-04-22",
   "access_type": null,
   "license": null,
   "significance": "A dense 27 billion parameter model that scores 46 on the Artificial Analysis Intelligence Index.",
   "organization_evidence": [
    "Two Years of Local AI on a Laptop: When Open Models Outpaced Moore's Law",
    "## The five snapshots",
    "| Oct 2024 | [Qwen 2.5 72B Instruct](https://huggingface.co/Qwen/Qwen2.5-72B-Instruct) | Q4 | 16 |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Two Years of Local AI on a Laptop: When Open Models Outpaced Moore's Law",
    "## Discontinuity 2: small dense reasoning catches up, huge MoE fits via Q2 (April 2026)",
    "[**Qwen3.6 27B (Reasoning)**](https://huggingface.co/Qwen/Qwen3.6-27B) arrived on April 22, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Two Years of Local AI on a Laptop: When Open Models Outpaced Moore's Law",
    "## Discontinuity 2: small dense reasoning catches up, huge MoE fits via Q2 (April 2026)",
    "[**Qwen3.6 27B (Reasoning)**](https://huggingface.co/Qwen/Qwen3.6-27B) arrived on April 22, 2026.",
    "A dense 27 billion parameter model that scores 46 on the Artificial Analysis Intelligence Index."
   ]
  },
  {
   "url": "https://huggingface.co/blog/mishig/local-moores-law",
   "title": "Two Years of Local AI on a Laptop: When Open Models Outpaced Moore's Law",
   "published_at": "2026-05-11T10:13:08",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "284 billion total parameters, 13 billion active.",
   "organization_evidence": [
    "Two Years of Local AI on a Laptop: When Open Models Outpaced Moore's Law",
    "## Discontinuity 2: small dense reasoning catches up, huge MoE fits via Q2 (April 2026)",
    "**DeepSeek V4 Flash** arrived on April 24, 2026."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Two Years of Local AI on a Laptop: When Open Models Outpaced Moore's Law",
    "## Discontinuity 2: small dense reasoning catches up, huge MoE fits via Q2 (April 2026)",
    "**DeepSeek V4 Flash** arrived on April 24, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Two Years of Local AI on a Laptop: When Open Models Outpaced Moore's Law",
    "## Discontinuity 2: small dense reasoning catches up, huge MoE fits via Q2 (April 2026)",
    "284 billion total parameters, 13 billion active."
   ]
  },
  {
   "url": "https://techcrunch.com/2026/08/26/surprise-z-ai-is-the-ai-lab-behind-the-mysterious-ox-alpha-model?post=",
   "title": "Surprise: Z.ai is the AI lab behind the mysterious Ox Alpha model",
   "published_at": "2026-08-26T14:19:29",
   "organization": "Z.ai",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "a reasoning model designed for coding, sustained agentic work, and production workloads",
   "organization_evidence": [
    "# Surprise: Z.ai is the AI lab behind the mysterious Ox Alpha model",
    "As many had expected, Ox Alpha was spawned by GLM-maker Z.ai, according to [Bloomberg](https://www.bloomberg.com/news/articles/2026-08-26/china-s-z-ai-made-ox-alpha-stealth-model-that-rivals-deepseek?taid=6a8eae9d6aa06700017f2021&utm_campaign=trueanthem&utm_content=business&utm_medium=social&utm_source=twitter)."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Surprise: Z.ai is the AI lab behind the mysterious Ox Alpha model",
    "The company describes Ox Alpha as \u201ca reasoning model designed for coding, sustained agentic work, and production workloads."
   ]
  },
  {
   "url": "https://venturebeat.com/infrastructure/poolside-drops-laguna-s-2-1-an-open-weight-coding-model-that-beats-rivals-10x-its-size",
   "title": "Poolside drops Laguna S 2.1, an open-weight coding model that beats rivals 10x its size",
   "published_at": "2026-07-21T21:49:43",
   "organization": "Poolside",
   "model": "Laguna S 2.1",
   "release_date": null,
   "access_type": "open weights",
   "license": "OpenMDW-1.1",
   "significance": "118-billion-parameter Mixture-of-Experts system that activates only 8 billion parameters per token, supports a context window of up to 1 million tokens, and matches or beats open models several times its size on agentic coding tasks",
   "organization_evidence": [
    "Poolside",
    "[Poolside](http://poolside.ai/), the San Francisco AI lab that has spent most of its three-year existence quietly selling coding models to governments and defense agencies, released its most capable model to date on Tuesday \u2014 and made an unusually aggressive bet that radical transparency, not raw scale, is how a smaller lab competes at the frontier."
   ],
   "model_evidence": [
    "Poolside",
    "The model, [Laguna S 2.1](https://poolside.ai/blog/introducing-laguna-s-2-1), is a 118-billion-parameter[ Mixture-of-Experts (MoE) system](https://huggingface.co/blog/moe) that activates only 8 billion parameters per token, supports a context window of up to 1 million tokens, and \u2014 according to benchmarks published by the company \u2014 matches or beats open models several times its size on agentic coding tasks."
   ],
   "release_date_evidence": null,
   "access_type_evidence": [
    "Poolside"
   ],
   "license_evidence": [
    "Poolside",
    "The weights are [available immediately](https://huggingface.co/poolside/Laguna-S-2.1) on Hugging Face under the permissive OpenMDW-1.1 license."
   ],
   "significance_evidence": [
    "Poolside",
    "The model, [Laguna S 2.1](https://poolside.ai/blog/introducing-laguna-s-2-1), is a 118-billion-parameter[ Mixture-of-Experts (MoE) system](https://huggingface.co/blog/moe) that activates only 8 billion parameters per token, supports a context window of up to 1 million tokens, and \u2014 according to benchmarks published by the company \u2014 matches or beats open models several times its size on agentic coding tasks."
   ]
  },
  {
   "url": "https://arxiv.org/pdf/2604.07190v2",
   "title": "The ATOM Report: Measuring the Open Language Model Ecosystem",
   "published_at": null,
   "organization": "Qwen Team",
   "model": null,
   "release_date": "2026-02",
   "access_type": null,
   "license": null,
   "significance": "strong initial adoption and rollout across multiple model scales",
   "organization_evidence": [
    "The ATOM Report",
    "April 2026",
    "5.1\nEcosystem Leaders (Qwen, Llama, DeepSeek, Mistral, and OpenAI)",
    "The early adoption numbers of Qwen3.5 (Qwen Team, 2026) since its release in February 2026 are\nan indication that the dominance of Qwen relative to its peers will continue (see Section 6 and\nspecifically Fig. 15 for early data on Qwen 3.5\u2019s adoption)."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The ATOM Report",
    "April 2026",
    "5.1\nEcosystem Leaders (Qwen, Llama, DeepSeek, Mistral, and OpenAI)",
    "The early adoption numbers of Qwen3.5 (Qwen Team, 2026) since its release in February 2026 are\nan indication that the dominance of Qwen relative to its peers will continue (see Section 6 and\nspecifically Fig. 15 for early data on Qwen 3.5\u2019s adoption)."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The ATOM Report",
    "April 2026",
    "6.2\nUsing RAM to Measure New Models",
    "For example, using the 2026-Q2 RAM baseline, the RAM framework makes the February 2026\nQwen3.5 rollout easier to compare across scales."
   ]
  },
  {
   "url": "https://futureagi.com/blog/open-source-llm-releases",
   "title": "Open-Source AI News: July 2026 LLM Release Tracker",
   "published_at": "2026-07-20T00:00:00",
   "organization": "Moonshot",
   "model": null,
   "release_date": "2026-07-27",
   "access_type": null,
   "license": null,
   "significance": "A 2.8-trillion-parameter mixture-of-experts model with 104B active parameters, native image input, and a one-million-token context window.",
   "organization_evidence": [
    "Open-Source AI News: July 2026 LLM Release Tracker",
    "Nine notable LLM releases had downloadable weights in July by the time of this update.",
    "The list now includes [Kimi K3](https://huggingface.co/moonshotai/Kimi-K3): Moonshot [announced it on 16 July](https://www.kimi.com/code/docs/en/kimi-code/whats-new.html), and the full checkpoint files became available on 27 July."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Open-Source AI News: July 2026 LLM Release Tracker",
    "Nine notable LLM releases had downloadable weights in July by the time of this update.",
    "The list now includes [Kimi K3](https://huggingface.co/moonshotai/Kimi-K3): Moonshot [announced it on 16 July](https://www.kimi.com/code/docs/en/kimi-code/whats-new.html), and the full checkpoint files became available on 27 July."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Open-Source AI News: July 2026 LLM Release Tracker",
    "Kimi K3 Made July\u2019s Largest Checkpoint Downloadable",
    "Kimi K3 is a 2.8-trillion-parameter mixture-of-experts model with 104B active parameters, native image input, and a one-million-token context window."
   ]
  },
  {
   "url": "https://futureagi.com/blog/open-source-llm-releases",
   "title": "Open-Source AI News: July 2026 LLM Release Tracker",
   "published_at": "2026-07-20T00:00:00",
   "organization": "Tencent",
   "model": null,
   "release_date": "2026-07-06",
   "access_type": null,
   "license": null,
   "significance": "A large general model with 295B total parameters, 21B active parameters, and a 256K context window.",
   "organization_evidence": [
    "Open-Source AI News: July 2026 LLM Release Tracker",
    "### Hy3 Put a Large General Model Under Apache-2.0",
    "Tencent released Hy3 on 6 July with 295B total parameters, 21B active parameters, and a 256K context window."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Open-Source AI News: July 2026 LLM Release Tracker",
    "### Hy3 Put a Large General Model Under Apache-2.0",
    "Tencent released Hy3 on 6 July with 295B total parameters, 21B active parameters, and a 256K context window."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Open-Source AI News: July 2026 LLM Release Tracker",
    "### Hy3 Put a Large General Model Under Apache-2.0",
    "Tencent released Hy3 on 6 July with 295B total parameters, 21B active parameters, and a 256K context window."
   ]
  },
  {
   "url": "https://gcn.com/meta-opens-muse-glimmer-billion-parameter/21144",
   "title": "Meta opens Muse Glimmer's 30-billion-parameter weights and pledges to release Muse Spark 1.2 openly",
   "published_at": "2026-08-27T12:00:00",
   "organization": "Meta",
   "model": null,
   "release_date": "2026-08-10",
   "access_type": null,
   "license": null,
   "significance": "a 30-billion-parameter open-weight model designed for agentic tasks that can run on a Mac or PC with a single graphics card",
   "organization_evidence": [
    "# Meta opens Muse Glimmer\u2019s 30-billion-parameter weights and pledges to release Muse Spark 1.2 openly",
    "Meta opened the weights of its 30-billion-parameter Muse Glimmer artificial intelligence model on August 10, while CEO Mark Zuckerberg said the company would open the weights for its latest AI model, Muse Spark 1.2, meaning it can be downloaded and used by the public."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Meta opens Muse Glimmer\u2019s 30-billion-parameter weights and pledges to release Muse Spark 1.2 openly",
    "Meta opened the weights of its 30-billion-parameter Muse Glimmer artificial intelligence model on August 10, while CEO Mark Zuckerberg said the company would open the weights for its latest AI model, Muse Spark 1.2, meaning it can be downloaded and used by the public."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## What Muse Glimmer is and what it can do",
    "Muse Glimmer is an open-weight model designed for agentic tasks that can run on a Mac or PC with a single graphics card, aiming to tap demand for AI systems that run directly on people\u2019s devices."
   ]
  },
  {
   "url": "https://huggingface.co/blog/Svngoku/ai-models-week-july-09-2026",
   "title": "AI Model Releases \u2014 Week of July 9, 2026",
   "published_at": "2026-07-09T20:30:27",
   "organization": "Tencent",
   "model": null,
   "release_date": "2026-07-06",
   "access_type": null,
   "license": null,
   "significance": "a 295B-parameter Mixture-of-Experts model with 21B parameters activated per token and a separate 3.8B-parameter multi-token-prediction layer",
   "organization_evidence": [
    "At the same time, Tencent, Meituan, Mistral, Poolside, and Sber expanded the open-weight frontier with models aimed at agents, coding, formal verification, and long-context work.",
    "## Executive Summary",
    "## 5. \ud83d\udfe0 Tencent \u2014 Hy3"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "**Released:** July 6, 2026",
    "## 5. \ud83d\udfe0 Tencent \u2014 Hy3"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Hy3 is the production release that follows Tencent's April preview.",
    "It is a 295B-parameter Mixture-of-Experts model with 21B parameters activated per token and a separate 3.8B-parameter multi-token-prediction layer.",
    "## 5. \ud83d\udfe0 Tencent \u2014 Hy3"
   ]
  },
  {
   "url": "https://huggingface.co/blog/Svngoku/ai-models-week-july-09-2026",
   "title": "AI Model Releases \u2014 Week of July 9, 2026",
   "published_at": "2026-07-09T20:30:27",
   "organization": "Meituan",
   "model": null,
   "release_date": "2026-07-02",
   "access_type": null,
   "license": null,
   "significance": "1.6T total parameters with roughly 48B activated per token",
   "organization_evidence": [
    "At the same time, Tencent, Meituan, Mistral, Poolside, and Sber expanded the open-weight frontier with models aimed at agents, coding, formal verification, and long-context work.",
    "## Executive Summary",
    "## 6. \ud83d\udfe1 Meituan \u2014 LongCat\u20112.0"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "**Public weights:** Released during the July 2\u20139 window",
    "## 6. \ud83d\udfe1 Meituan \u2014 LongCat\u20112.0"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "LongCat\u20112.0 is one of the largest openly downloadable MoE releases of the week: **1.6T total parameters** with roughly **48B activated per token**.",
    "## 6. \ud83d\udfe1 Meituan \u2014 LongCat\u20112.0"
   ]
  },
  {
   "url": "https://huggingface.co/blog/Svngoku/ai-models-week-july-09-2026",
   "title": "AI Model Releases \u2014 Week of July 9, 2026",
   "published_at": "2026-07-09T20:30:27",
   "organization": "Mistral",
   "model": null,
   "release_date": "2026-07-02",
   "access_type": null,
   "license": null,
   "significance": "a 119B-total / 6B-active MoE model specialized for theorem proving and code verification in Lean 4",
   "organization_evidence": [
    "At the same time, Tencent, Meituan, Mistral, Poolside, and Sber expanded the open-weight frontier with models aimed at agents, coding, formal verification, and long-context work.",
    "## Executive Summary",
    "## 10. \ud83d\udd34 Mistral \u2014 Leanstral 1.5"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "**Released:** July 2, 2026",
    "## 10. \ud83d\udd34 Mistral \u2014 Leanstral 1.5"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Leanstral 1.5 is a **119B-total / 6B-active** MoE model specialized for theorem proving and code verification in Lean 4.",
    "## 10. \ud83d\udd34 Mistral \u2014 Leanstral 1.5"
   ]
  },
  {
   "url": "https://huggingface.co/blog/Svngoku/ai-models-week-july-09-2026",
   "title": "AI Model Releases \u2014 Week of July 9, 2026",
   "published_at": "2026-07-09T20:30:27",
   "organization": "Poolside",
   "model": null,
   "release_date": "2026-07-02",
   "access_type": null,
   "license": null,
   "significance": "a 33B-total / 3B-active MoE model designed to run agentic coding workloads on local hardware",
   "organization_evidence": [
    "At the same time, Tencent, Meituan, Mistral, Poolside, and Sber expanded the open-weight frontier with models aimed at agents, coding, formal verification, and long-context work.",
    "## Executive Summary",
    "## 11. \ud83d\udfe6 Poolside \u2014 Laguna XS 2.1"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "**Released:** July 2, 2026",
    "## 11. \ud83d\udfe6 Poolside \u2014 Laguna XS 2.1"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Laguna XS 2.1 is a **33B-total / 3B-active** MoE model designed to run agentic coding workloads on local hardware.",
    "## 11. \ud83d\udfe6 Poolside \u2014 Laguna XS 2.1"
   ]
  },
  {
   "url": "https://huggingface.co/blog/Svngoku/ai-models-week-july-09-2026",
   "title": "AI Model Releases \u2014 Week of July 9, 2026",
   "published_at": "2026-07-09T20:30:27",
   "organization": "Sber",
   "model": null,
   "release_date": "2026-07-06",
   "access_type": null,
   "license": null,
   "significance": "Sber's new flagship model for coding, mathematics, long-document analysis, and autonomous workflows",
   "organization_evidence": [
    "At the same time, Tencent, Meituan, Mistral, Poolside, and Sber expanded the open-weight frontier with models aimed at agents, coding, formal verification, and long-context work.",
    "## Executive Summary",
    "## 12. \ud83d\udfe2 Sber \u2014 GigaChat 3.5 Ultra"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "**Released:** July 6, 2026",
    "## 12. \ud83d\udfe2 Sber \u2014 GigaChat 3.5 Ultra"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "GigaChat 3.5 Ultra is Sber's new flagship model for coding, mathematics, long-document analysis, and autonomous workflows.",
    "## 12. \ud83d\udfe2 Sber \u2014 GigaChat 3.5 Ultra"
   ]
  },
  {
   "url": "https://techcrunch.com/2026/01/05/nvidia-launches-alpamayo-open-ai-models-that-allow-autonomous-vehicles-to-think-like-a-human",
   "title": "Nvidia launches Alpamayo, open AI models that allow autonomous vehicles to \u2018think like a human\u2019",
   "published_at": "2026-01-05T21:52:22",
   "organization": "Nvidia",
   "model": null,
   "release_date": "2026-01",
   "access_type": null,
   "license": null,
   "significance": "10 billion-parameter chain-of-thought, reason-based vision language action (VLA) model that allows an AV to think more like a human so it can solve complex edge cases",
   "organization_evidence": [
    "At [CES 2026](https://techcrunch.com/storyline/ces-2026-follow-live-as-nvidia-lego-amd-amazon-and-more-make-their-big-reveals/), Nvidia launched Alpamayo, a new family of open source AI models, simulation tools, and datasets for training physical robots and vehicles that are designed to help autonomous vehicles reason through complex driving situations."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "At [CES 2026](https://techcrunch.com/storyline/ces-2026-follow-live-as-nvidia-lego-amd-amazon-and-more-make-their-big-reveals/), Nvidia launched Alpamayo, a new family of open source AI models, simulation tools, and datasets for training physical robots and vehicles that are designed to help autonomous vehicles reason through complex driving situations."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "At the core of Nvidia\u2019s new family is Alpamayo 1, a 10 billion-parameter chain-of-thought, reason-based vision language action (VLA) model that allows an AV to think more like a human so it can solve complex edge cases \u2014 like how to navigate a traffic light outage at a busy intersection \u2014 without previous experience."
   ]
  },
  {
   "url": "https://venturebeat.com/technology/qwen3-8-27b-runs-frontier-class-coding-agents-and-reasoning-locally-no-cloud-api-required",
   "title": "Qwen3.8-27B runs frontier-class coding agents and reasoning on a high-end laptop, no cloud API required",
   "published_at": "2026-08-18T00:06:53",
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-08-28",
   "access_type": null,
   "license": null,
   "significance": "A dense multimodal model with native image and video understanding, a 262,144-token context window, configurable reasoning and support for coding and agentic workflows.",
   "organization_evidence": [
    "The biggest AI model release of the past few days, at least among the developers and AI power users on social media, wasn't a frontier cloud model from OpenAI, Anthropic or Google.",
    "It was a 27-billion-parameter model from Alibaba: [Qwen3.8-27B landed on Hugging Face](https://huggingface.co/Qwen/Qwen3.8-27B) on Friday under an enterprise-friendly, open source Apache 2.0 license, giving developers downloadable weights for a dense multimodal model."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "It was a 27-billion-parameter model from Alibaba: [Qwen3.8-27B landed on Hugging Face](https://huggingface.co/Qwen/Qwen3.8-27B) on Friday under an enterprise-friendly, open source Apache 2.0 license, giving developers downloadable weights for a dense multimodal model."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The biggest AI model release of the past few days, at least among the developers and AI power users on social media, wasn't a frontier cloud model from OpenAI, Anthropic or Google.",
    "But Qwen3.8-27B isn't a garden variety small local model: it includes native image and video understanding, a 262,144-token context window, configurable reasoning and support for coding and agentic workflows \u2014 a \u201ccompact, deployment-friendly\u201d version of the capabilities developed for its Qwen3.8 generation."
   ]
  },
  {
   "url": "https://ghacks.net/2026/07/16/thinking-machines-lab-releases-inkling-a-975-billion-parameter-open-weights-ai-model-under-apache-2-0",
   "title": "Thinking Machines Lab Releases Inkling, a 975 Billion Parameter Open Weights AI Model Under Apache 2.0",
   "published_at": "2026-07-16T08:50:06",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-16",
   "access_type": null,
   "license": null,
   "significance": "975 billion parameters open weights AI model",
   "organization_evidence": [
    "# Thinking Machines Lab Releases Inkling, a 975 Billion Parameter Open Weights AI Model Under Apache 2.0",
    "Thinking Machines Lab released Inkling on Wednesday, an AI model with 975 billion parameters."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Updated \u2022 Jul 16, 2026",
    "Thinking Machines Lab released Inkling on Wednesday, an AI model with 975 billion parameters."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Thinking Machines Lab Releases Inkling, a 975 Billion Parameter Open Weights AI Model Under Apache 2.0",
    "Thinking Machines Lab released Inkling on Wednesday, an AI model with 975 billion parameters."
   ]
  },
  {
   "url": "https://superintelligencenews.com/ai-fields/large-language-models/open-weight-model-inkling-thinking-machines",
   "title": "Thinking Machines Debuts Inkling, Its First Open Model Built for Enterprise Customization",
   "published_at": "2026-07-15T18:37:03",
   "organization": "Thinking Machines",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "Mixture-of-experts model with 975 billion total parameters and multimodal reasoning across text, audio, images and video.",
   "organization_evidence": [
    "Thinking Machines has released Inkling, its first open-weight multimodal model for enterprise customization, while WIRED adds that it was trained from scratch on text, audio and video, briefly shed natural-language reasoning during training, and is being positioned against leading Chinese open-weight rivals.",
    "Thinking Machines Lab, the artificial intelligence startup founded by former OpenAI chief technology officer Mira Murati, has unveiled Inkling, its first AI model and its first major public test of a strategy built around customizable, open-weight systems."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Updated July 15, 2026 6:25 pm",
    "| July 15, 2026 | Inkling is released publicly | First major model launch and a test of the company\u2019s strategy |",
    "Timeline: how Thinking Machines got here"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "- Inkling uses a mixture-of-experts design with 975 billion total parameters and about 41 billion active per task.",
    "- Thinking Machines says the model was trained on 45 trillion tokens and can reason across text, images, audio and video.",
    "In short"
   ]
  },
  {
   "url": "https://pondero.ai/news/2026-07-07-meituan-longcat-2-chinese-frontier-coding-model",
   "title": "Meituan open-sources LongCat-2.0, a 1.6T-parameter coding model trained entirely on 50,000 domestic Chinese chips",
   "published_at": "2026-07-07T00:00:00",
   "organization": "Meituan",
   "model": null,
   "release_date": "2026-06-30",
   "access_type": null,
   "license": null,
   "significance": "A 1.6-trillion-parameter MoE coding model with native 1M context trained entirely on 50,000 domestic Chinese chips.",
   "organization_evidence": [
    "## Meituan open-sources LongCat-2.0, a 1.6T-parameter coding model trained entirely on 50,000 domestic Chinese chips",
    "China's food-delivery giant Meituan officially released LongCat-2.0 on June 30, 2026."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "China's food-delivery giant Meituan officially released LongCat-2.0 on June 30, 2026.",
    "## Meituan open-sources LongCat-2.0, a 1.6T-parameter coding model trained entirely on 50,000 domestic Chinese chips"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## Meituan open-sources LongCat-2.0, a 1.6T-parameter coding model trained entirely on 50,000 domestic Chinese chips",
    "The model is a Mixture-of-Experts architecture with 1.6 trillion total parameters."
   ]
  },
  {
   "url": "https://llm-releases.com/changelog",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-07-31",
   "access_type": null,
   "license": null,
   "significance": "DeepSeek-V4-Flash-0731 production build featuring 284B total / 13B active MoE and 1M context",
   "organization_evidence": [
    "A single feed of releases, updates, deprecations, and retractions across every lab we track.",
    "DeepSeek ships DeepSeek-V4-Flash-0731, the production build of V4-Flash: the April preview retrained on an improved post-training pipeline for coding, agents, reasoning, and tool use, with the architecture untouched (284B total / 13B active MoE, 1M context)."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Last updated Jul 31, 2026",
    "DeepSeek ships DeepSeek-V4-Flash-0731, the production build of V4-Flash: the April preview retrained on an improved post-training pipeline for coding, agents, reasoning, and tool use, with the architecture untouched (284B total / 13B active MoE, 1M context)."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "A single feed of releases, updates, deprecations, and retractions across every lab we track.",
    "DeepSeek ships DeepSeek-V4-Flash-0731, the production build of V4-Flash: the April preview retrained on an improved post-training pipeline for coding, agents, reasoning, and tool use, with the architecture untouched (284B total / 13B active MoE, 1M context)."
   ]
  },
  {
   "url": "https://llm-releases.com/changelog",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-26",
   "access_type": null,
   "license": null,
   "significance": "2.8T-parameter MoE, the largest open-weight model publicly available",
   "organization_evidence": [
    "A single feed of releases, updates, deprecations, and retractions across every lab we track.",
    "Moonshot AI publishes the full Kimi K3 weights to Hugging Face under a Modified MIT license on July 26 \u2014 a day ahead of its announced July 27 target \u2014 making the 2.8T-parameter MoE freely downloadable, modifiable, and self-hostable, and cementing K3 as the largest open-weight model publicly available."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Last updated Jul 31, 2026",
    "Moonshot AI publishes the full Kimi K3 weights to Hugging Face under a Modified MIT license on July 26 \u2014 a day ahead of its announced July 27 target \u2014 making the 2.8T-parameter MoE freely downloadable, modifiable, and self-hostable, and cementing K3 as the largest open-weight model publicly available."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "A single feed of releases, updates, deprecations, and retractions across every lab we track.",
    "Moonshot AI publishes the full Kimi K3 weights to Hugging Face under a Modified MIT license on July 26 \u2014 a day ahead of its announced July 27 target \u2014 making the 2.8T-parameter MoE freely downloadable, modifiable, and self-hostable, and cementing K3 as the largest open-weight model publicly available."
   ]
  },
  {
   "url": "https://llm-releases.com/changelog",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Ant Group (inclusionAI)",
   "model": null,
   "release_date": "2026-07-24",
   "access_type": null,
   "license": null,
   "significance": "Hybrid-reasoning Mixture-of-Experts model with 124B total parameters and ~5.1B active per token",
   "organization_evidence": [
    "A single feed of releases, updates, deprecations, and retractions across every lab we track.",
    "Ant Group's inclusionAI lab releases Ling-3.0-flash, a hybrid-reasoning Mixture-of-Experts model with 124B total parameters and ~5.1B active per token (1/64 expert activation), built for production-scale agents."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Last updated Jul 31, 2026",
    "Announced as open-weight under Apache 2.0, but weights and a model card were not yet posted to Hugging Face as of July 24; usable only via hosted API, free on OpenRouter and Vercel AI Gateway through August 3 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "A single feed of releases, updates, deprecations, and retractions across every lab we track.",
    "Ant Group's inclusionAI lab releases Ling-3.0-flash, a hybrid-reasoning Mixture-of-Experts model with 124B total parameters and ~5.1B active per token (1/64 expert activation), built for production-scale agents."
   ]
  },
  {
   "url": "https://llm-releases.com/changelog",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Poolside",
   "model": null,
   "release_date": "2026-07-31",
   "access_type": null,
   "license": null,
   "significance": "118B-total / 8B-active open-weight MoE coding model with a 1M-token context",
   "organization_evidence": [
    "A single feed of releases, updates, deprecations, and retractions across every lab we track.",
    "Poolside releases Laguna S 2.1, a 118B-total / 8B-active open-weight MoE coding model with a 1M-token context, pitched as 'the West's most capable open-weight model' for its weight class."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Last updated Jul 31, 2026"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "A single feed of releases, updates, deprecations, and retractions across every lab we track.",
    "Poolside releases Laguna S 2.1, a 118B-total / 8B-active open-weight MoE coding model with a 1M-token context, pitched as 'the West's most capable open-weight model' for its weight class."
   ]
  },
  {
   "url": "https://llm-releases.com/changelog",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-31",
   "access_type": null,
   "license": null,
   "significance": "975B-total / 41B-active multimodal MoE pretrained on ~45T tokens",
   "organization_evidence": [
    "A single feed of releases, updates, deprecations, and retractions across every lab we track.",
    "Mira Murati's lab ships its first model \u2014 a 975B-total / 41B-active multimodal MoE (text/image/audio in, text out) pretrained on ~45T tokens, released under Apache-2.0 with weights on Hugging Face (1M context) and hosted on the Tinker API (256K)."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Last updated Jul 31, 2026"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "A single feed of releases, updates, deprecations, and retractions across every lab we track.",
    "Mira Murati's lab ships its first model \u2014 a 975B-total / 41B-active multimodal MoE (text/image/audio in, text out) pretrained on ~45T tokens, released under Apache-2.0 with weights on Hugging Face (1M context) and hosted on the Tinker API (256K)."
   ]
  },
  {
   "url": "https://llm-releases.com/changelog",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "NVIDIA",
   "model": null,
   "release_date": "2026-07-31",
   "access_type": null,
   "license": null,
   "significance": "Compressed variant of Nemotron-3-Super trimming parent to 75.3B total / 9.3B active with hybrid Mamba-Transformer LatentMoE design",
   "organization_evidence": [
    "A single feed of releases, updates, deprecations, and retractions across every lab we track.",
    "NVIDIA releases Nemotron-Labs-3-Puzzle-75B-A9B"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Last updated Jul 31, 2026"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "A single feed of releases, updates, deprecations, and retractions across every lab we track.",
    "A compressed variant of Nemotron-3-Super produced with \"Iterative Puzzle\", trimming the parent to 75.3B total / 9.3B active while keeping the hybrid Mamba-Transformer LatentMoE design and 1M context."
   ]
  },
  {
   "url": "https://llm-releases.com/changelog",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Tencent Hunyuan",
   "model": null,
   "release_date": "2026-07-31",
   "access_type": null,
   "license": null,
   "significance": "295B-total / 21B-active MoE with a 256K context and three selectable fast/slow inference modes",
   "organization_evidence": [
    "A single feed of releases, updates, deprecations, and retractions across every lab we track.",
    "Tencent officially launches Hunyuan 3.0 (Hy3), the GA of its rebuilt third-generation model, and open-sources it under Apache-2.0."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Last updated Jul 31, 2026",
    "Weights on Hugging Face (tencent/Hy3) and ModelScope, with a free OpenRouter route (tencent/hy3:free) through July 21, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "A single feed of releases, updates, deprecations, and retractions across every lab we track.",
    "A 295B-total / 21B-active MoE (plus a 3.8B multi-token-prediction layer) with a 256K context and three selectable fast/slow inference modes; Tencent reports it rivals GLM-5.2 and DeepSeek-V4 and matches or surpasses GPT-5.5 on several science benchmarks, with 78.0 on SWE-bench Verified."
   ]
  },
  {
   "url": "https://llm-releases.com/changelog",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Poolside",
   "model": null,
   "release_date": "2026-07-31",
   "access_type": null,
   "license": null,
   "significance": "Upgraded 33B-A3B open-weight MoE coding model served at 256K context",
   "organization_evidence": [
    "A single feed of releases, updates, deprecations, and retractions across every lab we track.",
    "Poolside releases Laguna XS 2.1, an upgraded 33B-A3B open-weight MoE coding model served at 256K context."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Last updated Jul 31, 2026"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "A single feed of releases, updates, deprecations, and retractions across every lab we track.",
    "Poolside releases Laguna XS 2.1, an upgraded 33B-A3B open-weight MoE coding model served at 256K context."
   ]
  },
  {
   "url": "https://llm-releases.com/changelog",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Meituan (LongCat)",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "1.6T-parameter MoE (~48B active) with a 1M-token context for agentic coding",
   "organization_evidence": [
    "A single feed of releases, updates, deprecations, and retractions across every lab we track.",
    "Meituan releases LongCat-2.0, a 1.6T-parameter MoE (~48B active) with a 1M-token context for agentic coding, open-sourced under the MIT license on Hugging Face and GitHub."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Last updated Jul 31, 2026",
    "## June 2026",
    "1. Released",
    "Meituan open-sources LongCat-2.0, trained entirely on Chinese chips"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "A single feed of releases, updates, deprecations, and retractions across every lab we track.",
    "Meituan releases LongCat-2.0, a 1.6T-parameter MoE (~48B active) with a 1M-token context for agentic coding, open-sourced under the MIT license on Hugging Face and GitHub."
   ]
  },
  {
   "url": "https://llm-releases.com/changelog",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Liquid AI",
   "model": null,
   "release_date": "2026-05-02",
   "access_type": null,
   "license": null,
   "significance": "On-device MoE (8.3B total / ~1.5B active) with a 131K context that runs in under ~6GB of memory",
   "organization_evidence": [
    "A single feed of releases, updates, deprecations, and retractions across every lab we track.",
    "Liquid AI ships its on-device MoE (8.3B total / ~1.5B active) with a 131K context that runs in under ~6GB of memory, under the LFM Open License, scaling pretraining to 38T tokens over the October 2025 LFM2-8B-A1B."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Last updated Jul 31, 2026",
    "## May 2026",
    "2. Released",
    "Liquid AI releases LFM2.5-8B-A1B"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "A single feed of releases, updates, deprecations, and retractions across every lab we track.",
    "Liquid AI ships its on-device MoE (8.3B total / ~1.5B active) with a 131K context that runs in under ~6GB of memory, under the LFM Open License, scaling pretraining to 38T tokens over the October 2025 LFM2-8B-A1B."
   ]
  },
  {
   "url": "https://llm-releases.com/changelog",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Tencent Hunyuan",
   "model": null,
   "release_date": "2026-04-03",
   "access_type": null,
   "license": null,
   "significance": "295B-total / 21B-active MoE with a 256K context",
   "organization_evidence": [
    "A single feed of releases, updates, deprecations, and retractions across every lab we track.",
    "Tencent releases and open-sources the Hy3 preview, its rebuilt third-generation Hunyuan: a 295B-total / 21B-active MoE with a 256K context, positioned as a leading open reasoning-and-agent model for its size."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Last updated Jul 31, 2026",
    "## April 2026",
    "3. Released",
    "Tencent open-sources Hunyuan Hy3-preview"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "A single feed of releases, updates, deprecations, and retractions across every lab we track.",
    "Tencent releases and open-sources the Hy3 preview, its rebuilt third-generation Hunyuan: a 295B-total / 21B-active MoE with a 256K context, positioned as a leading open reasoning-and-agent model for its size."
   ]
  },
  {
   "url": "https://llm-releases.com/changelog",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Xiaomi (MiMo)",
   "model": null,
   "release_date": "2026-04-04",
   "access_type": null,
   "license": null,
   "significance": "~310B/15B-active sparse-MoE model trained on ~48T tokens with a 1M-token context",
   "organization_evidence": [
    "A single feed of releases, updates, deprecations, and retractions across every lab we track.",
    "Alongside the Pro flagship, Xiaomi releases MiMo-V2.5, a ~310B/15B-active sparse-MoE model trained on ~48T tokens with a 1M-token context, under the MIT license."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Last updated Jul 31, 2026",
    "## April 2026",
    "4. Released",
    "Xiaomi open-sources MiMo-V2.5"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "A single feed of releases, updates, deprecations, and retractions across every lab we track.",
    "Alongside the Pro flagship, Xiaomi releases MiMo-V2.5, a ~310B/15B-active sparse-MoE model trained on ~48T tokens with a 1M-token context, under the MIT license."
   ]
  },
  {
   "url": "https://llm-releases.com/changelog",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Xiaomi (MiMo)",
   "model": null,
   "release_date": "2026-04-05",
   "access_type": null,
   "license": null,
   "significance": "1.02T-parameter MoE (~42B active) with hybrid attention and a 1M-token context, tuned for frontier-class agentic coding",
   "organization_evidence": [
    "A single feed of releases, updates, deprecations, and retractions across every lab we track.",
    "Xiaomi releases its open-weight flagship MiMo-V2.5-Pro: a 1.02T-parameter MoE (~42B active) with hybrid attention and a 1M-token context, tuned for frontier-class agentic coding."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Last updated Jul 31, 2026",
    "## April 2026",
    "5. Released",
    "Xiaomi open-sources MiMo-V2.5-Pro"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "A single feed of releases, updates, deprecations, and retractions across every lab we track.",
    "Xiaomi releases its open-weight flagship MiMo-V2.5-Pro: a 1.02T-parameter MoE (~42B active) with hybrid attention and a 1M-token context, tuned for frontier-class agentic coding."
   ]
  },
  {
   "url": "https://arxiv.org/pdf/2604.07190",
   "title": "The ATOM Report: Measuring the Open Language Model Ecosystem",
   "published_at": "2026-04-09T00:00:00",
   "organization": "Alibaba",
   "model": "Qwen3.5",
   "release_date": "2026-02",
   "access_type": "open weight",
   "license": null,
   "significance": "strong adoption and broad model family covering multiple parameter scales",
   "organization_evidence": [
    "The growth of the Chinese model ecosystem can largely be attributed to Alibaba\u2019s Qwen, which is responsible for almost a billion cumulative downloads by March 2026.",
    "## 1\u2003\u200aIntroduction"
   ],
   "model_evidence": [
    "The early adoption numbers of Qwen3.5\u00a0(29) since its release in February 2026 are an indication that the dominance of Qwen relative to its peers will continue (see Section\u00a06 and specifically Fig.\u00a015 for early data on Qwen 3.5\u2019s adoption).",
    "## 5\u2003\u200aModel Adoption by Organization",
    "### 5.1\u2003\u200aEcosystem Leaders (Qwen, Llama, DeepSeek, Mistral, and OpenAI)"
   ],
   "release_date_evidence": [
    "The early adoption numbers of Qwen3.5\u00a0(29) since its release in February 2026 are an indication that the dominance of Qwen relative to its peers will continue (see Section\u00a06 and specifically Fig.\u00a015 for early data on Qwen 3.5\u2019s adoption).",
    "## 5\u2003\u200aModel Adoption by Organization",
    "### 5.1\u2003\u200aEcosystem Leaders (Qwen, Llama, DeepSeek, Mistral, and OpenAI)"
   ],
   "access_type_evidence": [
    "Open weight AI models are becoming foundational infrastructure across research, startups, and governments negotiating their future in understanding, building, and deploying increasingly powerful AI systems.",
    "## 1\u2003\u200aIntroduction"
   ],
   "license_evidence": null,
   "significance_evidence": [
    "The early adoption numbers of Qwen3.5\u00a0(29) since its release in February 2026 are an indication that the dominance of Qwen relative to its peers will continue (see Section\u00a06 and specifically Fig.\u00a015 for early data on Qwen 3.5\u2019s adoption).",
    "## 5\u2003\u200aModel Adoption by Organization",
    "### 5.1\u2003\u200aEcosystem Leaders (Qwen, Llama, DeepSeek, Mistral, and OpenAI)"
   ]
  },
  {
   "url": "https://medium.com/@ffguci8/nvidia-nemotron-3-the-sota-open-weight-ai-model-family-of-2026-4612ae7aefb4",
   "title": "NVIDIA Nemotron 3: The SOTA Open-Weight AI Model Family of 2026",
   "published_at": "2026-06-06T19:25:30",
   "organization": "NVIDIA",
   "model": null,
   "release_date": "2026-03-11",
   "access_type": null,
   "license": null,
   "significance": "120.6 billion total parameters with a 1-million-token context window and high inference throughput",
   "organization_evidence": [
    "# NVIDIA Nemotron 3: The SOTA Open-Weight AI Model Family of 2026",
    "On June 4, 2026, NVIDIA quietly dropped Nemotron 3 Ultra to Hugging Face, two days after Jensen Huang announced it from the Computex stage in Taipei."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# NVIDIA Nemotron 3: The SOTA Open-Weight AI Model Family of 2026",
    "*The Nemotron 3 family: Nano (December 2025), Super (March 2026), and Ultra (June 2026). Source: NVIDIA Research / Artificial Analysis*",
    "**Nemotron 3 Super** released at GTC on March 11, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# NVIDIA Nemotron 3: The SOTA Open-Weight AI Model Family of 2026",
    "**Nemotron 3 Super** released at GTC on March 11, 2026."
   ]
  },
  {
   "url": "https://medium.com/@ffguci8/nvidia-nemotron-3-the-sota-open-weight-ai-model-family-of-2026-4612ae7aefb4",
   "title": "NVIDIA Nemotron 3: The SOTA Open-Weight AI Model Family of 2026",
   "published_at": "2026-06-06T19:25:30",
   "organization": "NVIDIA",
   "model": null,
   "release_date": "2026-06-04",
   "access_type": null,
   "license": null,
   "significance": "550 billion parameters and the highest Intelligence Index score of any US-developed open-weight model",
   "organization_evidence": [
    "# NVIDIA Nemotron 3: The SOTA Open-Weight AI Model Family of 2026",
    "On June 4, 2026, NVIDIA quietly dropped Nemotron 3 Ultra to Hugging Face, two days after Jensen Huang announced it from the Computex stage in Taipei."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# NVIDIA Nemotron 3: The SOTA Open-Weight AI Model Family of 2026",
    "*The Nemotron 3 family: Nano (December 2025), Super (March 2026), and Ultra (June 2026). Source: NVIDIA Research / Artificial Analysis*",
    "On June 4, 2026, NVIDIA quietly dropped Nemotron 3 Ultra to Hugging Face, two days after Jensen Huang announced it from the Computex stage in Taipei."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# NVIDIA Nemotron 3: The SOTA Open-Weight AI Model Family of 2026",
    "On June 4, 2026, NVIDIA quietly dropped Nemotron 3 Ultra to Hugging Face, two days after Jensen Huang announced it from the Computex stage in Taipei."
   ]
  },
  {
   "url": "https://medium.com/@ffguci8/nvidia-nemotron-3-the-sota-open-weight-ai-model-family-of-2026-4612ae7aefb4",
   "title": "NVIDIA Nemotron 3: The SOTA Open-Weight AI Model Family of 2026",
   "published_at": "2026-06-06T19:25:30",
   "organization": "NVIDIA",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "Multimodal version of Nano that handles vision, audio, and language in a single model",
   "organization_evidence": [
    "# NVIDIA Nemotron 3: The SOTA Open-Weight AI Model Family of 2026",
    "On June 4, 2026, NVIDIA quietly dropped Nemotron 3 Ultra to Hugging Face, two days after Jensen Huang announced it from the Computex stage in Taipei."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# NVIDIA Nemotron 3: The SOTA Open-Weight AI Model Family of 2026",
    "There is also a fourth variant: **Nemotron 3 Nano Omni**, released in April 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# NVIDIA Nemotron 3: The SOTA Open-Weight AI Model Family of 2026",
    "There is also a fourth variant: **Nemotron 3 Nano Omni**, released in April 2026."
   ]
  },
  {
   "url": "https://medium.com/@blockgeni7/mira-muratis-thinking-machines-releases-inkling-an-open-weight-ai-model-built-for-customization-ba5c4d286c76",
   "title": "Mira Murati\u2019s Thinking Machines Releases Inkling, an Open-Weight AI Model Built for Customization",
   "published_at": "2026-07-28T09:45:16",
   "organization": "Thinking Machines",
   "model": null,
   "release_date": "2026-07-16",
   "access_type": null,
   "license": null,
   "significance": "trained from scratch with the full weights available and paired with a companion tool called Tinker for direct weight adjustment",
   "organization_evidence": [
    "# Mira Murati\u2019s Thinking Machines Releases Inkling, an Open-Weight AI Model Built for Customization",
    "Mira Murati, the former chief technology officer of OpenAI, has delivered the first public product from her AI startup Thinking Machines Lab: a fully open-weight model named Inkling, designed to be customized by businesses and developers rather than used as a fixed, proprietary service."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Mira Murati\u2019s Thinking Machines Releases Inkling, an Open-Weight AI Model Built for Customization",
    "The release of Inkling on July 16, 2026, is the first concrete signal of what that work produced."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Mira Murati\u2019s Thinking Machines Releases Inkling, an Open-Weight AI Model Built for Customization",
    "According to the company, Inkling is \u201ca model we trained from scratch with the full weights available, so that people can make it their own.\u201d",
    "To make that customization practical, Thinking Machines has paired Inkling with a companion tool called Tinker, which allows users to adjust the model\u2019s weights directly \u2014 effectively reshaping how the AI reasons and responds without requiring full retraining from scratch."
   ]
  },
  {
   "url": "https://cryptobriefing.com/nvidia-shortens-ai-release-cycles",
   "title": "Nvidia shortens AI model release cycles from 6-8 months to every 4-6 weeks",
   "published_at": "2026-08-28T19:42:43",
   "organization": "Nvidia",
   "model": null,
   "release_date": "2026-08-11",
   "access_type": null,
   "license": null,
   "significance": "Latest product of the pipeline built with synthetic data generation, multi-teacher distillation, and reinforcement learning environments, optimized for agentic use cases.",
   "organization_evidence": [
    "The chipmaker's open-weight Nemotron models are now shipping at a pace that makes traditional software updates look glacial",
    "Nvidia is now pushing out new versions of its open-weight AI models roughly every month and a half, a dramatic acceleration from the six-to-eight-month cadence it maintained previously."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The latest product of this pipeline is Nemotron 3.5 Lightning, which shipped on August 11, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The latest product of this pipeline is Nemotron 3.5 Lightning, which shipped on August 11, 2026.",
    "Nvidia has been explicit that its Nemotron models are optimized for agentic use cases, meaning AI systems that can take actions, use tools, and operate semi-autonomously rather than just generating text."
   ]
  },
  {
   "url": "https://techriseups.com/articles/open-weight-ai-models-caught-up-in-2026-the-catch-is-you-host-them-yourself",
   "title": "Open-Weight AI Models Caught Up in 2026 \u2014 The Catch Is You Host Them Yourself",
   "published_at": "2026-06-27T21:43:24",
   "organization": "Zhipu AI",
   "model": null,
   "release_date": "2026-06",
   "access_type": null,
   "license": null,
   "significance": "Released June 2026 as part of open-weight frontier models closing the gap with closed APIs.",
   "organization_evidence": [
    "Open-Weight AI Models Caught Up in 2026 \u2014 The Catch Is You Host Them Yourself",
    "## Recent open-weight releases worth knowing",
    "| Model | Maker | Licence | Reported result |",
    "| **GLM-5.2** | Zhipu AI | Open-source | Released June 2026 (per llm-stats) |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Open-Weight AI Models Caught Up in 2026 \u2014 The Catch Is You Host Them Yourself",
    "## Recent open-weight releases worth knowing",
    "| Model | Maker | Licence | Reported result |",
    "| **GLM-5.2** | Zhipu AI | Open-source | Released June 2026 (per llm-stats) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Open-Weight AI Models Caught Up in 2026 \u2014 The Catch Is You Host Them Yourself",
    "According to model trackers, [Zhipu AI released GLM-5.2 in June 2026](https://llm-stats.com/llm-updates) as an open-source model \u2014 part of a busy year for open weights.",
    "## Recent open-weight releases worth knowing",
    "| Model | Maker | Licence | Reported result |",
    "| **GLM-5.2** | Zhipu AI | Open-source | Released June 2026 (per llm-stats) |"
   ]
  },
  {
   "url": "https://techriseups.com/articles/open-weight-ai-models-caught-up-in-2026-the-catch-is-you-host-them-yourself",
   "title": "Open-Weight AI Models Caught Up in 2026 \u2014 The Catch Is You Host Them Yourself",
   "published_at": "2026-06-27T21:43:24",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-16",
   "access_type": null,
   "license": null,
   "significance": "2.8-trillion-parameter open MoE model with a 1M-token context",
   "organization_evidence": [
    "AI & ML",
    "### Kimi K3 Is the Largest Open AI Model Ever \u2014 With a Catch",
    "Kimi K3 is Moonshot AI\u2019s 2.8-trillion-parameter open MoE model with a 1M-token context, launched July 16, 2026."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "AI & ML",
    "### Kimi K3 Is the Largest Open AI Model Ever \u2014 With a Catch",
    "Kimi K3 is Moonshot AI\u2019s 2.8-trillion-parameter open MoE model with a 1M-token context, launched July 16, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "AI & ML",
    "### Kimi K3 Is the Largest Open AI Model Ever \u2014 With a Catch",
    "Kimi K3 is Moonshot AI\u2019s 2.8-trillion-parameter open MoE model with a 1M-token context, launched July 16, 2026."
   ]
  },
  {
   "url": "https://www.technologyreview.com/2026/02/12/1132811/whats-next-for-chinese-open-source-ai/",
   "title": "What\u2019s next for Chinese open-source AI",
   "published_at": "2026-02-12T10:00:00",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "came close to top proprietary systems such as Anthropic\u2019s Claude Opus on some early benchmarks",
   "organization_evidence": [
    "# What\u2019s next for Chinese open-source AI",
    "Just last week the Chinese firm Moonshot AI released its latest open-weight model, Kimi K2.5, which came close to top proprietary systems such as Anthropic\u2019s Claude Opus on some early benchmarks."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# What\u2019s next for Chinese open-source AI",
    "Just last week the Chinese firm Moonshot AI released its latest open-weight model, Kimi K2.5, which came close to top proprietary systems such as Anthropic\u2019s Claude Opus on some early benchmarks."
   ]
  },
  {
   "url": "https://huggingface.co/blog/CohereLabs/introducing-north-mini-code",
   "title": "Introducing North Mini Code: Cohere\u2019s First Model For Developers",
   "published_at": "2026-06-09T15:56:23",
   "organization": "Cohere",
   "model": null,
   "release_date": "2026-06",
   "access_type": null,
   "license": null,
   "significance": "a 30B-parameter Mixture-of-Experts model with 3B active parameters with powerful agentic coding capabilities",
   "organization_evidence": [
    "North Mini Code is the first model in Cohere\u2019s new family of models, and is specifically designed and trained for agentic software engineering tasks."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "author = {{Team Cohere}},\n    month = {June},\n    year = {2026}"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Today, we are releasing North Mini Code, a 30B-parameter Mixture-of-Experts model with 3B active parameters with powerful agentic coding capabilities, available on Hugging Face under the Apache 2.0 license."
   ]
  },
  {
   "url": "https://dynamicbusiness.com/ai-tools/inkling-open-weights-multimodal-foundation-model.html",
   "title": "Inkling: Open-Weights Multimodal Foundation Model",
   "published_at": "2026-08-21T16:24:00",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "A mixture-of-experts transformer with 975 billion parameters supporting a context length of up to one million tokens and native multimodality.",
   "organization_evidence": [
    "# Inkling: Open-Weights Multimodal Foundation Model",
    "Inkling is the latest open-weights multimodal foundation model from Thinking Machines Lab, released on July 15, 2026."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Inkling: Open-Weights Multimodal Foundation Model",
    "Inkling is the latest open-weights multimodal foundation model from Thinking Machines Lab, released on July 15, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Inkling: Open-Weights Multimodal Foundation Model",
    "Inkling is the latest open-weights multimodal foundation model from Thinking Machines Lab, released on July 15, 2026.",
    "Built as a mixture-of-experts transformer, Inkling features 975 billion parameters with 41 billion active, supports a context length of up to one million tokens, and accommodates inputs in text, image, and audio formats."
   ]
  },
  {
   "url": "https://promptzone.com/ai-model-releases?bb=4",
   "title": "AI Model Releases Timeline: Latest Launches with Dates \u2014 PromptZone - Leading AI Community for Prompt Engineering and AI Enthusiasts",
   "published_at": null,
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-08-03",
   "access_type": null,
   "license": null,
   "significance": "2.4T-parameter MoE flagship with ~95B active parameters and a 1M context window, hosted via Qwen Studio and QwenCloud at launch; open weights promised the following week.",
   "organization_evidence": [
    "A chronological timeline of major AI model releases across 2025 and 2026 \u2014 LLMs, reasoning models, open-weights families, and image & video generators, each with its date, why it mattered, and the official announcement.",
    "## August 2026 \u2014 latest AI model releases",
    "1. Aug 3, 2026 \u00b7 Alibaba"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "A chronological timeline of major AI model releases across 2025 and 2026 \u2014 LLMs, reasoning models, open-weights families, and image & video generators, each with its date, why it mattered, and the official announcement.",
    "## August 2026 \u2014 latest AI model releases",
    "1. Aug 3, 2026 \u00b7 Alibaba"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "A chronological timeline of major AI model releases across 2025 and 2026 \u2014 LLMs, reasoning models, open-weights families, and image & video generators, each with its date, why it mattered, and the official announcement.",
    "## August 2026 \u2014 latest AI model releases",
    "2.4T-parameter MoE flagship with ~95B active parameters and a 1M context window, hosted via Qwen Studio and QwenCloud at launch; open weights promised the following week."
   ]
  },
  {
   "url": "https://promptzone.com/ai-model-releases?bb=4",
   "title": "AI Model Releases Timeline: Latest Launches with Dates \u2014 PromptZone - Leading AI Community for Prompt Engineering and AI Enthusiasts",
   "published_at": null,
   "organization": "Liquid AI",
   "model": null,
   "release_date": "2026-08-04",
   "access_type": null,
   "license": null,
   "significance": "Compact 2.6B-parameter model aimed at on-device use.",
   "organization_evidence": [
    "A chronological timeline of major AI model releases across 2025 and 2026 \u2014 LLMs, reasoning models, open-weights families, and image & video generators, each with its date, why it mattered, and the official announcement.",
    "## August 2026 \u2014 latest AI model releases",
    "4. Aug 4, 2026 \u00b7 Liquid AI"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "A chronological timeline of major AI model releases across 2025 and 2026 \u2014 LLMs, reasoning models, open-weights families, and image & video generators, each with its date, why it mattered, and the official announcement.",
    "## August 2026 \u2014 latest AI model releases",
    "4. Aug 4, 2026 \u00b7 Liquid AI"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "A chronological timeline of major AI model releases across 2025 and 2026 \u2014 LLMs, reasoning models, open-weights families, and image & video generators, each with its date, why it mattered, and the official announcement.",
    "## August 2026 \u2014 latest AI model releases",
    "Compact 2.6B-parameter model aimed at on-device use."
   ]
  },
  {
   "url": "https://promptzone.com/ai-model-releases?bb=4",
   "title": "AI Model Releases Timeline: Latest Launches with Dates \u2014 PromptZone - Leading AI Community for Prompt Engineering and AI Enthusiasts",
   "published_at": null,
   "organization": "Zhipu AI",
   "model": null,
   "release_date": "2026-06-16",
   "access_type": null,
   "license": null,
   "significance": "Open-weights flagship that topped Hugging Face trending and runs on surprisingly modest hardware.",
   "organization_evidence": [
    "A chronological timeline of major AI model releases across 2025 and 2026 \u2014 LLMs, reasoning models, open-weights families, and image & video generators, each with its date, why it mattered, and the official announcement.",
    "## June 2026",
    "3. Jun 16, 2026 \u00b7 Zhipu AI"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "A chronological timeline of major AI model releases across 2025 and 2026 \u2014 LLMs, reasoning models, open-weights families, and image & video generators, each with its date, why it mattered, and the official announcement.",
    "## June 2026",
    "3. Jun 16, 2026 \u00b7 Zhipu AI"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "A chronological timeline of major AI model releases across 2025 and 2026 \u2014 LLMs, reasoning models, open-weights families, and image & video generators, each with its date, why it mattered, and the official announcement.",
    "## June 2026",
    "Open-weights flagship that topped Hugging Face trending and runs on surprisingly modest hardware."
   ]
  },
  {
   "url": "https://promptzone.com/ai-model-releases?bb=4",
   "title": "AI Model Releases Timeline: Latest Launches with Dates \u2014 PromptZone - Leading AI Community for Prompt Engineering and AI Enthusiasts",
   "published_at": null,
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "Successor to Qwen3.5 (March 2026), continuing Alibaba's fast open-weights release cadence.",
   "organization_evidence": [
    "A chronological timeline of major AI model releases across 2025 and 2026 \u2014 LLMs, reasoning models, open-weights families, and image & video generators, each with its date, why it mattered, and the official announcement.",
    "## Earlier in 2026",
    "2. Apr 2026 \u00b7 Alibaba"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "A chronological timeline of major AI model releases across 2025 and 2026 \u2014 LLMs, reasoning models, open-weights families, and image & video generators, each with its date, why it mattered, and the official announcement.",
    "## Earlier in 2026",
    "2. Apr 2026 \u00b7 Alibaba"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "A chronological timeline of major AI model releases across 2025 and 2026 \u2014 LLMs, reasoning models, open-weights families, and image & video generators, each with its date, why it mattered, and the official announcement.",
    "## Earlier in 2026",
    "Successor to Qwen3.5 (March 2026), continuing Alibaba's fast open-weights release cadence."
   ]
  },
  {
   "url": "https://promptzone.com/ai-model-releases?bb=4",
   "title": "AI Model Releases Timeline: Latest Launches with Dates \u2014 PromptZone - Leading AI Community for Prompt Engineering and AI Enthusiasts",
   "published_at": null,
   "organization": "Mistral AI",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "Compact update to the Small line, following the Voxtral TTS release in March 2026.",
   "organization_evidence": [
    "A chronological timeline of major AI model releases across 2025 and 2026 \u2014 LLMs, reasoning models, open-weights families, and image & video generators, each with its date, why it mattered, and the official announcement.",
    "## Earlier in 2026",
    "3. Apr 2026 \u00b7 Mistral AI"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "A chronological timeline of major AI model releases across 2025 and 2026 \u2014 LLMs, reasoning models, open-weights families, and image & video generators, each with its date, why it mattered, and the official announcement.",
    "## Earlier in 2026",
    "3. Apr 2026 \u00b7 Mistral AI"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "A chronological timeline of major AI model releases across 2025 and 2026 \u2014 LLMs, reasoning models, open-weights families, and image & video generators, each with its date, why it mattered, and the official announcement.",
    "## Earlier in 2026",
    "Compact update to the Small line, following the Voxtral TTS release in March 2026."
   ]
  },
  {
   "url": "https://promptzone.com/ai-model-releases?bb=4",
   "title": "AI Model Releases Timeline: Latest Launches with Dates \u2014 PromptZone - Leading AI Community for Prompt Engineering and AI Enthusiasts",
   "published_at": null,
   "organization": "Meta",
   "model": null,
   "release_date": "2026-03",
   "access_type": null,
   "license": null,
   "significance": "Mixture-of-experts entry in the Llama 4 line.",
   "organization_evidence": [
    "A chronological timeline of major AI model releases across 2025 and 2026 \u2014 LLMs, reasoning models, open-weights families, and image & video generators, each with its date, why it mattered, and the official announcement.",
    "## Earlier in 2026",
    "4. Mar 2026 \u00b7 Meta"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "A chronological timeline of major AI model releases across 2025 and 2026 \u2014 LLMs, reasoning models, open-weights families, and image & video generators, each with its date, why it mattered, and the official announcement.",
    "## Earlier in 2026",
    "4. Mar 2026 \u00b7 Meta"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "A chronological timeline of major AI model releases across 2025 and 2026 \u2014 LLMs, reasoning models, open-weights families, and image & video generators, each with its date, why it mattered, and the official announcement.",
    "## Earlier in 2026",
    "Mixture-of-experts entry in the Llama 4 line."
   ]
  },
  {
   "url": "https://promptzone.com/ai-model-releases?bb=4",
   "title": "AI Model Releases Timeline: Latest Launches with Dates \u2014 PromptZone - Leading AI Community for Prompt Engineering and AI Enthusiasts",
   "published_at": null,
   "organization": "Google DeepMind",
   "model": null,
   "release_date": "2026-02",
   "access_type": null,
   "license": null,
   "significance": "Open-weights Gemma refresh with vision capability across every size in the family.",
   "organization_evidence": [
    "A chronological timeline of major AI model releases across 2025 and 2026 \u2014 LLMs, reasoning models, open-weights families, and image & video generators, each with its date, why it mattered, and the official announcement.",
    "## Earlier in 2026",
    "5. Feb 2026 \u00b7 Google DeepMind"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "A chronological timeline of major AI model releases across 2025 and 2026 \u2014 LLMs, reasoning models, open-weights families, and image & video generators, each with its date, why it mattered, and the official announcement.",
    "## Earlier in 2026",
    "5. Feb 2026 \u00b7 Google DeepMind"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "A chronological timeline of major AI model releases across 2025 and 2026 \u2014 LLMs, reasoning models, open-weights families, and image & video generators, each with its date, why it mattered, and the official announcement.",
    "## Earlier in 2026",
    "Open-weights Gemma refresh with vision capability across every size in the family."
   ]
  },
  {
   "url": "https://tech.yahoo.com/ai/articles/china-longcat-2-0-becomes-134258951.html",
   "title": "China\u2019s LongCat-2.0 Becomes the Biggest AI Model Without Nvidia Chips",
   "published_at": "2026-06-30T13:42:58",
   "organization": "Meituan",
   "model": null,
   "release_date": "2026-06-30",
   "access_type": null,
   "license": null,
   "significance": "1.6 trillion parameter open-source large language model trained entirely without NVIDIA chips with a 1-million-token context window",
   "organization_evidence": [
    "# China\u2019s LongCat-2.0 Becomes the Biggest AI Model Without Nvidia Chips",
    "Meituan unveiled LongCat-2.0, a 1.6 trillion parameter open-source large language model."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# China\u2019s LongCat-2.0 Becomes the Biggest AI Model Without Nvidia Chips",
    "\u2014 Meituan LongCat (@Meituan_LongCat) [June 30, 2026](https://x.com/Meituan_LongCat/status/2071783587205308721?ref_src=twsrc%5Etfw)"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# China\u2019s LongCat-2.0 Becomes the Biggest AI Model Without Nvidia Chips",
    "Meituan unveiled LongCat-2.0, a 1.6 trillion parameter open-source large language model.",
    "LongCat-2.0 ranks among the largest ever, with 1.6 trillion parameters and a 1-million-token context window."
   ]
  },
  {
   "url": "https://medium.com/@vinoth.lingam333/kimi-k3-the-2-8-trillion-parameter-open-model-that-just-reset-the-frontier-01ff9df198da",
   "title": "Kimi K3: The 2.8 Trillion Parameter Open Model That Just Reset the Frontier",
   "published_at": "2026-07-18T00:00:13",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-14",
   "access_type": null,
   "license": null,
   "significance": "A 2.8 trillion parameter open-weight model with a 1 million token context window that makes K3 the largest open-weight model ever released",
   "organization_evidence": [
    "# Kimi K3: The 2.8 Trillion Parameter Open Model That Just Reset the Frontier",
    "Two days ago, Moonshot AI launched Kimi K3 \u2014 a 2.8 trillion parameter open-weight model with a 1 million token context window."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Kimi K3: The 2.8 Trillion Parameter Open Model That Just Reset the Frontier",
    "Moonshot AI (the Beijing-based company behind the Kimi chatbot) published K3 on July 14, 2026, with general availability following on July 16."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Kimi K3: The 2.8 Trillion Parameter Open Model That Just Reset the Frontier",
    "Two days ago, Moonshot AI launched Kimi K3 \u2014 a 2.8 trillion parameter open-weight model with a 1 million token context window."
   ]
  },
  {
   "url": "https://atomproject.ai/atom_report.pdf",
   "title": "The ATOM Report: Measuring the Open Language Model Ecosystem",
   "published_at": "2026-04-08T15:11:09",
   "organization": "Qwen Team",
   "model": null,
   "release_date": "2026-02",
   "access_type": null,
   "license": null,
   "significance": "major model series continuation showing strong early adoption",
   "organization_evidence": [
    "The ATOM Report",
    "April 2026",
    "The early adoption numbers of Qwen3.5 (Qwen Team, 2026) since its release in February 2026 are\nan indication that the dominance of Qwen relative to its peers will continue (see Section 6 and\nspecifically Fig. 15 for early data on Qwen 3.5\u2019s adoption)."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The ATOM Report",
    "April 2026",
    "The early adoption numbers of Qwen3.5 (Qwen Team, 2026) since its release in February 2026 are\nan indication that the dominance of Qwen relative to its peers will continue (see Section 6 and\nspecifically Fig. 15 for early data on Qwen 3.5\u2019s adoption)."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The ATOM Report",
    "April 2026",
    "The growth of Qwen-based models has even accelerated\nespecially with the updated model series of Qwen3 2507 and Qwen3.5, respectively."
   ]
  },
  {
   "url": "https://laurencemoroney.com/2026/08/21/small-ai-open-weights-on-device.html",
   "title": "Small AI in 2026: Open Weights Go On-Device",
   "published_at": "2026-08-21T00:00:00",
   "organization": "Google DeepMind",
   "model": null,
   "release_date": "2026-04-02",
   "access_type": null,
   "license": null,
   "significance": "Spans phones through workstations with variants ranging from 2.3B to 31B parameters, including multimodal capabilities and local execution.",
   "organization_evidence": [
    "## Gemma 4, and why the license change matters",
    "Google DeepMind shipped [Gemma 4](https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/) on April 2, 2026 under Apache 2.0."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## Gemma 4, and why the license change matters",
    "Google DeepMind shipped [Gemma 4](https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/) on April 2, 2026 under Apache 2.0."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## Gemma 4, and why the license change matters",
    "The sizes span phones through workstations: E2B at 2.3B effective parameters (5.1B with embeddings), E4B at 4.5B effective / 8B, a 26B A4B MoE with 3.8B active / 25.2B total, and a 31B dense model."
   ]
  },
  {
   "url": "https://laurencemoroney.com/2026/08/21/small-ai-open-weights-on-device.html",
   "title": "Small AI in 2026: Open Weights Go On-Device",
   "published_at": "2026-08-21T00:00:00",
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-02-16",
   "access_type": null,
   "license": null,
   "significance": "Natively multimodal with expanded language coverage and dense SKUs fitting local work.",
   "organization_evidence": [
    "## The rest of the permissive edge",
    "Alibaba shipped Qwen3.5 on February 16, 2026, also Apache 2.0 and natively multimodal, with language coverage moving from 119 to 201."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## The rest of the permissive edge",
    "Alibaba shipped Qwen3.5 on February 16, 2026, also Apache 2.0 and natively multimodal, with language coverage moving from 119 to 201."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## The rest of the permissive edge",
    "Alibaba shipped Qwen3.5 on February 16, 2026, also Apache 2.0 and natively multimodal, with language coverage moving from 119 to 201."
   ]
  },
  {
   "url": "https://laurencemoroney.com/2026/08/21/small-ai-open-weights-on-device.html",
   "title": "Small AI in 2026: Open Weights Go On-Device",
   "published_at": "2026-08-21T00:00:00",
   "organization": "Mistral",
   "model": null,
   "release_date": "2026-03-16",
   "access_type": null,
   "license": null,
   "significance": "119B MoE with roughly 6B active parameters and a 256K window.",
   "organization_evidence": [
    "## The rest of the permissive edge",
    "Mistral Small 4, announced March 16, 2026, needs a careful read."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## The rest of the permissive edge",
    "Mistral Small 4, announced March 16, 2026, needs a careful read."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## The rest of the permissive edge",
    "This is a 119B MoE with roughly 6B active parameters."
   ]
  },
  {
   "url": "https://huggingface.co/blog/mlabonne/glm-5",
   "title": "GLM-5: China's First Public AI Company Ships a Frontier Model",
   "published_at": "2026-02-17T12:55:21",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-02-11",
   "access_type": null,
   "license": null,
   "significance": "744B-parameter Mixture-of-Experts model with 40B active parameters per token, 200K context window, #1 open-weight model on Artificial Analysis and #1 among open models on LMArena's Text Arena",
   "organization_evidence": [
    "On February 11th, 2026, just days before the Lunar New Year, [Z.ai](https://www.linkedin.com/company/zdotai/) officially released GLM-5, its new frontier large language model.",
    "First, congrats to the [Z.ai](https://www.linkedin.com/company/zdotai/) team on a strong release."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "On February 11th, 2026, just days before the Lunar New Year, [Z.ai](https://www.linkedin.com/company/zdotai/) officially released GLM-5, its new frontier large language model."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "GLM-5 is a 744B-parameter Mixture-of-Experts model with 40B active parameters per token.",
    "The model also switched its attention mechanism to DeepSeek Sparse Attention (DSA) for efficient long-context handling, supporting a 200K-token context window.",
    "GLM-5 is the new #1 open-weight model on Artificial Analysis and hit #1 among open models on LMArena\u2019s Text Arena (score 1452, #11 overall)."
   ]
  },
  {
   "url": "https://edgen.tech/fr/news/post/mira-muratis-inkling-debuts-as-best-open-source-model-in-the-west",
   "title": "Mira Murati's Inkling debuts as best open-source model in the West",
   "published_at": "2026-07-26T14:42:40",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-26",
   "access_type": null,
   "license": null,
   "significance": "A 975-billion-parameter MoE model with 41 billion active parameters that is the best Western-trained open-weight AI model and scores 74.1% on MCP Atlas for agentic tool use.",
   "organization_evidence": [
    "Mira Murati's Inkling debuts as best open-source model in the West",
    "Mira Murati's Thinking Machines Lab released Inkling, a 975-billion-parameter open-source model that is the best Western-trained open-weight AI model \u2014 but its narrow strengths and heavy censorship limit its appeal to most developers."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Mira Murati's Inkling debuts as best open-source model in the West",
    "[![Rachel Kim](https://static.edgen.tech/authors/rachel-kim.png?image_process=resize,w_32/quality,q_75/format,webp/sharpen,100)\n## Rachel Kim](https://www.edgen.tech/fr/author/news/rachel-kim)\u00b7Jul 26 2026, 14:42"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Mira Murati's Inkling debuts as best open-source model in the West",
    "Mira Murati's Thinking Machines Lab released Inkling, a 975-billion-parameter open-source model that is the best Western-trained open-weight AI model \u2014 but its narrow strengths and heavy censorship limit its appeal to most developers.",
    "Inkling is a 975-billion-parameter MoE model with 41 billion active parameters at inference"
   ]
  },
  {
   "url": "https://huggingface.co/blog/LiquidAI/qad",
   "title": "LFM2.5 Q4\\_0 Checkpoints from Quantization-Aware Distillation",
   "published_at": "2026-08-19T13:48:49",
   "organization": "Liquid AI",
   "model": null,
   "release_date": "2026-08",
   "access_type": null,
   "license": null,
   "significance": "Quantization-Aware Distillation (QAD) 4-bit checkpoints that maintain high accuracy, low memory footprint, and high throughput across four models for edge deployment.",
   "organization_evidence": [
    "Liquid AI, \"LFM2.5 Q4_0: Quantization-Aware Distillation for Edge Deployment\", Liquid AI Blog, Aug 2026.",
    "@article{liquidAI2026Q40,",
    "  author = {Liquid AI},"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Liquid AI, \"LFM2.5 Q4_0: Quantization-Aware Distillation for Edge Deployment\", Liquid AI Blog, Aug 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Today, we release QAD Q4_0 GGUFs.",
    "These are updated 4-bit checkpoints for LFM2.5-230M, LFM2.5-350M, LFM2.5-1.2B-Instruct, and LFM2.5-2.6B.",
    "Liquid AI, \"LFM2.5 Q4_0: Quantization-Aware Distillation for Edge Deployment\", Liquid AI Blog, Aug 2026."
   ]
  },
  {
   "url": "https://machineherald.io/article/2026-07/06-meituan-open-sources-longcat-20-a-16-trillion-parameter-coding-model-it-says-was-trained-end-to-end-on-chinese-chips",
   "title": "Meituan Open-Sources LongCat-2.0, a 1.6-Trillion-Parameter Coding Model It Says Was Trained End-to-End on Chinese Chips",
   "published_at": "2026-07-06T12:25:03",
   "organization": "Meituan",
   "model": "LongCat-2.0",
   "release_date": "2026-06-30",
   "access_type": "open-sourced",
   "license": "MIT license",
   "significance": "a 1.6T-parameter sparse mixture-of-experts model with a native 1-million-token context window trained and served entirely on domestic ASICs",
   "organization_evidence": [
    "# Meituan Open-Sources LongCat-2.0, a 1.6-Trillion-Parameter Coding Model It Says Was Trained End-to-End on Chinese Chips",
    "Meituan released LongCat-2.0, a 1.6T-parameter MoE model under an MIT license, claiming it is the first trillion-parameter model trained and served entirely on domestic ASICs."
   ],
   "model_evidence": [
    "# Meituan Open-Sources LongCat-2.0, a 1.6-Trillion-Parameter Coding Model It Says Was Trained End-to-End on Chinese Chips",
    "Meituan released LongCat-2.0, a 1.6T-parameter MoE model under an MIT license, claiming it is the first trillion-parameter model trained and served entirely on domestic ASICs."
   ],
   "release_date_evidence": [
    "# Meituan Open-Sources LongCat-2.0, a 1.6-Trillion-Parameter Coding Model It Says Was Trained End-to-End on Chinese Chips",
    "The model was released on June 30, 2026 under an MIT license, and its weights are published on Hugging Face, according to the official [Hugging Face](https://huggingface.co/meituan-longcat/LongCat-2.0) model card."
   ],
   "access_type_evidence": [
    "# Meituan Open-Sources LongCat-2.0, a 1.6-Trillion-Parameter Coding Model It Says Was Trained End-to-End on Chinese Chips",
    "Meituan, the Chinese food-delivery and local-services company, has open-sourced LongCat-2.0, a large language model with 1.6 trillion parameters and a native 1-million-token context window, according to the [South China Morning Post](https://www.scmp.com/tech/tech-trends/article/3358854/china-debuts-biggest-ai-model-trained-local-chips-meituan-releases-longcat-20)."
   ],
   "license_evidence": [
    "Meituan released LongCat-2.0, a 1.6T-parameter MoE model under an MIT license, claiming it is the first trillion-parameter model trained and served entirely on domestic ASICs.",
    "# Meituan Open-Sources LongCat-2.0, a 1.6-Trillion-Parameter Coding Model It Says Was Trained End-to-End on Chinese Chips",
    "The model was released on June 30, 2026 under an MIT license, and its weights are published on Hugging Face, according to the official [Hugging Face](https://huggingface.co/meituan-longcat/LongCat-2.0) model card."
   ],
   "significance_evidence": [
    "# Meituan Open-Sources LongCat-2.0, a 1.6-Trillion-Parameter Coding Model It Says Was Trained End-to-End on Chinese Chips",
    "Meituan released LongCat-2.0, a 1.6T-parameter MoE model under an MIT license, claiming it is the first trillion-parameter model trained and served entirely on domestic ASICs.",
    "Meituan, the Chinese food-delivery and local-services company, has open-sourced LongCat-2.0, a large language model with 1.6 trillion parameters and a native 1-million-token context window, according to the [South China Morning Post](https://www.scmp.com/tech/tech-trends/article/3358854/china-debuts-biggest-ai-model-trained-local-chips-meituan-releases-longcat-20)."
   ]
  },
  {
   "url": "https://tech-insider.org/best-open-source-llm-2026",
   "title": "Best Open Source LLM 2026: DeepSeek, Kimi, Qwen Ranked",
   "published_at": "2026-06-03T10:00:00",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "80.6% SWE-Bench Verified, 90.1% GPQA Diamond, and a LiveCodeBench of 93.5 with a 1M-token context window",
   "organization_evidence": [
    "Full Specifications Compared",
    "| Developer | DeepSeek | Moonshot AI | Zhipu / Z.ai | Alibaba | Meta |",
    "DeepSeek V4-Pro: The Open-Weight Leader",
    "Released April 24, 2026 under the MIT license, DeepSeek V4-Pro is the model that sits at the top of nearly every open leaderboard in mid-2026."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Full Specifications Compared",
    "| Release date | Apr 24, 2026 | Apr 20, 2026 | Sep 30, 2025 | 2025 (2507 update) | Apr 5, 2025 |",
    "DeepSeek V4-Pro: The Open-Weight Leader",
    "Released April 24, 2026 under the MIT license, DeepSeek V4-Pro is the model that sits at the top of nearly every open leaderboard in mid-2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "DeepSeek V4-Pro: The Open-Weight Leader",
    "Its 1.6-trillion-parameter MoE (49B active) pairs a 1M-token context window with the strongest all-round open scores: 80.6% SWE-Bench Verified, 90.1% GPQA Diamond, and a LiveCodeBench of 93.5."
   ]
  },
  {
   "url": "https://tech-insider.org/best-open-source-llm-2026",
   "title": "Best Open Source LLM 2026: DeepSeek, Kimi, Qwen Ranked",
   "published_at": "2026-06-03T10:00:00",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-04-20",
   "access_type": null,
   "license": null,
   "significance": "leads the SWE-Bench Pro board at 58.6% with agentic endurance and tool-use capabilities",
   "organization_evidence": [
    "Full Specifications Compared",
    "| Developer | DeepSeek | Moonshot AI | Zhipu / Z.ai | Alibaba | Meta |",
    "Kimi K2.6: The Agentic Coding Specialist",
    "Moonshot AI\u2019s Kimi K2.6, released April 20, 2026, is a 1-trillion-parameter MoE (~32B active) with a 256K context window and a Modified MIT license."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Full Specifications Compared",
    "| Release date | Apr 24, 2026 | Apr 20, 2026 | Sep 30, 2025 | 2025 (2507 update) | Apr 5, 2025 |",
    "Kimi K2.6: The Agentic Coding Specialist",
    "Moonshot AI\u2019s Kimi K2.6, released April 20, 2026, is a 1-trillion-parameter MoE (~32B active) with a 256K context window and a Modified MIT license."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Kimi K2.6: The Agentic Coding Specialist",
    "Its standout trait is agentic endurance: Kimi is built to plan, call tools, and run multi-step tasks without losing the thread, and it leads the SWE-Bench Pro board at 58.6%."
   ]
  },
  {
   "url": "https://iaexpertos.net/en/blog/kimi-k3-the-worlds-largest-open-weight-model-and-its-impact-on-the-global-ai-balance-en",
   "title": "Kimi K3: The World\u2019s Largest Open-Weight Model and Its Impact on the Global AI Balance",
   "published_at": "2026-07-17T09:00:45",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-17",
   "access_type": null,
   "license": null,
   "significance": "a 2.8 trillion parameter language model that becomes the largest open-weight model ever created",
   "organization_evidence": [
    "# Kimi K3: The World\u2019s Largest Open-Weight Model and Its Impact on the Global AI Balance",
    "On July 17, 2026, Chinese startup Moonshot AI, backed by Alibaba, released Kimi K3, a 2.8 trillion parameter language model that becomes the largest open-weight model ever created."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Kimi K3: The World\u2019s Largest Open-Weight Model and Its Impact on the Global AI Balance",
    "On July 17, 2026, Chinese startup Moonshot AI, backed by Alibaba, released Kimi K3, a 2.8 trillion parameter language model that becomes the largest open-weight model ever created."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Kimi K3: The World\u2019s Largest Open-Weight Model and Its Impact on the Global AI Balance",
    "On July 17, 2026, Chinese startup Moonshot AI, backed by Alibaba, released Kimi K3, a 2.8 trillion parameter language model that becomes the largest open-weight model ever created."
   ]
  },
  {
   "url": "https://edgen.tech/fr/news/post/zai-confirms-ox-alpha-as-glm-model-with-usage-2x-deepseek",
   "title": "Z.AI confirms Ox Alpha as GLM model with usage 2x DeepSeek",
   "published_at": "2026-08-26T12:16:39",
   "organization": "Z.AI",
   "model": null,
   "release_date": "2026-08-26",
   "access_type": null,
   "license": null,
   "significance": "reasoning model carrying a 1,048,600-token context window that more than doubled DeepSeek's usage on OpenRouter",
   "organization_evidence": [
    "# Z.AI confirms Ox Alpha as GLM model with usage 2x DeepSeek",
    "The mystery AI model that swept to the top of OpenRouter's usage charts is a new GLM-family release from China's Z.AI, which confirmed the launch Wednesday and said it will release the model's weights tonight."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Z.AI confirms Ox Alpha as GLM model with usage 2x DeepSeek",
    "[![Alex Nguyen](https://static.edwealth.ai/authors/alex-nguyen.png?w=32)\n## Alex Nguyen](https://www.edgen.tech/fr/author/news/alex-nguyen)\u00b7Aug 26 2026, 05:16",
    "The mystery AI model that swept to the top of OpenRouter's usage charts is a new GLM-family release from China's Z.AI, which confirmed the launch Wednesday and said it will release the model's weights tonight."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Z.AI confirms Ox Alpha as GLM model with usage 2x DeepSeek",
    "Ox Alpha's usage more than doubled DeepSeek's on OpenRouter's leaderboard, making it the biggest launch in the marketplace's history, according to the company."
   ]
  },
  {
   "url": "https://pioneer.ai/blog/closing-the-gap-open-weight-vs.-proprietary-frontier-language-models",
   "title": "Closing the Gap: Open-Weight vs. Proprietary Frontier Language Models",
   "published_at": "2026-07-20T00:00:00",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-06-16",
   "access_type": null,
   "license": null,
   "significance": "A 744-billion-parameter mixture-of-experts model with a 1-million-token context window that scores 51 on the Artificial Analysis Intelligence Index.",
   "organization_evidence": [
    "Research",
    "## Who are Z.ai and Moonshot AI?",
    "Z.ai shipped GLM-5.2 in mid-June under an MIT license, API live on June 16."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Research",
    "## What they actually shipped",
    "Z.ai shipped GLM-5.2 in mid-June under an MIT license, API live on June 16."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Research",
    "**On capability:** Moonshot AI's Kimi K3 scores 57 on the Artificial Analysis Intelligence Index, a composite of nine evaluations.",
    "[Z.ai](http://z.ai/)'s GLM-5.2 follows with a score of 51.",
    "**GLM-5.2** is a 744-billion-parameter mixture-of-experts (MoE) model with about 40 billion active parameters per forward pass, a 1-million-token context window, and up to 128K output tokens."
   ]
  },
  {
   "url": "https://pioneer.ai/blog/closing-the-gap-open-weight-vs.-proprietary-frontier-language-models",
   "title": "Closing the Gap: Open-Weight vs. Proprietary Frontier Language Models",
   "published_at": "2026-07-20T00:00:00",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-27",
   "access_type": null,
   "license": null,
   "significance": "A 2.8-trillion-parameter MoE model, the first open model to reach the 3-trillion class, scoring 57 on the Artificial Analysis Intelligence Index.",
   "organization_evidence": [
    "Research",
    "## Who are Z.ai and Moonshot AI?",
    "A month later, on July 16, Moonshot AI announced Kimi K3, with API access at launch and full weights arriving on Hugging Face by July 27."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Research",
    "## What they actually shipped",
    "A month later, on July 16, Moonshot AI announced Kimi K3, with API access at launch and full weights arriving on Hugging Face by July 27."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Research",
    "**On capability:** Moonshot AI's Kimi K3 scores 57 on the Artificial Analysis Intelligence Index, a composite of nine evaluations.",
    "**Kimi K3** is a 2.8-trillion-parameter MoE model, the first open model to reach the 3-trillion class."
   ]
  },
  {
   "url": "https://venturebeat.com/technology/z-ai-debuts-faster-cheaper-glm-5-turbo-model-for-agents-and-claws-but-its",
   "title": "z.ai debuts faster, cheaper GLM-5 Turbo model for agents and 'claws' \u2014 but it's not open-source",
   "published_at": "2026-03-16T19:36:30",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-02",
   "access_type": null,
   "license": null,
   "significance": "scaled to 744 billion parameters with 40 billion active per token in a mixture-of-experts architecture, used 28.5 trillion pretraining tokens, and achieved a record-low hallucination score on the AA-Omniscience Index",
   "organization_evidence": [
    "[Z.ai\u2019s last major release, GLM-5](https://venturebeat.com/technology/z-ais-open-source-glm-5-achieves-record-low-hallucination-rate-and-leverages), which debuted in February 2026, gives useful context for what the company is now trying to do with GLM-5-Turbo.",
    "## **Background: z.ai and GLM-5 set the stage for Turbo**"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "[Z.ai\u2019s last major release, GLM-5](https://venturebeat.com/technology/z-ais-open-source-glm-5-achieves-record-low-hallucination-rate-and-leverages), which debuted in February 2026, gives useful context for what the company is now trying to do with GLM-5-Turbo.",
    "## **Background: z.ai and GLM-5 set the stage for Turbo**"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "GLM-5 is an open-source flagship model carrying an MIT license, posting a record-low hallucination score on the AA-Omniscience Index, and debuted a native \u201cAgent Mode\u201d that could turn prompts or source materials into ready-to-use .docx, .pdf and .xlsx files.",
    "That earlier release was also framed as a major technical step up for the company.",
    "GLM-5 scaled to 744 billion parameters with 40 billion active per token in a mixture-of-experts architecture, used 28.5 trillion pretraining tokens, and relied on a new asynchronous reinforcement-learning infrastructure called \u201cslime\u201d to reduce training bottlenecks and support more complex agentic behavior.",
    "## **Background: z.ai and GLM-5 set the stage for Turbo**"
   ]
  },
  {
   "url": "https://artificialanalysis.ai/articles/sarvam-105b-Sarvam-30b-everything-you-need-to-know",
   "title": "Sarvam 105B and Sarvam 30B: India enters the open-weights race",
   "published_at": "2026-04-02T00:00:00",
   "organization": "SarvamAI",
   "model": null,
   "release_date": "2026",
   "access_type": null,
   "license": null,
   "significance": "India's largest models pre-trained from scratch, scoring 18 and 12 on the Artificial Analysis Intelligence Index with reasoning and non-reasoning support",
   "organization_evidence": [
    "# Sarvam 105B and Sarvam 30B: India enters the open-weights race",
    "SarvamAI's Sarvam 105B and Sarvam 30B score 18 and 12 on the Artificial Analysis Intelligence Index respectively."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Sarvam 105B and Sarvam 30B: India enters the open-weights race",
    "Announced at the India AI Impact Summit 2026 and open-sourced under Apache 2.0, both are Mixture-of-Experts models trained entirely in India using compute provided under the IndiaAI Mission."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Sarvam 105B and Sarvam 30B: India enters the open-weights race",
    "**India enters the open-weights AI race with its largest models pre-trained from scratch: Sarvam 105B and Sarvam 30B**",
    "SarvamAI's Sarvam 105B and Sarvam 30B score 18 and 12 on the Artificial Analysis Intelligence Index respectively."
   ]
  },
  {
   "url": "https://techstartups.com/2026/07/15/thinking-machines-unveils-inkling-a-powerful-open-weights-ai-model-that-challenges-kimi-and-nemotron",
   "title": "Thinking Machines launches Inkling, a powerful open-weights AI model that challenges Kimi and Nemotron",
   "published_at": "2026-07-15T22:59:58",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "A 975-billion-parameter Mixture-of-Experts foundation model built for multimodal reasoning, coding, tool use, and long-context workloads.",
   "organization_evidence": [
    "# Thinking Machines launches Inkling, a powerful open-weights AI model that challenges Kimi and Nemotron",
    "Former OpenAI CTO Mira Murati\u2019s AI startup, [Thinking Machines Lab](https://thinkingmachines.ai/), has entered the race for open AI models with a release that is already drawing attention across the developer community."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Thinking Machines launches Inkling, a powerful open-weights AI model that challenges Kimi and Nemotron",
    "Posted On July 15, 2026"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Thinking Machines launches Inkling, a powerful open-weights AI model that challenges Kimi and Nemotron",
    "The company on Tuesday introduced Inkling, its first open-weights foundation model, a 975-billion-parameter Mixture-of-Experts (MoE) system built for multimodal reasoning, coding, tool use, and long-context workloads."
   ]
  },
  {
   "url": "https://llm-releases.com/latest",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-07-31",
   "access_type": null,
   "license": null,
   "significance": "the production release of DeepSeek's V4-Flash tier \u2014 the April V4-Flash preview retrained on a substantially improved post-training pipeline",
   "organization_evidence": [
    "Fresh model coverage",
    "# Latest LLM releases",
    "The newest model releases, previews, and notable announcements in the catalog, sorted by release or announcement date and backed by source links.",
    "Weights released on Hugging Face under the MIT license; API pricing held at $0.14 / $0.28 per Mtok."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Fresh model coverage",
    "# Latest LLM releases",
    "The newest model releases, previews, and notable announcements in the catalog, sorted by release or announcement date and backed by source links.",
    "DeepSeek reports the 0731 build scoring higher than its own larger V4-Pro-Preview on all nine agent and coding benchmarks it published \u2014 a vendor-reported result, with independent replication still limited at launch."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Fresh model coverage",
    "# Latest LLM releases",
    "The newest model releases, previews, and notable announcements in the catalog, sorted by release or announcement date and backed by source links.",
    "The production release of DeepSeek's V4-Flash tier \u2014 the April V4-Flash preview retrained on a substantially improved post-training pipeline targeting coding, agents, reasoning, and tool use, with no change to the base architecture."
   ]
  },
  {
   "url": "https://writingmate.ai/blog/inkling-writingmate-agentic-coding-test-2026",
   "title": "Inkling Is on Writingmate: Testing Thinking Machines Lab's First Open-Weight Model for Agentic Coding and Tool Use",
   "published_at": "2026-07-18T12:00:00",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-17",
   "access_type": null,
   "license": null,
   "significance": "975 billion total parameters, only 41 billion active per token, a declared 1,048,576-token context window, and native audio input alongside text and image",
   "organization_evidence": [
    "Thinking Machines Lab shipped its first open-weight model on July 17, 2026, and the spec sheet is specific enough to actually test instead of skim: 975 billion total parameters, only 41 billion active per token, a declared 1,048,576-token context window, and native audio input alongside text and image.",
    "## What's actually new in Inkling",
    "| Field | Inkling | Reader takeaway |",
    "| Provider | Thinking Machines Lab | Mira Murati's first open-weight release since founding the company in 2025 |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Thinking Machines Lab shipped its first open-weight model on July 17, 2026, and the spec sheet is specific enough to actually test instead of skim: 975 billion total parameters, only 41 billion active per token, a declared 1,048,576-token context window, and native audio input alongside text and image.",
    "## What's actually new in Inkling",
    "| Field | Inkling | Reader takeaway |",
    "| Availability | July 17, 2026 in the Writingmate catalog | Live to test today, not waitlisted |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Thinking Machines Lab shipped its first open-weight model on July 17, 2026, and the spec sheet is specific enough to actually test instead of skim: 975 billion total parameters, only 41 billion active per token, a declared 1,048,576-token context window, and native audio input alongside text and image.",
    "## What's actually new in Inkling"
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/qwen3-8-27b",
   "title": "Qwen3.8 27B (xhigh) - Intelligence, Performance & Price Analysis | Artificial Analysis",
   "published_at": null,
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-08-14",
   "access_type": null,
   "license": null,
   "significance": "A leading reasoning model with 27B parameters, multimodal support for text, image, and video, and a 256k context window.",
   "organization_evidence": [
    "Qwen3.8 27B (xhigh) was created by Alibaba.",
    "Qwen3.8 27B (xhigh) Intelligence, Performance & Price Analysis"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Qwen3.8 27B (xhigh) was released on August 14, 2026.",
    "Qwen3.8 27B (xhigh) Intelligence, Performance & Price Analysis"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Qwen3.8 27B (xhigh) is amongst the leading models in intelligence, but particularly expensive when comparing to other open weight models of similar size.",
    "The model supports text, image, and video input, outputs text, and has a 256k tokens context window.",
    "Qwen3.8 27B (xhigh) Intelligence, Performance & Price Analysis"
   ]
  },
  {
   "url": "https://medium.com/@ai-engineering-trend/kimi-k3-launches-2-8t-parameters-push-chinese-open-source-ai-to-new-heights-my-first-hand-test-5849341e9840",
   "title": "Kimi K3 Officially Launches: 2.8T",
   "published_at": "2026-07-17T05:24:06",
   "organization": "Kimi",
   "model": null,
   "release_date": "2026-07-17",
   "access_type": null,
   "license": null,
   "significance": "boasts 2.8T total parameters, native visual understanding support, and a 1M token context window",
   "organization_evidence": [
    "# Kimi K3 Officially Launches: 2.8T Parameters, Chinese Open-Source AI Reaches A New Milestone, My Fresh First-Hand Review",
    "By winkrun, *AI Engineering* \u00b7 July 17, 2026 \u00b7 Beijing, China",
    "I never expected Kimi K3 would land directly at 2.8 trillion parameters, pushing Chinese open-source large language models to an entirely new level."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Kimi K3 Officially Launches: 2.8T Parameters, Chinese Open-Source AI Reaches A New Milestone, My Fresh First-Hand Review",
    "By winkrun, *AI Engineering* \u00b7 July 17, 2026 \u00b7 Beijing, China",
    "Last week, I saw people speculating in AI communities whether Kimi\u2019s next model would break the trillion-parameter barrier."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Kimi K3 Officially Launches: 2.8T Parameters, Chinese Open-Source AI Reaches A New Milestone, My Fresh First-Hand Review",
    "The newly launched Kimi K3 boasts 2.8T total parameters, native visual understanding support, and a 1M token context window."
   ]
  },
  {
   "url": "https://huggingface.co/blog/muse-glimmer",
   "title": "Meta is back with Muse Glimmer: local, agentic, multimodal, and open source",
   "published_at": "2026-08-10T00:00:00",
   "organization": "Meta",
   "model": null,
   "release_date": "2026-08-10",
   "access_type": null,
   "license": null,
   "significance": "A 30B multimodal model designed for local agentic use cases with hybrid attention and multimodal tool calling.",
   "organization_evidence": [
    "Published August 10, 2026",
    "Great news from the OGs of open source LLMs! Muse Glimmer, released today, is Meta\u2019s new multimodal model, especially designed for local agentic use cases."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Published August 10, 2026",
    "Great news from the OGs of open source LLMs! Muse Glimmer, released today, is Meta\u2019s new multimodal model, especially designed for local agentic use cases."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Published August 10, 2026",
    "Great news from the OGs of open source LLMs! Muse Glimmer, released today, is Meta\u2019s new multimodal model, especially designed for local agentic use cases."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/inkling",
   "title": "Inkling - Intelligence, Performance & Price Analysis | Artificial Analysis",
   "published_at": null,
   "organization": "Thinking Machines",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "975 billion total parameters, 41 billion active parameters, 1M tokens context window, and scores 42 on the Artificial Analysis Intelligence Index",
   "organization_evidence": [
    "Inkling (xhigh) Intelligence, Performance & Price Analysis",
    "Inkling (xhigh) was created by Thinking Machines."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Inkling (xhigh) Intelligence, Performance & Price Analysis",
    "Inkling (xhigh) was released on July 15, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Inkling (xhigh) Intelligence, Performance & Price Analysis",
    "Inkling (xhigh) scores 42 on the Artificial Analysis Intelligence Index, placing it well above average among comparable models (median: 29).",
    "The model supports text, image, and speech input, outputs text, and has a 1M tokens context window.",
    "Total parameters | 975B",
    "Active parameters | 41B Number of parameters active per token during inference"
   ]
  },
  {
   "url": "https://codersera.com/blog/best-open-source-llm-2026-llama-4-qwen-3-5-deepseek-v4-gemma-4-mistral/amp",
   "title": "Best Open-Source LLM 2026: Kimi, DeepSeek V4, Qwen, Llama 4",
   "published_at": "2026-05-03T08:27:05",
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-08-12",
   "access_type": null,
   "license": null,
   "significance": "Posts the highest neutral AA Index of any open model (~57.7) for frontier agentic coding at datacentre-scale only",
   "organization_evidence": [
    "# Best Open-Source LLM 2026: Llama 4, DeepSeek V4, Qwen, Kimi",
    "**Quick answer.** The best open-source LLM as of August 2026 depends on workload.",
    "**Qwen3.8-2.4T-A95B** (opened August 12 under a custom licence) now posts the highest neutral AA Index of any open model (~57.7)"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Best Open-Source LLM 2026: Llama 4, DeepSeek V4, Qwen, Kimi",
    "**Quick answer.** The best open-source LLM as of August 2026 depends on workload.",
    "**Qwen3.8-2.4T-A95B** (opened August 12 under a custom licence) now posts the highest neutral AA Index of any open model (~57.7)"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Best Open-Source LLM 2026: Llama 4, DeepSeek V4, Qwen, Kimi",
    "**Quick answer.** The best open-source LLM as of August 2026 depends on workload.",
    "**Qwen3.8-2.4T-A95B** (opened August 12 under a custom licence) now posts the highest neutral AA Index of any open model (~57.7)"
   ]
  },
  {
   "url": "https://codersera.com/blog/best-open-source-llm-2026-llama-4-qwen-3-5-deepseek-v4-gemma-4-mistral/amp",
   "title": "Best Open-Source LLM 2026: Kimi, DeepSeek V4, Qwen, Llama 4",
   "published_at": "2026-05-03T08:27:05",
   "organization": "Moonshot",
   "model": null,
   "release_date": "2026-04-20",
   "access_type": null,
   "license": null,
   "significance": "Leads the rest at 54 on the Artificial Analysis Intelligence Index, landing at #4 across all models",
   "organization_evidence": [
    "# Best Open-Source LLM 2026: Llama 4, DeepSeek V4, Qwen, Kimi",
    "**Quick answer.** The best open-source LLM as of August 2026 depends on workload.",
    "**Kimi K2.6** (Moonshot) leads the rest at 54"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Best Open-Source LLM 2026: Llama 4, DeepSeek V4, Qwen, Kimi",
    "**Quick answer.** The best open-source LLM as of August 2026 depends on workload.",
    "Moonshot AI released [Kimi K2.6](https://huggingface.co/moonshotai/Kimi-K2.6?ref=codersera.com) on April 20, 2026: a Mixture-of-Experts model with **1 trillion total parameters, 32B active per token**, a **256K context window**, and native image + video input."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Best Open-Source LLM 2026: Llama 4, DeepSeek V4, Qwen, Kimi",
    "**Quick answer.** The best open-source LLM as of August 2026 depends on workload.",
    "**Kimi K2.6** (Moonshot) leads the rest at 54"
   ]
  },
  {
   "url": "https://codersera.com/blog/best-open-source-llm-2026-llama-4-qwen-3-5-deepseek-v4-gemma-4-mistral/amp",
   "title": "Best Open-Source LLM 2026: Kimi, DeepSeek V4, Qwen, Llama 4",
   "published_at": "2026-05-03T08:27:05",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "Wins agentic coding and ties the closed frontier on SWE-Bench",
   "organization_evidence": [
    "# Best Open-Source LLM 2026: Llama 4, DeepSeek V4, Qwen, Kimi",
    "**Quick answer.** The best open-source LLM as of August 2026 depends on workload.",
    "**DeepSeek V4 Pro** wins agentic coding and ties the closed frontier on SWE-Bench"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Best Open-Source LLM 2026: Llama 4, DeepSeek V4, Qwen, Kimi",
    "**Quick answer.** The best open-source LLM as of August 2026 depends on workload.",
    "DeepSeek released [V4 Pro and V4 Flash](https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro?ref=codersera.com) on April 24, 2026, both **MIT-licensed** with a **1M-token context**."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Best Open-Source LLM 2026: Llama 4, DeepSeek V4, Qwen, Kimi",
    "**Quick answer.** The best open-source LLM as of August 2026 depends on workload.",
    "**DeepSeek V4 Pro** wins agentic coding and ties the closed frontier on SWE-Bench"
   ]
  },
  {
   "url": "https://codersera.com/blog/best-open-source-llm-2026-llama-4-qwen-3-5-deepseek-v4-gemma-4-mistral/amp",
   "title": "Best Open-Source LLM 2026: Kimi, DeepSeek V4, Qwen, Llama 4",
   "published_at": "2026-05-03T08:27:05",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-04-01",
   "access_type": null,
   "license": null,
   "significance": "Has the cleanest MIT license and ranks second only to DeepSeek V4 Pro on the GDPval-AA agentic leaderboard",
   "organization_evidence": [
    "# Best Open-Source LLM 2026: Llama 4, DeepSeek V4, Qwen, Kimi",
    "**Quick answer.** The best open-source LLM as of August 2026 depends on workload.",
    "**GLM-5.1** has the cleanest MIT license"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Best Open-Source LLM 2026: Llama 4, DeepSeek V4, Qwen, Kimi",
    "Between early April and mid-May 2026, Moonshot shipped **Kimi K2.6**, Z.ai shipped **GLM-5.1**, DeepSeek shipped **V4 Pro and V4 Flash**, Xiaomi shipped **MiMo-V2.5-Pro**, MiniMax open-sourced **M2.7**, Google released **Gemma 4**, Alibaba released **Qwen 3.6**, and Ant Group's inclusionAI released **Ring-2.6-1T**."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Best Open-Source LLM 2026: Llama 4, DeepSeek V4, Qwen, Kimi",
    "**Quick answer.** The best open-source LLM as of August 2026 depends on workload.",
    "**GLM-5.1** has the cleanest MIT license"
   ]
  },
  {
   "url": "https://codersera.com/blog/best-open-source-llm-2026-llama-4-qwen-3-5-deepseek-v4-gemma-4-mistral/amp",
   "title": "Best Open-Source LLM 2026: Kimi, DeepSeek V4, Qwen, Llama 4",
   "published_at": "2026-05-03T08:27:05",
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-08-14",
   "access_type": null,
   "license": null,
   "significance": "Best small dense coder under Apache-2.0 with native vision-language and 262K context",
   "organization_evidence": [
    "# Best Open-Source LLM 2026: Llama 4, DeepSeek V4, Qwen, Kimi",
    "**Quick answer.** The best open-source LLM as of August 2026 depends on workload.",
    "**Qwen3.8-27B** (August 14) is the best small dense coder under Apache-2.0"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Best Open-Source LLM 2026: Llama 4, DeepSeek V4, Qwen, Kimi",
    "**Quick answer.** The best open-source LLM as of August 2026 depends on workload.",
    "**Qwen3.8-27B** (August 14) is the best small dense coder under Apache-2.0"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Best Open-Source LLM 2026: Llama 4, DeepSeek V4, Qwen, Kimi",
    "**Quick answer.** The best open-source LLM as of August 2026 depends on workload.",
    "**Qwen3.8-27B** (August 14) is the best small dense coder under Apache-2.0"
   ]
  },
  {
   "url": "https://codersera.com/blog/best-open-source-llm-2026-llama-4-qwen-3-5-deepseek-v4-gemma-4-mistral/amp",
   "title": "Best Open-Source LLM 2026: Kimi, DeepSeek V4, Qwen, Llama 4",
   "published_at": "2026-05-03T08:27:05",
   "organization": "Meta",
   "model": null,
   "release_date": "2026-08-10",
   "access_type": null,
   "license": null,
   "significance": "Strongest tool-calling model in the 30B class with built-in vision encoder under Apache-2.0",
   "organization_evidence": [
    "# Best Open-Source LLM 2026: Llama 4, DeepSeek V4, Qwen, Kimi",
    "**Quick answer.** The best open-source LLM as of August 2026 depends on workload.",
    "**Muse Glimmer** (August 10) is the strongest tool-calling model in the 30B class."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Best Open-Source LLM 2026: Llama 4, DeepSeek V4, Qwen, Kimi",
    "**Quick answer.** The best open-source LLM as of August 2026 depends on workload.",
    "Yes \u2014 as of August 10, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Best Open-Source LLM 2026: Llama 4, DeepSeek V4, Qwen, Kimi",
    "**Quick answer.** The best open-source LLM as of August 2026 depends on workload.",
    "**Muse Glimmer** (August 10) is the strongest tool-calling model in the 30B class."
   ]
  },
  {
   "url": "https://codersera.com/blog/best-open-source-llm-2026-llama-4-qwen-3-5-deepseek-v4-gemma-4-mistral/amp",
   "title": "Best Open-Source LLM 2026: Kimi, DeepSeek V4, Qwen, Llama 4",
   "published_at": "2026-05-03T08:27:05",
   "organization": "Xiaomi",
   "model": null,
   "release_date": "2026-04-01",
   "access_type": null,
   "license": null,
   "significance": "Ties #1 by index with a 1M context window under Apache-2.0",
   "organization_evidence": [
    "# Best Open-Source LLM 2026: Llama 4, DeepSeek V4, Qwen, Kimi",
    "Between early April and mid-May 2026, Moonshot shipped **Kimi K2.6**, Z.ai shipped **GLM-5.1**, DeepSeek shipped **V4 Pro and V4 Flash**, Xiaomi shipped **MiMo-V2.5-Pro**, MiniMax open-sourced **M2.7**, Google released **Gemma 4**, Alibaba released **Qwen 3.6**, and Ant Group's inclusionAI released **Ring-2.6-1T**."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Best Open-Source LLM 2026: Llama 4, DeepSeek V4, Qwen, Kimi",
    "Between early April and mid-May 2026, Moonshot shipped **Kimi K2.6**, Z.ai shipped **GLM-5.1**, DeepSeek shipped **V4 Pro and V4 Flash**, Xiaomi shipped **MiMo-V2.5-Pro**, MiniMax open-sourced **M2.7**, Google released **Gemma 4**, Alibaba released **Qwen 3.6**, and Ant Group's inclusionAI released **Ring-2.6-1T**."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Best Open-Source LLM 2026: Llama 4, DeepSeek V4, Qwen, Kimi",
    "One row per model. The \"Neutral score\" column is the Artificial Analysis Intelligence Index (independent composite); a dash means AA has not published an index score for that exact variant. SWE-Bench figures are vendor-reported unless noted; we label them.",
    "| **1=** | **MiMo-V2.5-Pro** (Xiaomi) | 42B / 1T (MoE) | Apache-2.0 | **54** | Tied #1 by index; 1M context |"
   ]
  },
  {
   "url": "https://codersera.com/blog/best-open-source-llm-2026-llama-4-qwen-3-5-deepseek-v4-gemma-4-mistral/amp",
   "title": "Best Open-Source LLM 2026: Kimi, DeepSeek V4, Qwen, Llama 4",
   "published_at": "2026-05-03T08:27:05",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-04-01",
   "access_type": null,
   "license": null,
   "significance": "Cheapest frontier-class inference with an active parameter count of only 10B",
   "organization_evidence": [
    "# Best Open-Source LLM 2026: Llama 4, DeepSeek V4, Qwen, Kimi",
    "Between early April and mid-May 2026, Moonshot shipped **Kimi K2.6**, Z.ai shipped **GLM-5.1**, DeepSeek shipped **V4 Pro and V4 Flash**, Xiaomi shipped **MiMo-V2.5-Pro**, MiniMax open-sourced **M2.7**, Google released **Gemma 4**, Alibaba released **Qwen 3.6**, and Ant Group's inclusionAI released **Ring-2.6-1T**."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Best Open-Source LLM 2026: Llama 4, DeepSeek V4, Qwen, Kimi",
    "Between early April and mid-May 2026, Moonshot shipped **Kimi K2.6**, Z.ai shipped **GLM-5.1**, DeepSeek shipped **V4 Pro and V4 Flash**, Xiaomi shipped **MiMo-V2.5-Pro**, MiniMax open-sourced **M2.7**, Google released **Gemma 4**, Alibaba released **Qwen 3.6**, and Ant Group's inclusionAI released **Ring-2.6-1T**."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Best Open-Source LLM 2026: Llama 4, DeepSeek V4, Qwen, Kimi",
    "One row per model. The \"Neutral score\" column is the Artificial Analysis Intelligence Index (independent composite); a dash means AA has not published an index score for that exact variant. SWE-Bench figures are vendor-reported unless noted; we label them.",
    "| **5** | **MiniMax-M2.7** | 10B / 230B (MoE) | Apache-2.0 | 50 | Cheapest frontier-class inference |"
   ]
  },
  {
   "url": "https://codersera.com/blog/best-open-source-llm-2026-llama-4-qwen-3-5-deepseek-v4-gemma-4-mistral/amp",
   "title": "Best Open-Source LLM 2026: Kimi, DeepSeek V4, Qwen, Llama 4",
   "published_at": "2026-05-03T08:27:05",
   "organization": "Google",
   "model": null,
   "release_date": "2026-04-02",
   "access_type": null,
   "license": null,
   "significance": "On-device or laptop-class deployment with Apache-2.0 license",
   "organization_evidence": [
    "# Best Open-Source LLM 2026: Llama 4, DeepSeek V4, Qwen, Kimi",
    "Between early April and mid-May 2026, Moonshot shipped **Kimi K2.6**, Z.ai shipped **GLM-5.1**, DeepSeek shipped **V4 Pro and V4 Flash**, Xiaomi shipped **MiMo-V2.5-Pro**, MiniMax open-sourced **M2.7**, Google released **Gemma 4**, Alibaba released **Qwen 3.6**, and Ant Group's inclusionAI released **Ring-2.6-1T**."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Best Open-Source LLM 2026: Llama 4, DeepSeek V4, Qwen, Kimi",
    "**Gemma 4** (Google, April 2, 2026) is now **Apache-2.0** (a meaningful upgrade from prior Gemma terms) and ships in 2B/4B effective sizes, a 31B dense, and a 26B MoE (3.8B active)."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Best Open-Source LLM 2026: Llama 4, DeepSeek V4, Qwen, Kimi",
    "One row per model. The \"Neutral score\" column is the Artificial Analysis Intelligence Index (independent composite); a dash means AA has not published an index score for that exact variant. SWE-Bench figures are vendor-reported unless noted; we label them.",
    "| **9** | **Gemma 4** (Google) | 2B\u201331B (dense + 26B MoE) | Apache-2.0 | \u2014 | On-device / laptop-class |"
   ]
  },
  {
   "url": "https://codersera.com/blog/best-open-source-llm-2026-llama-4-qwen-3-5-deepseek-v4-gemma-4-mistral/amp",
   "title": "Best Open-Source LLM 2026: Kimi, DeepSeek V4, Qwen, Llama 4",
   "published_at": "2026-05-03T08:27:05",
   "organization": "Ant Group",
   "model": null,
   "release_date": "2026-05-08",
   "access_type": null,
   "license": null,
   "significance": "Adaptive-effort reasoning MoE under MIT",
   "organization_evidence": [
    "# Best Open-Source LLM 2026: Llama 4, DeepSeek V4, Qwen, Kimi",
    "Between early April and mid-May 2026, Moonshot shipped **Kimi K2.6**, Z.ai shipped **GLM-5.1**, DeepSeek shipped **V4 Pro and V4 Flash**, Xiaomi shipped **MiMo-V2.5-Pro**, MiniMax open-sourced **M2.7**, Google released **Gemma 4**, Alibaba released **Qwen 3.6**, and Ant Group's inclusionAI released **Ring-2.6-1T**."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Best Open-Source LLM 2026: Llama 4, DeepSeek V4, Qwen, Kimi",
    "**Ring-2.6-1T** (inclusionAI / Ant Group, ~May 8, 2026) is a ~1T-total / ~63B-active MoE under MIT with adaptive reasoning-effort modes."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Best Open-Source LLM 2026: Llama 4, DeepSeek V4, Qwen, Kimi",
    "One row per model. The \"Neutral score\" column is the Artificial Analysis Intelligence Index (independent composite); a dash means AA has not published an index score for that exact variant. SWE-Bench figures are vendor-reported unless noted; we label them.",
    "| **10** | **Ring-2.6-1T** (inclusionAI) | ~63B / 1T (MoE) | MIT | \u2014 (vendor numbers only) | Adaptive-effort reasoning (unverified) |"
   ]
  },
  {
   "url": "https://techcrunch.com/2026/01/27/chinas-moonshot-releases-a-new-open-source-model-kimi-k2-5-and-a-coding-agent",
   "title": "China\u2019s Moonshot releases a new open source model Kimi K2.5 and a coding agent",
   "published_at": "2026-01-27T14:11:10",
   "organization": "Moonshot AI",
   "model": "Kimi K2.5",
   "release_date": "2026-01-27",
   "access_type": "open source",
   "license": null,
   "significance": "trained on 15 trillion mixed visual and text tokens natively understanding text, image, and video, matching or beating proprietary peers like Gemini 3 Pro and GPT 5.2 in benchmarks",
   "organization_evidence": [
    "China\u2019s [Moonshot AI](https://www.moonshot.ai/), which is backed by the likes of Alibaba and HongShan (formerly Sequoia China), today released a new open source model, Kimi K2.5, which understands text, image, and video."
   ],
   "model_evidence": [
    "China\u2019s [Moonshot AI](https://www.moonshot.ai/), which is backed by the likes of Alibaba and HongShan (formerly Sequoia China), today released a new open source model, Kimi K2.5, which understands text, image, and video."
   ],
   "release_date_evidence": [
    "China\u2019s [Moonshot AI](https://www.moonshot.ai/), which is backed by the likes of Alibaba and HongShan (formerly Sequoia China), today released a new open source model, Kimi K2.5, which understands text, image, and video."
   ],
   "access_type_evidence": [
    "China\u2019s [Moonshot AI](https://www.moonshot.ai/), which is backed by the likes of Alibaba and HongShan (formerly Sequoia China), today released a new open source model, Kimi K2.5, which understands text, image, and video."
   ],
   "license_evidence": null,
   "significance_evidence": [
    "China\u2019s [Moonshot AI](https://www.moonshot.ai/), which is backed by the likes of Alibaba and HongShan (formerly Sequoia China), today released a new open source model, Kimi K2.5, which understands text, image, and video.",
    "The company said that the model was trained on 15 trillion mixed visual and text tokens, and that\u2019s why it is natively multimodal.",
    "In released benchmarks, the model matches the performance of the proprietary peers and even beats them in certain tasks."
   ]
  },
  {
   "url": "https://faq.com.tw/en/ai-ml/2026-05-11-china-open-weights-coding-blitz-en",
   "title": "China's 12-Day Coding Model Blitz: How Four AI Labs Reshaped Open Source",
   "published_at": "2026-05-11T00:00:00",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-04-07",
   "access_type": null,
   "license": null,
   "significance": "Scored 83 on the BenchLM composite, with particular strength in multi-step code reasoning and strong results in C++ and Rust systems programming tasks.",
   "organization_evidence": [
    "# China's 12-Day Coding Model Blitz: How Four AI Labs Reshaped Open Source",
    "In a window spanning just 12 days between April 7 and April 24, 2026, four Chinese AI laboratories released open-weight coding models that collectively rattled the assumptions of every frontier model developer in the world.",
    "The releases \u2014 Z.ai\u2019s GLM-5.1, MiniMax\u2019s M2.7, Moonshot AI\u2019s Kimi K2.6, and DeepSeek\u2019s V4 in two variants \u2014 represent the most concentrated burst of high-capability open-model releases the industry has ever seen from a single country, and they arrive at a moment when the competitive gap between Chinese and American AI is narrowing faster than most Western observers anticipated."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# China's 12-Day Coding Model Blitz: How Four AI Labs Reshaped Open Source",
    "In a window spanning just 12 days between April 7 and April 24, 2026, four Chinese AI laboratories released open-weight coding models that collectively rattled the assumptions of every frontier model developer in the world."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# China's 12-Day Coding Model Blitz: How Four AI Labs Reshaped Open Source",
    "## The Models: What They Actually Do",
    "**GLM-5.1** from Z.ai (which spun out from Tsinghua University\u2019s Institute for AI Industry Research) scored 83 on the BenchLM composite, with particular strength in multi-step code reasoning tasks."
   ]
  },
  {
   "url": "https://faq.com.tw/en/ai-ml/2026-05-11-china-open-weights-coding-blitz-en",
   "title": "China's 12-Day Coding Model Blitz: How Four AI Labs Reshaped Open Source",
   "published_at": "2026-05-11T00:00:00",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-04-07",
   "access_type": null,
   "license": null,
   "significance": "Placed at the top of the multilingual coding category, outperforming all non-Chinese open-weight alternatives on cross-lingual coding benchmarks.",
   "organization_evidence": [
    "# China's 12-Day Coding Model Blitz: How Four AI Labs Reshaped Open Source",
    "In a window spanning just 12 days between April 7 and April 24, 2026, four Chinese AI laboratories released open-weight coding models that collectively rattled the assumptions of every frontier model developer in the world.",
    "The releases \u2014 Z.ai\u2019s GLM-5.1, MiniMax\u2019s M2.7, Moonshot AI\u2019s Kimi K2.6, and DeepSeek\u2019s V4 in two variants \u2014 represent the most concentrated burst of high-capability open-model releases the industry has ever seen from a single country, and they arrive at a moment when the competitive gap between Chinese and American AI is narrowing faster than most Western observers anticipated."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# China's 12-Day Coding Model Blitz: How Four AI Labs Reshaped Open Source",
    "In a window spanning just 12 days between April 7 and April 24, 2026, four Chinese AI laboratories released open-weight coding models that collectively rattled the assumptions of every frontier model developer in the world."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# China's 12-Day Coding Model Blitz: How Four AI Labs Reshaped Open Source",
    "## The Models: What They Actually Do",
    "**MiniMax M2.7** rounded out the quartet with benchmark results placing it at the top of the multilingual coding category."
   ]
  },
  {
   "url": "https://faq.com.tw/en/ai-ml/2026-05-11-china-open-weights-coding-blitz-en",
   "title": "China's 12-Day Coding Model Blitz: How Four AI Labs Reshaped Open Source",
   "published_at": "2026-05-11T00:00:00",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-04-20",
   "access_type": null,
   "license": null,
   "significance": "First open-weight model in history to surpass GPT-5.4 on SWE-Bench Pro, scoring 58.6.",
   "organization_evidence": [
    "# China's 12-Day Coding Model Blitz: How Four AI Labs Reshaped Open Source",
    "In a window spanning just 12 days between April 7 and April 24, 2026, four Chinese AI laboratories released open-weight coding models that collectively rattled the assumptions of every frontier model developer in the world.",
    "The releases \u2014 Z.ai\u2019s GLM-5.1, MiniMax\u2019s M2.7, Moonshot AI\u2019s Kimi K2.6, and DeepSeek\u2019s V4 in two variants \u2014 represent the most concentrated burst of high-capability open-model releases the industry has ever seen from a single country, and they arrive at a moment when the competitive gap between Chinese and American AI is narrowing faster than most Western observers anticipated."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# China's 12-Day Coding Model Blitz: How Four AI Labs Reshaped Open Source",
    "## The Models: What They Actually Do",
    "**Kimi K2.6** (released April 20) made perhaps the most dramatic debut of the four."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# China's 12-Day Coding Model Blitz: How Four AI Labs Reshaped Open Source",
    "## The Models: What They Actually Do",
    "It became the first open-weight model in history to surpass GPT-5.4 (xhigh) on SWE-Bench Pro \u2014 the benchmark now widely considered the most rigorous test of real-world software engineering capability, evaluating models on their ability to resolve actual GitHub issues in large production codebases."
   ]
  },
  {
   "url": "https://faq.com.tw/en/ai-ml/2026-05-11-china-open-weights-coding-blitz-en",
   "title": "China's 12-Day Coding Model Blitz: How Four AI Labs Reshaped Open Source",
   "published_at": "2026-05-11T00:00:00",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-07",
   "access_type": null,
   "license": null,
   "significance": "Pro version scored 87 on BenchLM\u2019s composite leaderboard, the highest recorded for any Chinese model to date.",
   "organization_evidence": [
    "# China's 12-Day Coding Model Blitz: How Four AI Labs Reshaped Open Source",
    "In a window spanning just 12 days between April 7 and April 24, 2026, four Chinese AI laboratories released open-weight coding models that collectively rattled the assumptions of every frontier model developer in the world.",
    "The releases \u2014 Z.ai\u2019s GLM-5.1, MiniMax\u2019s M2.7, Moonshot AI\u2019s Kimi K2.6, and DeepSeek\u2019s V4 in two variants \u2014 represent the most concentrated burst of high-capability open-model releases the industry has ever seen from a single country, and they arrive at a moment when the competitive gap between Chinese and American AI is narrowing faster than most Western observers anticipated."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# China's 12-Day Coding Model Blitz: How Four AI Labs Reshaped Open Source",
    "In a window spanning just 12 days between April 7 and April 24, 2026, four Chinese AI laboratories released open-weight coding models that collectively rattled the assumptions of every frontier model developer in the world."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# China's 12-Day Coding Model Blitz: How Four AI Labs Reshaped Open Source",
    "## The Models: What They Actually Do",
    "The Pro version scored 87 on BenchLM\u2019s composite leaderboard, the highest recorded for any Chinese model to date."
   ]
  },
  {
   "url": "https://edgen.tech/news/post/zai-confirms-ox-alpha-as-glm-model-with-usage-2x-deepseek",
   "title": "Z.AI confirms Ox Alpha as GLM model with usage 2x DeepSeek",
   "published_at": "2026-08-26T12:16:39",
   "organization": "Z.AI",
   "model": null,
   "release_date": "2026-08-26",
   "access_type": null,
   "license": null,
   "significance": "Ox Alpha's usage more than doubled DeepSeek's on OpenRouter's leaderboard, making it the biggest launch in the marketplace's history, and passes 63 percent on DeepSWE.",
   "organization_evidence": [
    "Z.AI confirms Ox Alpha as GLM model with usage 2x DeepSeek",
    "The anonymous AI model that topped OpenRouter's usage charts within days is a new GLM-family release from China's Z.AI, which confirmed the launch and will open-source its weights."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The mystery AI model that swept to the top of OpenRouter's usage charts is a new GLM-family release from China's Z.AI, which confirmed the launch Wednesday and said it will release the model's weights tonight."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Z.AI confirms Ox Alpha as GLM model with usage 2x DeepSeek",
    "Ox Alpha's usage more than doubled DeepSeek's on OpenRouter's leaderboard, making it the biggest launch in the marketplace's history, according to the company."
   ]
  },
  {
   "url": "https://en.wikipedia.org/wiki/GLM-4.5?no_redirect=true",
   "title": "GLM (AI)",
   "published_at": "2026-08-26T23:05:36",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-08-26",
   "access_type": null,
   "license": null,
   "significance": "major model family generation released with weights and served using AI chips manufactured in China",
   "organization_evidence": [
    "GLM, short for General Language Model, is a series of open weight large language models developed by Chinese software company Z.ai."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Z.ai anonymously previewed a model called Ox Alpha on OpenRouter, which was released with weights on 26 August as GLM-5.3-Flash."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Z.ai anonymously previewed a model called Ox Alpha on OpenRouter, which was released with weights on 26 August as GLM-5.3-Flash."
   ]
  },
  {
   "url": "https://bivashvlog.com/thinking-machines-inkling-ai-model-specs-benchmarks",
   "title": "Thinking Machines Inkling AI Model: Specs, Benchmarks and Release Details",
   "published_at": "2026-07-16T01:47:10",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "a 975-billion-parameter open-weights system that reasons across text, images, and audio",
   "organization_evidence": [
    "Thinking Machines Inkling AI Model: Specs, Benchmarks and Release Details",
    "Thinking Machines Lab released its first public AI model on July 15, 2026, called Inkling, a 975-billion-parameter open-weights system that reasons across text, images, and audio, ships under an Apache 2.0 license, and is available for fine-tuning through the company\u2019s Tinker platform starting today."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Thinking Machines Inkling AI Model: Specs, Benchmarks and Release Details",
    "Thinking Machines Lab released its first public AI model on July 15, 2026, called Inkling, a 975-billion-parameter open-weights system that reasons across text, images, and audio, ships under an Apache 2.0 license, and is available for fine-tuning through the company\u2019s Tinker platform starting today."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Thinking Machines Inkling AI Model: Specs, Benchmarks and Release Details",
    "Thinking Machines Lab released its first public AI model on July 15, 2026, called Inkling, a 975-billion-parameter open-weights system that reasons across text, images, and audio, ships under an Apache 2.0 license, and is available for fine-tuning through the company\u2019s Tinker platform starting today."
   ]
  },
  {
   "url": "https://insideai.news/news/generative-ai/thinking-machines-launches-inkling-an-open-weight-ai-model-from-the-us/4503",
   "title": "Thinking Machines Launches Inkling: An Open-Weight AI Model from the US",
   "published_at": "2026-07-16T15:55:10",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-16",
   "access_type": null,
   "license": null,
   "significance": "A 975-billion-parameter open-weight model featuring a controllable thinking feature and 1-million token context window",
   "organization_evidence": [
    "# Thinking Machines Launches Inkling: An Open-Weight AI Model from the US",
    "Thinking Machines Lab, founded by ex-OpenAI CTO Mira Murati, has released Inkling, its first open-weight AI model."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "By Shamil Khan Published on: July 16, 2026",
    "**July 16, 2026, (Inside AI) \u2014** Thinking Machines Lab, the startup founded by former OpenAI CTO Mira Murati, launched its first in-house AI model on Wednesday."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Thinking Machines Launches Inkling: An Open-Weight AI Model from the US",
    "Thinking Machines Lab, founded by ex-OpenAI CTO Mira Murati, has released Inkling, its first open-weight AI model. With 975 billion parameters and a unique controllable thinking feature, it targets enterprises wanting to fine-tune their own AI rather than rely on closed, one-size-fits-all systems.",
    "Called **Inkling**, it is an open-weight, 975-billion-parameter large language model that lets users dial thinking effort up or down."
   ]
  },
  {
   "url": "https://pomegra.io/news/chinas-moonshot-ai-kimi-k3-challenges-western-llm-dominance",
   "title": "China's Moonshot AI: Kimi K3 Challenges Western LLM Dominance",
   "published_at": "2026-07-21T17:32:11",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-16",
   "access_type": null,
   "license": null,
   "significance": "the world's largest open-weight model at 2.8 trillion parameters, benchmarking within striking distance of OpenAI's and Anthropic's best proprietary systems",
   "organization_evidence": [
    "# China's Moonshot AI: Kimi K3 Challenges Western LLM Dominance",
    "**Moonshot AI** released **Kimi K3** on July 16, 2026, unveiling a 2.8-trillion-parameter large language model that the company describes as the largest open-weight model in existence."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# China's Moonshot AI: Kimi K3 Challenges Western LLM Dominance",
    "**Moonshot AI** released **Kimi K3** on July 16, 2026, unveiling a 2.8-trillion-parameter large language model that the company describes as the largest open-weight model in existence."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# China's Moonshot AI: Kimi K3 Challenges Western LLM Dominance",
    "China's Moonshot AI has released Kimi K3, the world's largest open-weight model at 2.8 trillion parameters, benchmarking within striking distance of OpenAI's and Anthropic's best proprietary systems while pricing at a fraction of their cost."
   ]
  },
  {
   "url": "https://huggingface.co/blog/ResterChed/deepseek-v4-flash-official-release",
   "title": "DeepSeek V4 Flash Is Now Official: What Changed in the 0731 Build",
   "published_at": "2026-07-31T14:09:29",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-07-31",
   "access_type": null,
   "license": null,
   "significance": "a 13B-active model that outruns the family's flagship preview on every published agentic benchmark, ships with native Codex support, and costs $0.14/$0.28 per million tokens with a 98% cache-hit discount",
   "organization_evidence": [
    "Today DeepSeek promoted V4-Flash from preview to official release.",
    "The DeepSeek V4 Flash official release is the most consequential open-model update of the summer for anyone building agents: a 13B-active model that outruns the family's flagship preview on every published agentic benchmark, ships with native Codex support, and costs $0.14/$0.28 per million tokens with a 98% cache-hit discount \u2014 with the same-day MIT-licensed weights to verify every claim."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The DeepSeek-V4-Flash-0731 checkpoint shipped on July 31, 2026 \u2014 the same 284B-parameter CSA+HCA backbone, re-post-trained for agents, now outscoring the V4-Pro preview on every agentic benchmark DeepSeek publishes.",
    "July 31, 2026 \u00b7 ~12 min read"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": null
  },
  {
   "url": "https://artificialanalysis.ai/models/qwen3-5-0-8b",
   "title": "Qwen3.5 0.8B - Intelligence, Performance & Price Analysis",
   "published_at": null,
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-03-02",
   "access_type": null,
   "license": null,
   "significance": "above average in intelligence and well priced when comparing to other open weight models of similar size, supporting text, image, and video input, outputs text, and having a 262k tokens context window",
   "organization_evidence": [
    "Qwen3.5 0.8B (Reasoning) Intelligence, Performance & Price Analysis",
    "Qwen3.5 0.8B (Reasoning) was created by Alibaba."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Qwen3.5 0.8B (Reasoning) Intelligence, Performance & Price Analysis",
    "Qwen3.5 0.8B (Reasoning) was released on March 2, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Qwen3.5 0.8B (Reasoning) Intelligence, Performance & Price Analysis",
    "Qwen3.5 0.8B (Reasoning) is above average in intelligence and well priced when comparing to other open weight models of similar size."
   ]
  },
  {
   "url": "https://edgen.tech/ja/news/post/zai-confirms-ox-alpha-as-glm-model-with-usage-2x-deepseek",
   "title": "Z.AI confirms Ox Alpha as GLM model with usage 2x DeepSeek",
   "published_at": "2026-08-26T12:16:39",
   "organization": "Z.AI",
   "model": null,
   "release_date": "2026-08-26",
   "access_type": null,
   "license": null,
   "significance": "Ox Alpha's OpenRouter usage more than doubled DeepSeek's, the marketplace's biggest launch.",
   "organization_evidence": [
    "Z.AI confirms Ox Alpha as GLM model with usage 2x DeepSeek",
    "The mystery AI model that swept to the top of OpenRouter's usage charts is a new GLM-family release from China's Z.AI, which confirmed the launch Wednesday and said it will release the model's weights tonight."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "[![Alex Nguyen](https://static.edwealth.ai/authors/alex-nguyen.png?w=32)\n## Alex Nguyen](https://www.edgen.tech/ja/author/news/alex-nguyen)\u00b7Aug 26 2026, 05:16",
    "Z.AI confirms Ox Alpha as GLM model with usage 2x DeepSeek",
    "The mystery AI model that swept to the top of OpenRouter's usage charts is a new GLM-family release from China's Z.AI, which confirmed the launch Wednesday and said it will release the model's weights tonight."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Z.AI confirms Ox Alpha as GLM model with usage 2x DeepSeek",
    "- Ox Alpha's OpenRouter usage more than doubled DeepSeek's, the marketplace's biggest launch."
   ]
  },
  {
   "url": "https://santageai.com/news/2026/08/10/meta-muse-glimmer-open-weight-agentic-model",
   "title": "Meta Returns to Open Weights With Muse Glimmer",
   "published_at": "2026-08-10T00:00:00",
   "organization": "Meta",
   "model": null,
   "release_date": "2026-08-10",
   "access_type": null,
   "license": null,
   "significance": "a 30-billion-parameter dense multimodal model tuned for local agents, tool use, coding, and LLM-as-judge evaluation",
   "organization_evidence": [
    "Meta puts a 30B agentic model back in the open",
    "Meta has returned to open weights."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "On August 10 the company's Superintelligence Labs published Muse Glimmer, a 30-billion-parameter dense multimodal model, and released the weights under an Apache 2.0 license, one of the most permissive terms available."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Meta Superintelligence Labs released Muse Glimmer, a 30-billion-parameter dense multimodal model, under a permissive Apache 2.0 license on Hugging Face with day-one support in Ollama, LM Studio, and vLLM.",
    "Compressed to 4-bit, the model needs under 20GB of memory and runs on a single consumer GPU such as a 24GB card or a MacBook, and is tuned for local agents, tool use, coding, and LLM-as-judge evaluation."
   ]
  },
  {
   "url": "https://techcrunch.com/2026/07/24/as-us-weighs-response-to-chinese-ai-industry-urges-against-broad-open-weight-restrictions",
   "title": "As US weighs response to Chinese AI, industry urges against broad open-weight restrictions",
   "published_at": "2026-07-24T15:51:49",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-18",
   "access_type": null,
   "license": null,
   "significance": "very impressive",
   "organization_evidence": null,
   "model_evidence": null,
   "release_date_evidence": [
    "The White House has even accused Moonshot AI of distilling Anthropic\u2019s Fable model to train its recently released and, by all measures, very impressive, [Kimi K3 model.](https://techcrunch.com/2026/07/18/kimi-threat-or-menace/)"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The White House has even accused Moonshot AI of distilling Anthropic\u2019s Fable model to train its recently released and, by all measures, very impressive, [Kimi K3 model.](https://techcrunch.com/2026/07/18/kimi-threat-or-menace/)"
   ]
  },
  {
   "url": "https://techcrunch.com/2026/07/24/as-us-weighs-response-to-chinese-ai-industry-urges-against-broad-open-weight-restrictions",
   "title": "As US weighs response to Chinese AI, industry urges against broad open-weight restrictions",
   "published_at": "2026-07-24T15:51:49",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": null,
   "organization_evidence": [
    "He pointed out that [Thinking Machines Lab\u2019s new open model, Inkling](https://techcrunch.com/2026/07/15/thinking-machines-amps-up-its-bet-against-one-size-fits-all-ai-with-its-first-open-model-inkling/), was trained with the help of Moonshot\u2019s Kimi 2.5."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "He pointed out that [Thinking Machines Lab\u2019s new open model, Inkling](https://techcrunch.com/2026/07/15/thinking-machines-amps-up-its-bet-against-one-size-fits-all-ai-with-its-first-open-model-inkling/), was trained with the help of Moonshot\u2019s Kimi 2.5."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": null
  },
  {
   "url": "https://boondockrebel.org/kimi-k3-china-ai",
   "title": "China Built the Largest Open AI Model in History. It Just Took the Top Coding Spot From Everyone.",
   "published_at": "2026-08-22T12:00:00",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-16",
   "access_type": null,
   "license": null,
   "significance": "2.8 trillion parameters in a sparse mixture-of-experts architecture with a one-million-token context window that took the top position on LMArena's Frontend Code Arena leaderboard",
   "organization_evidence": [
    "China Built the Largest Open AI Model in History. It Just Took the Top Coding Spot From Everyone.",
    "On July 16, 2026, Moonshot AI released Kimi K3 \u2014 2.8 trillion parameters, open weights, and a score that landed it at number one on the global coding benchmark."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "China Built the Largest Open AI Model in History. It Just Took the Top Coding Spot From Everyone.",
    "On July 16, 2026, Moonshot AI released Kimi K3 \u2014 2.8 trillion parameters, open weights, and a score that landed it at number one on the global coding benchmark."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "China Built the Largest Open AI Model in History. It Just Took the Top Coding Spot From Everyone.",
    "It is the largest open-weight AI model ever released to the public \u2014 2.8 trillion parameters in a sparse mixture-of-experts architecture, with a one-million-token context window that allows it to process and reason across an amount of text that would fill multiple books at once."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/north-mini-code",
   "title": "North Mini Code - Intelligence, Performance & Price Analysis | Artificial Analysis",
   "published_at": null,
   "organization": "Cohere",
   "model": null,
   "release_date": "2026-06-09",
   "access_type": null,
   "license": null,
   "significance": "A reasoning model with 30 billion parameters (3 billion active), 260k-256k tokens context window, and scoring 20 on the Artificial Analysis Intelligence Index.",
   "organization_evidence": [
    "North Mini Code Intelligence, Performance & Price Analysis",
    "North Mini Code was created by Cohere."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "North Mini Code Intelligence, Performance & Price Analysis",
    "North Mini Code was released on June 9, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "North Mini Code Intelligence, Performance & Price Analysis",
    "North Mini Code is amongst the leading models in intelligence and well priced when comparing to other open weight models of similar size.",
    "North Mini Code scores 20 on the Artificial Analysis Intelligence Index, placing it well above average among comparable models (median: 9).",
    "North Mini Code has a context window of 260k tokens. This determines how much text and conversation history the model can process in a single request.",
    "Yes, North Mini Code is a reasoning model. It uses extended thinking or chain-of-thought reasoning to work through complex problems before providing an answer.",
    "North Mini Code has 30 billion parameters (3 billion active)."
   ]
  },
  {
   "url": "https://geeksalad.org/exploring-china-s-rapid-ai-innovation-four-frontier-class-open-models",
   "title": "Exploring China\u2019s Rapid AI Innovation: Four Frontier-Class Open Models - Geek Salad",
   "published_at": "2026-07-15T06:14:13",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor",
   "organization_evidence": [
    "Starting with DeepSeek V4 on April 24, then MiniMax M3 on June 1, followed by Kimi K2.7-Code and GLM-5.2 in mid-June, Chinese labs have demonstrated an accelerated cadence of releasing high-capability models.",
    "Exploring China\u2019s Rapid AI Innovation: Four Frontier-Class Open Models"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor",
    "Starting with DeepSeek V4 on April 24, then MiniMax M3 on June 1, followed by Kimi K2.7-Code and GLM-5.2 in mid-June, Chinese labs have demonstrated an accelerated cadence of releasing high-capability models.",
    "Exploring China\u2019s Rapid AI Innovation: Four Frontier-Class Open Models",
    "### The production line \u2014 spring 2026"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor",
    "Exploring China\u2019s Rapid AI Innovation: Four Frontier-Class Open Models",
    "### The production line \u2014 spring 2026"
   ]
  },
  {
   "url": "https://geeksalad.org/exploring-china-s-rapid-ai-innovation-four-frontier-class-open-models",
   "title": "Exploring China\u2019s Rapid AI Innovation: Four Frontier-Class Open Models - Geek Salad",
   "published_at": "2026-07-15T06:14:13",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "cheap 1M-token context, native multimodal, modified-MIT",
   "organization_evidence": [
    "Starting with DeepSeek V4 on April 24, then MiniMax M3 on June 1, followed by Kimi K2.7-Code and GLM-5.2 in mid-June, Chinese labs have demonstrated an accelerated cadence of releasing high-capability models.",
    "Exploring China\u2019s Rapid AI Innovation: Four Frontier-Class Open Models"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "**MiniMax M3**cheap 1M-token context, native multimodal, modified-MIT",
    "Starting with DeepSeek V4 on April 24, then MiniMax M3 on June 1, followed by Kimi K2.7-Code and GLM-5.2 in mid-June, Chinese labs have demonstrated an accelerated cadence of releasing high-capability models.",
    "Exploring China\u2019s Rapid AI Innovation: Four Frontier-Class Open Models",
    "### The production line \u2014 spring 2026"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "**MiniMax M3**cheap 1M-token context, native multimodal, modified-MIT",
    "Exploring China\u2019s Rapid AI Innovation: Four Frontier-Class Open Models",
    "### The production line \u2014 spring 2026"
   ]
  },
  {
   "url": "https://geeksalad.org/exploring-china-s-rapid-ai-innovation-four-frontier-class-open-models",
   "title": "Exploring China\u2019s Rapid AI Innovation: Four Frontier-Class Open Models - Geek Salad",
   "published_at": "2026-07-15T06:14:13",
   "organization": "Moonshot",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "agent-run specialist, ~30% fewer thinking tokens than K2.6",
   "organization_evidence": [
    "Chinese labs\u2014DeepSeek, Z.ai, Moonshot, Alibaba\u2014each have distinct strategic focuses, from cost efficiency and long-horizon stability to broad deployment options.",
    "Exploring China\u2019s Rapid AI Innovation: Four Frontier-Class Open Models"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "**Kimi K2.7-Code (Moonshot)**agent-run specialist, ~30% fewer thinking tokens than K2.6",
    "Starting with DeepSeek V4 on April 24, then MiniMax M3 on June 1, followed by Kimi K2.7-Code and GLM-5.2 in mid-June, Chinese labs have demonstrated an accelerated cadence of releasing high-capability models.",
    "Exploring China\u2019s Rapid AI Innovation: Four Frontier-Class Open Models",
    "### The production line \u2014 spring 2026"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "**Kimi K2.7-Code (Moonshot)**agent-run specialist, ~30% fewer thinking tokens than K2.6",
    "Exploring China\u2019s Rapid AI Innovation: Four Frontier-Class Open Models",
    "### The production line \u2014 spring 2026"
   ]
  },
  {
   "url": "https://geeksalad.org/exploring-china-s-rapid-ai-innovation-four-frontier-class-open-models",
   "title": "Exploring China\u2019s Rapid AI Innovation: Four Frontier-Class Open Models - Geek Salad",
   "published_at": "2026-07-15T06:14:13",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-06-16",
   "access_type": null,
   "license": null,
   "significance": "753B MoE, MIT, top open-weight on Artificial Analysis index",
   "organization_evidence": [
    "Chinese labs\u2014DeepSeek, Z.ai, Moonshot, Alibaba\u2014each have distinct strategic focuses, from cost efficiency and long-horizon stability to broad deployment options.",
    "Exploring China\u2019s Rapid AI Innovation: Four Frontier-Class Open Models"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "**GLM-5.2 (Z.ai)**753B MoE, MIT, top open-weight on Artificial Analysis index",
    "Starting with DeepSeek V4 on April 24, then MiniMax M3 on June 1, followed by Kimi K2.7-Code and GLM-5.2 in mid-June, Chinese labs have demonstrated an accelerated cadence of releasing high-capability models.",
    "Exploring China\u2019s Rapid AI Innovation: Four Frontier-Class Open Models",
    "### The production line \u2014 spring 2026"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "**GLM-5.2 (Z.ai)**753B MoE, MIT, top open-weight on Artificial Analysis index",
    "Exploring China\u2019s Rapid AI Innovation: Four Frontier-Class Open Models",
    "### The production line \u2014 spring 2026"
   ]
  },
  {
   "url": "https://changegamer.ai/resources/open-weight-models-for-agents",
   "title": "Open-Weight Models for Agents",
   "published_at": null,
   "organization": "Mistral",
   "model": null,
   "release_date": "2026-03",
   "access_type": null,
   "license": null,
   "significance": "Small 4 unifies Magistral reasoning + Pixtral vision + Devstral coding in one model",
   "organization_evidence": [
    "# Open-Weight Models for Agents",
    "Cross-vendor comparison table of major open-weight LLM families \u2014 license, tool-calling support, context window, and agent-builder notes \u2014 as of July 2026.",
    "## Comparison table",
    "Comparison table",
    "| Family | Latest open-weight release | License | Native tool/function-calling | Notable for agents |",
    "| **Mistral** | Mistral Small 4 (119B total / 6.5B active MoE, Mar 2026); Mistral Medium 3.5 (128B dense, Apr 2026); Mistral Large 3 (675B total / 41B active MoE, Dec 2025) | Apache 2.0 (Small 4, Large 3); Modified MIT with revenue cap (Medium 3.5) | Yes \u2014 function calling and structured output supported across all three; Small 4 also unifies reasoning and vision | Small 4 (256K ctx) unifies Magistral reasoning + Pixtral vision + Devstral coding in one model. Medium 3.5 (256K ctx) is a frontier-class open coding/agentic model. Large 3 (256K ctx) is the largest open-weight Mistral. Mistral Small 3.2 deprecated April 30, 2026. |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Open-Weight Models for Agents",
    "Cross-vendor comparison table of major open-weight LLM families \u2014 license, tool-calling support, context window, and agent-builder notes \u2014 as of July 2026.",
    "## Comparison table",
    "Comparison table",
    "| Family | Latest open-weight release | License | Native tool/function-calling | Notable for agents |",
    "| **Mistral** | Mistral Small 4 (119B total / 6.5B active MoE, Mar 2026); Mistral Medium 3.5 (128B dense, Apr 2026); Mistral Large 3 (675B total / 41B active MoE, Dec 2025) | Apache 2.0 (Small 4, Large 3); Modified MIT with revenue cap (Medium 3.5) | Yes \u2014 function calling and structured output supported across all three; Small 4 also unifies reasoning and vision | Small 4 (256K ctx) unifies Magistral reasoning + Pixtral vision + Devstral coding in one model. Medium 3.5 (256K ctx) is a frontier-class open coding/agentic model. Large 3 (256K ctx) is the largest open-weight Mistral. Mistral Small 3.2 deprecated April 30, 2026. |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Open-Weight Models for Agents",
    "Cross-vendor comparison table of major open-weight LLM families \u2014 license, tool-calling support, context window, and agent-builder notes \u2014 as of July 2026.",
    "## Comparison table",
    "Comparison table",
    "| Family | Latest open-weight release | License | Native tool/function-calling | Notable for agents |",
    "| **Mistral** | Mistral Small 4 (119B total / 6.5B active MoE, Mar 2026); Mistral Medium 3.5 (128B dense, Apr 2026); Mistral Large 3 (675B total / 41B active MoE, Dec 2025) | Apache 2.0 (Small 4, Large 3); Modified MIT with revenue cap (Medium 3.5) | Yes \u2014 function calling and structured output supported across all three; Small 4 also unifies reasoning and vision | Small 4 (256K ctx) unifies Magistral reasoning + Pixtral vision + Devstral coding in one model. Medium 3.5 (256K ctx) is a frontier-class open coding/agentic model. Large 3 (256K ctx) is the largest open-weight Mistral. Mistral Small 3.2 deprecated April 30, 2026. |"
   ]
  },
  {
   "url": "https://changegamer.ai/resources/open-weight-models-for-agents",
   "title": "Open-Weight Models for Agents",
   "published_at": null,
   "organization": "Mistral",
   "model": "Mistral Medium 3.5",
   "release_date": "2026-04",
   "access_type": "open-weight",
   "license": "Modified MIT with revenue cap",
   "significance": "A frontier-class open coding/agentic model",
   "organization_evidence": [
    "# Open-Weight Models for Agents",
    "Cross-vendor comparison table of major open-weight LLM families \u2014 license, tool-calling support, context window, and agent-builder notes \u2014 as of July 2026.",
    "## Comparison table",
    "Comparison table",
    "| Family | Latest open-weight release | License | Native tool/function-calling | Notable for agents |",
    "| **Mistral** | Mistral Small 4 (119B total / 6.5B active MoE, Mar 2026); Mistral Medium 3.5 (128B dense, Apr 2026); Mistral Large 3 (675B total / 41B active MoE, Dec 2025) | Apache 2.0 (Small 4, Large 3); Modified MIT with revenue cap (Medium 3.5) | Yes \u2014 function calling and structured output supported across all three; Small 4 also unifies reasoning and vision | Small 4 (256K ctx) unifies Magistral reasoning + Pixtral vision + Devstral coding in one model. Medium 3.5 (256K ctx) is a frontier-class open coding/agentic model. Large 3 (256K ctx) is the largest open-weight Mistral. Mistral Small 3.2 deprecated April 30, 2026. |"
   ],
   "model_evidence": [
    "# Open-Weight Models for Agents",
    "Cross-vendor comparison table of major open-weight LLM families \u2014 license, tool-calling support, context window, and agent-builder notes \u2014 as of July 2026.",
    "## Comparison table",
    "Comparison table",
    "| Family | Latest open-weight release | License | Native tool/function-calling | Notable for agents |",
    "| **Mistral** | Mistral Small 4 (119B total / 6.5B active MoE, Mar 2026); Mistral Medium 3.5 (128B dense, Apr 2026); Mistral Large 3 (675B total / 41B active MoE, Dec 2025) | Apache 2.0 (Small 4, Large 3); Modified MIT with revenue cap (Medium 3.5) | Yes \u2014 function calling and structured output supported across all three; Small 4 also unifies reasoning and vision | Small 4 (256K ctx) unifies Magistral reasoning + Pixtral vision + Devstral coding in one model. Medium 3.5 (256K ctx) is a frontier-class open coding/agentic model. Large 3 (256K ctx) is the largest open-weight Mistral. Mistral Small 3.2 deprecated April 30, 2026. |"
   ],
   "release_date_evidence": [
    "# Open-Weight Models for Agents",
    "Cross-vendor comparison table of major open-weight LLM families \u2014 license, tool-calling support, context window, and agent-builder notes \u2014 as of July 2026.",
    "## Comparison table",
    "Comparison table",
    "| Family | Latest open-weight release | License | Native tool/function-calling | Notable for agents |",
    "| **Mistral** | Mistral Small 4 (119B total / 6.5B active MoE, Mar 2026); Mistral Medium 3.5 (128B dense, Apr 2026); Mistral Large 3 (675B total / 41B active MoE, Dec 2025) | Apache 2.0 (Small 4, Large 3); Modified MIT with revenue cap (Medium 3.5) | Yes \u2014 function calling and structured output supported across all three; Small 4 also unifies reasoning and vision | Small 4 (256K ctx) unifies Magistral reasoning + Pixtral vision + Devstral coding in one model. Medium 3.5 (256K ctx) is a frontier-class open coding/agentic model. Large 3 (256K ctx) is the largest open-weight Mistral. Mistral Small 3.2 deprecated April 30, 2026. |"
   ],
   "access_type_evidence": [
    "# Open-Weight Models for Agents",
    "Cross-vendor comparison table of major open-weight LLM families \u2014 license, tool-calling support, context window, and agent-builder notes \u2014 as of July 2026.",
    "## Comparison table",
    "Comparison table",
    "| Family | Latest open-weight release | License | Native tool/function-calling | Notable for agents |",
    "| **Mistral** | Mistral Small 4 (119B total / 6.5B active MoE, Mar 2026); Mistral Medium 3.5 (128B dense, Apr 2026); Mistral Large 3 (675B total / 41B active MoE, Dec 2025) | Apache 2.0 (Small 4, Large 3); Modified MIT with revenue cap (Medium 3.5) | Yes \u2014 function calling and structured output supported across all three; Small 4 also unifies reasoning and vision | Small 4 (256K ctx) unifies Magistral reasoning + Pixtral vision + Devstral coding in one model. Medium 3.5 (256K ctx) is a frontier-class open coding/agentic model. Large 3 (256K ctx) is the largest open-weight Mistral. Mistral Small 3.2 deprecated April 30, 2026. |"
   ],
   "license_evidence": [
    "# Open-Weight Models for Agents",
    "Cross-vendor comparison table of major open-weight LLM families \u2014 license, tool-calling support, context window, and agent-builder notes \u2014 as of July 2026.",
    "## Comparison table",
    "Comparison table",
    "| Family | Latest open-weight release | License | Native tool/function-calling | Notable for agents |",
    "| **Mistral** | Mistral Small 4 (119B total / 6.5B active MoE, Mar 2026); Mistral Medium 3.5 (128B dense, Apr 2026); Mistral Large 3 (675B total / 41B active MoE, Dec 2025) | Apache 2.0 (Small 4, Large 3); Modified MIT with revenue cap (Medium 3.5) | Yes \u2014 function calling and structured output supported across all three; Small 4 also unifies reasoning and vision | Small 4 (256K ctx) unifies Magistral reasoning + Pixtral vision + Devstral coding in one model. Medium 3.5 (256K ctx) is a frontier-class open coding/agentic model. Large 3 (256K ctx) is the largest open-weight Mistral. Mistral Small 3.2 deprecated April 30, 2026. |"
   ],
   "significance_evidence": [
    "# Open-Weight Models for Agents",
    "Cross-vendor comparison table of major open-weight LLM families \u2014 license, tool-calling support, context window, and agent-builder notes \u2014 as of July 2026.",
    "## Comparison table",
    "Comparison table",
    "| Family | Latest open-weight release | License | Native tool/function-calling | Notable for agents |",
    "| **Mistral** | Mistral Small 4 (119B total / 6.5B active MoE, Mar 2026); Mistral Medium 3.5 (128B dense, Apr 2026); Mistral Large 3 (675B total / 41B active MoE, Dec 2025) | Apache 2.0 (Small 4, Large 3); Modified MIT with revenue cap (Medium 3.5) | Yes \u2014 function calling and structured output supported across all three; Small 4 also unifies reasoning and vision | Small 4 (256K ctx) unifies Magistral reasoning + Pixtral vision + Devstral coding in one model. Medium 3.5 (256K ctx) is a frontier-class open coding/agentic model. Large 3 (256K ctx) is the largest open-weight Mistral. Mistral Small 3.2 deprecated April 30, 2026. |"
   ]
  },
  {
   "url": "https://changegamer.ai/resources/open-weight-models-for-agents",
   "title": "Open-Weight Models for Agents",
   "published_at": null,
   "organization": "Alibaba Qwen",
   "model": "Qwen 3.6",
   "release_date": "2026-04",
   "access_type": "open-weight",
   "license": "Apache 2.0",
   "significance": "Dense and MoE variants from 0.6B to 235B with up to 262K context",
   "organization_evidence": [
    "# Open-Weight Models for Agents",
    "Cross-vendor comparison table of major open-weight LLM families \u2014 license, tool-calling support, context window, and agent-builder notes \u2014 as of July 2026.",
    "## Comparison table",
    "Comparison table",
    "| Family | Latest open-weight release | License | Native tool/function-calling | Notable for agents |",
    "| **Alibaba Qwen** | Qwen 3.6-27B / 3.6-35B-A3B (Apr 2026); Qwen3 base series (Apr 2025) | Apache 2.0 | Yes \u2014 native tool-calling and MCP support via Qwen-Agent; all sizes | Dense and MoE variants from 0.6B to 235B. Up to 262K context (extensible to 1M via YaRN). Hybrid thinking/non-thinking mode. Qwen 3.7 is closed-weight API-only as of Jun 2026. |"
   ],
   "model_evidence": [
    "# Open-Weight Models for Agents",
    "Cross-vendor comparison table of major open-weight LLM families \u2014 license, tool-calling support, context window, and agent-builder notes \u2014 as of July 2026.",
    "## Comparison table",
    "Comparison table",
    "| Family | Latest open-weight release | License | Native tool/function-calling | Notable for agents |",
    "| **Alibaba Qwen** | Qwen 3.6-27B / 3.6-35B-A3B (Apr 2026); Qwen3 base series (Apr 2025) | Apache 2.0 | Yes \u2014 native tool-calling and MCP support via Qwen-Agent; all sizes | Dense and MoE variants from 0.6B to 235B. Up to 262K context (extensible to 1M via YaRN). Hybrid thinking/non-thinking mode. Qwen 3.7 is closed-weight API-only as of Jun 2026. |"
   ],
   "release_date_evidence": [
    "# Open-Weight Models for Agents",
    "Cross-vendor comparison table of major open-weight LLM families \u2014 license, tool-calling support, context window, and agent-builder notes \u2014 as of July 2026.",
    "## Comparison table",
    "Comparison table",
    "| Family | Latest open-weight release | License | Native tool/function-calling | Notable for agents |",
    "| **Alibaba Qwen** | Qwen 3.6-27B / 3.6-35B-A3B (Apr 2026); Qwen3 base series (Apr 2025) | Apache 2.0 | Yes \u2014 native tool-calling and MCP support via Qwen-Agent; all sizes | Dense and MoE variants from 0.6B to 235B. Up to 262K context (extensible to 1M via YaRN). Hybrid thinking/non-thinking mode. Qwen 3.7 is closed-weight API-only as of Jun 2026. |"
   ],
   "access_type_evidence": [
    "# Open-Weight Models for Agents",
    "Cross-vendor comparison table of major open-weight LLM families \u2014 license, tool-calling support, context window, and agent-builder notes \u2014 as of July 2026.",
    "## Comparison table",
    "Comparison table",
    "| Family | Latest open-weight release | License | Native tool/function-calling | Notable for agents |",
    "| **Alibaba Qwen** | Qwen 3.6-27B / 3.6-35B-A3B (Apr 2026); Qwen3 base series (Apr 2025) | Apache 2.0 | Yes \u2014 native tool-calling and MCP support via Qwen-Agent; all sizes | Dense and MoE variants from 0.6B to 235B. Up to 262K context (extensible to 1M via YaRN). Hybrid thinking/non-thinking mode. Qwen 3.7 is closed-weight API-only as of Jun 2026. |"
   ],
   "license_evidence": [
    "# Open-Weight Models for Agents",
    "Cross-vendor comparison table of major open-weight LLM families \u2014 license, tool-calling support, context window, and agent-builder notes \u2014 as of July 2026.",
    "## Comparison table",
    "Comparison table",
    "| Family | Latest open-weight release | License | Native tool/function-calling | Notable for agents |",
    "| **Alibaba Qwen** | Qwen 3.6-27B / 3.6-35B-A3B (Apr 2026); Qwen3 base series (Apr 2025) | Apache 2.0 | Yes \u2014 native tool-calling and MCP support via Qwen-Agent; all sizes | Dense and MoE variants from 0.6B to 235B. Up to 262K context (extensible to 1M via YaRN). Hybrid thinking/non-thinking mode. Qwen 3.7 is closed-weight API-only as of Jun 2026. |"
   ],
   "significance_evidence": [
    "# Open-Weight Models for Agents",
    "Cross-vendor comparison table of major open-weight LLM families \u2014 license, tool-calling support, context window, and agent-builder notes \u2014 as of July 2026.",
    "## Comparison table",
    "Comparison table",
    "| Family | Latest open-weight release | License | Native tool/function-calling | Notable for agents |",
    "| **Alibaba Qwen** | Qwen 3.6-27B / 3.6-35B-A3B (Apr 2026); Qwen3 base series (Apr 2025) | Apache 2.0 | Yes \u2014 native tool-calling and MCP support via Qwen-Agent; all sizes | Dense and MoE variants from 0.6B to 235B. Up to 262K context (extensible to 1M via YaRN). Hybrid thinking/non-thinking mode. Qwen 3.7 is closed-weight API-only as of Jun 2026. |"
   ]
  },
  {
   "url": "https://changegamer.ai/resources/open-weight-models-for-agents",
   "title": "Open-Weight Models for Agents",
   "published_at": null,
   "organization": "DeepSeek",
   "model": "DeepSeek V4",
   "release_date": "2026-04",
   "access_type": "open-weight",
   "license": "MIT",
   "significance": "1M context window with frontier-class agentic coding and fast/cheap inference variants",
   "organization_evidence": [
    "# Open-Weight Models for Agents",
    "Cross-vendor comparison table of major open-weight LLM families \u2014 license, tool-calling support, context window, and agent-builder notes \u2014 as of July 2026.",
    "## Comparison table",
    "Comparison table",
    "| Family | Latest open-weight release | License | Native tool/function-calling | Notable for agents |",
    "| **DeepSeek** | DeepSeek V4-Pro (1.6T total / 49B active MoE) and V4-Flash (284B total / 13B active), both Apr 2026 preview | MIT | Yes \u2014 V4 natively supports function calling, JSON output, tool calls, and thinking / non-thinking modes | 1M context window (default across both V4 variants). V4-Pro: frontier-class agentic coding. V4-Flash: fast/cheap inference. Weights on Hugging Face (deepseek-ai). V4 labeled preview; stable release expected later 2026. |"
   ],
   "model_evidence": [
    "# Open-Weight Models for Agents",
    "Cross-vendor comparison table of major open-weight LLM families \u2014 license, tool-calling support, context window, and agent-builder notes \u2014 as of July 2026.",
    "## Comparison table",
    "Comparison table",
    "| Family | Latest open-weight release | License | Native tool/function-calling | Notable for agents |",
    "| **DeepSeek** | DeepSeek V4-Pro (1.6T total / 49B active MoE) and V4-Flash (284B total / 13B active), both Apr 2026 preview | MIT | Yes \u2014 V4 natively supports function calling, JSON output, tool calls, and thinking / non-thinking modes | 1M context window (default across both V4 variants). V4-Pro: frontier-class agentic coding. V4-Flash: fast/cheap inference. Weights on Hugging Face (deepseek-ai). V4 labeled preview; stable release expected later 2026. |"
   ],
   "release_date_evidence": [
    "# Open-Weight Models for Agents",
    "Cross-vendor comparison table of major open-weight LLM families \u2014 license, tool-calling support, context window, and agent-builder notes \u2014 as of July 2026.",
    "## Comparison table",
    "Comparison table",
    "| Family | Latest open-weight release | License | Native tool/function-calling | Notable for agents |",
    "| **DeepSeek** | DeepSeek V4-Pro (1.6T total / 49B active MoE) and V4-Flash (284B total / 13B active), both Apr 2026 preview | MIT | Yes \u2014 V4 natively supports function calling, JSON output, tool calls, and thinking / non-thinking modes | 1M context window (default across both V4 variants). V4-Pro: frontier-class agentic coding. V4-Flash: fast/cheap inference. Weights on Hugging Face (deepseek-ai). V4 labeled preview; stable release expected later 2026. |"
   ],
   "access_type_evidence": [
    "# Open-Weight Models for Agents",
    "Cross-vendor comparison table of major open-weight LLM families \u2014 license, tool-calling support, context window, and agent-builder notes \u2014 as of July 2026.",
    "## Comparison table",
    "Comparison table",
    "| Family | Latest open-weight release | License | Native tool/function-calling | Notable for agents |",
    "| **DeepSeek** | DeepSeek V4-Pro (1.6T total / 49B active MoE) and V4-Flash (284B total / 13B active), both Apr 2026 preview | MIT | Yes \u2014 V4 natively supports function calling, JSON output, tool calls, and thinking / non-thinking modes | 1M context window (default across both V4 variants). V4-Pro: frontier-class agentic coding. V4-Flash: fast/cheap inference. Weights on Hugging Face (deepseek-ai). V4 labeled preview; stable release expected later 2026. |"
   ],
   "license_evidence": [
    "# Open-Weight Models for Agents",
    "Cross-vendor comparison table of major open-weight LLM families \u2014 license, tool-calling support, context window, and agent-builder notes \u2014 as of July 2026.",
    "## Comparison table",
    "Comparison table",
    "| Family | Latest open-weight release | License | Native tool/function-calling | Notable for agents |",
    "| **DeepSeek** | DeepSeek V4-Pro (1.6T total / 49B active MoE) and V4-Flash (284B total / 13B active), both Apr 2026 preview | MIT | Yes \u2014 V4 natively supports function calling, JSON output, tool calls, and thinking / non-thinking modes | 1M context window (default across both V4 variants). V4-Pro: frontier-class agentic coding. V4-Flash: fast/cheap inference. Weights on Hugging Face (deepseek-ai). V4 labeled preview; stable release expected later 2026. |"
   ],
   "significance_evidence": [
    "# Open-Weight Models for Agents",
    "Cross-vendor comparison table of major open-weight LLM families \u2014 license, tool-calling support, context window, and agent-builder notes \u2014 as of July 2026.",
    "## Comparison table",
    "Comparison table",
    "| Family | Latest open-weight release | License | Native tool/function-calling | Notable for agents |",
    "| **DeepSeek** | DeepSeek V4-Pro (1.6T total / 49B active MoE) and V4-Flash (284B total / 13B active), both Apr 2026 preview | MIT | Yes \u2014 V4 natively supports function calling, JSON output, tool calls, and thinking / non-thinking modes | 1M context window (default across both V4 variants). V4-Pro: frontier-class agentic coding. V4-Flash: fast/cheap inference. Weights on Hugging Face (deepseek-ai). V4 labeled preview; stable release expected later 2026. |"
   ]
  },
  {
   "url": "https://changegamer.ai/resources/open-weight-models-for-agents",
   "title": "Open-Weight Models for Agents",
   "published_at": null,
   "organization": "Google",
   "model": "Google Gemma 4",
   "release_date": "2026-03",
   "access_type": "open-weight",
   "license": "Apache 2.0",
   "significance": "First Gemma release under true Apache 2.0 with multimodal capabilities and native audio + video support",
   "organization_evidence": [
    "# Open-Weight Models for Agents",
    "Cross-vendor comparison table of major open-weight LLM families \u2014 license, tool-calling support, context window, and agent-builder notes \u2014 as of July 2026.",
    "## Comparison table",
    "Comparison table",
    "| Family | Latest open-weight release | License | Native tool/function-calling | Notable for agents |",
    "| **Google Gemma 4** | Gemma 4: E2B, E4B, 26B-A4B, 31B (Mar\u2013Apr 2026); Gemma 4 12B Unified (Jun 2026, encoder-free, native audio) | Apache 2.0 (first Gemma release under true Apache 2.0) | Yes \u2014 native function-calling built into Gemma 4; FunctionGemma 270M for edge/on-device | 128K context (E2B/E4B); 256K context (12B+, 26B, 31B). Gemma 4 12B Unified (Jun 3 2026) adds native audio + video via encoder-free architecture; runs on 16 GB RAM. Multimodal across the family. |"
   ],
   "model_evidence": [
    "# Open-Weight Models for Agents",
    "Cross-vendor comparison table of major open-weight LLM families \u2014 license, tool-calling support, context window, and agent-builder notes \u2014 as of July 2026.",
    "## Comparison table",
    "Comparison table",
    "| Family | Latest open-weight release | License | Native tool/function-calling | Notable for agents |",
    "| **Google Gemma 4** | Gemma 4: E2B, E4B, 26B-A4B, 31B (Mar\u2013Apr 2026); Gemma 4 12B Unified (Jun 2026, encoder-free, native audio) | Apache 2.0 (first Gemma release under true Apache 2.0) | Yes \u2014 native function-calling built into Gemma 4; FunctionGemma 270M for edge/on-device | 128K context (E2B/E4B); 256K context (12B+, 26B, 31B). Gemma 4 12B Unified (Jun 3 2026) adds native audio + video via encoder-free architecture; runs on 16 GB RAM. Multimodal across the family. |"
   ],
   "release_date_evidence": [
    "# Open-Weight Models for Agents",
    "Cross-vendor comparison table of major open-weight LLM families \u2014 license, tool-calling support, context window, and agent-builder notes \u2014 as of July 2026.",
    "## Comparison table",
    "Comparison table",
    "| Family | Latest open-weight release | License | Native tool/function-calling | Notable for agents |",
    "| **Google Gemma 4** | Gemma 4: E2B, E4B, 26B-A4B, 31B (Mar\u2013Apr 2026); Gemma 4 12B Unified (Jun 2026, encoder-free, native audio) | Apache 2.0 (first Gemma release under true Apache 2.0) | Yes \u2014 native function-calling built into Gemma 4; FunctionGemma 270M for edge/on-device | 128K context (E2B/E4B); 256K context (12B+, 26B, 31B). Gemma 4 12B Unified (Jun 3 2026) adds native audio + video via encoder-free architecture; runs on 16 GB RAM. Multimodal across the family. |"
   ],
   "access_type_evidence": [
    "# Open-Weight Models for Agents",
    "Cross-vendor comparison table of major open-weight LLM families \u2014 license, tool-calling support, context window, and agent-builder notes \u2014 as of July 2026.",
    "## Comparison table",
    "Comparison table",
    "| Family | Latest open-weight release | License | Native tool/function-calling | Notable for agents |",
    "| **Google Gemma 4** | Gemma 4: E2B, E4B, 26B-A4B, 31B (Mar\u2013Apr 2026); Gemma 4 12B Unified (Jun 2026, encoder-free, native audio) | Apache 2.0 (first Gemma release under true Apache 2.0) | Yes \u2014 native function-calling built into Gemma 4; FunctionGemma 270M for edge/on-device | 128K context (E2B/E4B); 256K context (12B+, 26B, 31B). Gemma 4 12B Unified (Jun 3 2026) adds native audio + video via encoder-free architecture; runs on 16 GB RAM. Multimodal across the family. |"
   ],
   "license_evidence": [
    "# Open-Weight Models for Agents",
    "Cross-vendor comparison table of major open-weight LLM families \u2014 license, tool-calling support, context window, and agent-builder notes \u2014 as of July 2026.",
    "## Comparison table",
    "Comparison table",
    "| Family | Latest open-weight release | License | Native tool/function-calling | Notable for agents |",
    "| **Google Gemma 4** | Gemma 4: E2B, E4B, 26B-A4B, 31B (Mar\u2013Apr 2026); Gemma 4 12B Unified (Jun 2026, encoder-free, native audio) | Apache 2.0 (first Gemma release under true Apache 2.0) | Yes \u2014 native function-calling built into Gemma 4; FunctionGemma 270M for edge/on-device | 128K context (E2B/E4B); 256K context (12B+, 26B, 31B). Gemma 4 12B Unified (Jun 3 2026) adds native audio + video via encoder-free architecture; runs on 16 GB RAM. Multimodal across the family. |"
   ],
   "significance_evidence": [
    "# Open-Weight Models for Agents",
    "Cross-vendor comparison table of major open-weight LLM families \u2014 license, tool-calling support, context window, and agent-builder notes \u2014 as of July 2026.",
    "## Comparison table",
    "Comparison table",
    "| Family | Latest open-weight release | License | Native tool/function-calling | Notable for agents |",
    "| **Google Gemma 4** | Gemma 4: E2B, E4B, 26B-A4B, 31B (Mar\u2013Apr 2026); Gemma 4 12B Unified (Jun 2026, encoder-free, native audio) | Apache 2.0 (first Gemma release under true Apache 2.0) | Yes \u2014 native function-calling built into Gemma 4; FunctionGemma 270M for edge/on-device | 128K context (E2B/E4B); 256K context (12B+, 26B, 31B). Gemma 4 12B Unified (Jun 3 2026) adds native audio + video via encoder-free architecture; runs on 16 GB RAM. Multimodal across the family. |"
   ]
  },
  {
   "url": "https://changegamer.ai/resources/open-weight-models-for-agents",
   "title": "Open-Weight Models for Agents",
   "published_at": null,
   "organization": "Microsoft",
   "model": "Microsoft Phi-4-reasoning-vision-15B",
   "release_date": "2026-03",
   "access_type": "open-weight",
   "license": "MIT",
   "significance": "Efficiency-first model with strong reasoning per parameter, selective thinking mode, and high-res vision",
   "organization_evidence": [
    "# Open-Weight Models for Agents",
    "Cross-vendor comparison table of major open-weight LLM families \u2014 license, tool-calling support, context window, and agent-builder notes \u2014 as of July 2026.",
    "## Comparison table",
    "Comparison table",
    "| Family | Latest open-weight release | License | Native tool/function-calling | Notable for agents |",
    "| **Microsoft Phi-4** | Phi-4-reasoning-vision-15B (15B, Mar 2026); Phi-4-reasoning (14B, May 2025); Phi-4-mini (3.8B) | MIT | Yes \u2014 Phi-4-mini has built-in function calling; the Phi-4 line supports tool use; Phi-4-reasoning for chain-of-thought agentic tasks | Efficiency-first: strong reasoning per parameter. MIT license. Phi-4-reasoning-vision-15B adds selective thinking mode + high-res vision. Phi-4-multimodal adds audio+vision. |"
   ],
   "model_evidence": [
    "# Open-Weight Models for Agents",
    "Cross-vendor comparison table of major open-weight LLM families \u2014 license, tool-calling support, context window, and agent-builder notes \u2014 as of July 2026.",
    "## Comparison table",
    "Comparison table",
    "| Family | Latest open-weight release | License | Native tool/function-calling | Notable for agents |",
    "| **Microsoft Phi-4** | Phi-4-reasoning-vision-15B (15B, Mar 2026); Phi-4-reasoning (14B, May 2025); Phi-4-mini (3.8B) | MIT | Yes \u2014 Phi-4-mini has built-in function calling; the Phi-4 line supports tool use; Phi-4-reasoning for chain-of-thought agentic tasks | Efficiency-first: strong reasoning per parameter. MIT license. Phi-4-reasoning-vision-15B adds selective thinking mode + high-res vision. Phi-4-multimodal adds audio+vision. |"
   ],
   "release_date_evidence": [
    "# Open-Weight Models for Agents",
    "Cross-vendor comparison table of major open-weight LLM families \u2014 license, tool-calling support, context window, and agent-builder notes \u2014 as of July 2026.",
    "## Comparison table",
    "Comparison table",
    "| Family | Latest open-weight release | License | Native tool/function-calling | Notable for agents |",
    "| **Microsoft Phi-4** | Phi-4-reasoning-vision-15B (15B, Mar 2026); Phi-4-reasoning (14B, May 2025); Phi-4-mini (3.8B) | MIT | Yes \u2014 Phi-4-mini has built-in function calling; the Phi-4 line supports tool use; Phi-4-reasoning for chain-of-thought agentic tasks | Efficiency-first: strong reasoning per parameter. MIT license. Phi-4-reasoning-vision-15B adds selective thinking mode + high-res vision. Phi-4-multimodal adds audio+vision. |"
   ],
   "access_type_evidence": [
    "# Open-Weight Models for Agents",
    "Cross-vendor comparison table of major open-weight LLM families \u2014 license, tool-calling support, context window, and agent-builder notes \u2014 as of July 2026.",
    "## Comparison table",
    "Comparison table",
    "| Family | Latest open-weight release | License | Native tool/function-calling | Notable for agents |",
    "| **Microsoft Phi-4** | Phi-4-reasoning-vision-15B (15B, Mar 2026); Phi-4-reasoning (14B, May 2025); Phi-4-mini (3.8B) | MIT | Yes \u2014 Phi-4-mini has built-in function calling; the Phi-4 line supports tool use; Phi-4-reasoning for chain-of-thought agentic tasks | Efficiency-first: strong reasoning per parameter. MIT license. Phi-4-reasoning-vision-15B adds selective thinking mode + high-res vision. Phi-4-multimodal adds audio+vision. |"
   ],
   "license_evidence": [
    "# Open-Weight Models for Agents",
    "Cross-vendor comparison table of major open-weight LLM families \u2014 license, tool-calling support, context window, and agent-builder notes \u2014 as of July 2026.",
    "## Comparison table",
    "Comparison table",
    "| Family | Latest open-weight release | License | Native tool/function-calling | Notable for agents |",
    "| **Microsoft Phi-4** | Phi-4-reasoning-vision-15B (15B, Mar 2026); Phi-4-reasoning (14B, May 2025); Phi-4-mini (3.8B) | MIT | Yes \u2014 Phi-4-mini has built-in function calling; the Phi-4 line supports tool use; Phi-4-reasoning for chain-of-thought agentic tasks | Efficiency-first: strong reasoning per parameter. MIT license. Phi-4-reasoning-vision-15B adds selective thinking mode + high-res vision. Phi-4-multimodal adds audio+vision. |"
   ],
   "significance_evidence": [
    "# Open-Weight Models for Agents",
    "Cross-vendor comparison table of major open-weight LLM families \u2014 license, tool-calling support, context window, and agent-builder notes \u2014 as of July 2026.",
    "## Comparison table",
    "Comparison table",
    "| Family | Latest open-weight release | License | Native tool/function-calling | Notable for agents |",
    "| **Microsoft Phi-4** | Phi-4-reasoning-vision-15B (15B, Mar 2026); Phi-4-reasoning (14B, May 2025); Phi-4-mini (3.8B) | MIT | Yes \u2014 Phi-4-mini has built-in function calling; the Phi-4 line supports tool use; Phi-4-reasoning for chain-of-thought agentic tasks | Efficiency-first: strong reasoning per parameter. MIT license. Phi-4-reasoning-vision-15B adds selective thinking mode + high-res vision. Phi-4-multimodal adds audio+vision. |"
   ]
  },
  {
   "url": "https://changegamer.ai/resources/open-weight-models-for-agents",
   "title": "Open-Weight Models for Agents",
   "published_at": null,
   "organization": "IBM",
   "model": "IBM Granite 4.1",
   "release_date": "2026-04",
   "access_type": "open-weight",
   "license": "Apache 2.0",
   "significance": "Enterprise-focused with a 512K context window and hybrid Mamba-Transformer architecture",
   "organization_evidence": [
    "# Open-Weight Models for Agents",
    "Cross-vendor comparison table of major open-weight LLM families \u2014 license, tool-calling support, context window, and agent-builder notes \u2014 as of July 2026.",
    "## Comparison table",
    "Comparison table",
    "| Family | Latest open-weight release | License | Native tool/function-calling | Notable for agents |",
    "| **IBM Granite 4.1** | Granite 4.1 (3B, 8B, 30B, Apr 2026) | Apache 2.0 | Yes \u2014 tool calling follows OpenAI function definition schema; benchmarked on Berkeley BFCL | 512K context window. Enterprise-focused; ISO 42001 certified (Granite 4.0 line). 30B uses hybrid Mamba-Transformer architecture for long-context efficiency; 3B/8B are dense. Sizes 3B\u201330B. |"
   ],
   "model_evidence": [
    "# Open-Weight Models for Agents",
    "Cross-vendor comparison table of major open-weight LLM families \u2014 license, tool-calling support, context window, and agent-builder notes \u2014 as of July 2026.",
    "## Comparison table",
    "Comparison table",
    "| Family | Latest open-weight release | License | Native tool/function-calling | Notable for agents |",
    "| **IBM Granite 4.1** | Granite 4.1 (3B, 8B, 30B, Apr 2026) | Apache 2.0 | Yes \u2014 tool calling follows OpenAI function definition schema; benchmarked on Berkeley BFCL | 512K context window. Enterprise-focused; ISO 42001 certified (Granite 4.0 line). 30B uses hybrid Mamba-Transformer architecture for long-context efficiency; 3B/8B are dense. Sizes 3B\u201330B. |"
   ],
   "release_date_evidence": [
    "# Open-Weight Models for Agents",
    "Cross-vendor comparison table of major open-weight LLM families \u2014 license, tool-calling support, context window, and agent-builder notes \u2014 as of July 2026.",
    "## Comparison table",
    "Comparison table",
    "| Family | Latest open-weight release | License | Native tool/function-calling | Notable for agents |",
    "| **IBM Granite 4.1** | Granite 4.1 (3B, 8B, 30B, Apr 2026) | Apache 2.0 | Yes \u2014 tool calling follows OpenAI function definition schema; benchmarked on Berkeley BFCL | 512K context window. Enterprise-focused; ISO 42001 certified (Granite 4.0 line). 30B uses hybrid Mamba-Transformer architecture for long-context efficiency; 3B/8B are dense. Sizes 3B\u201330B. |"
   ],
   "access_type_evidence": [
    "# Open-Weight Models for Agents",
    "Cross-vendor comparison table of major open-weight LLM families \u2014 license, tool-calling support, context window, and agent-builder notes \u2014 as of July 2026.",
    "## Comparison table",
    "Comparison table",
    "| Family | Latest open-weight release | License | Native tool/function-calling | Notable for agents |",
    "| **IBM Granite 4.1** | Granite 4.1 (3B, 8B, 30B, Apr 2026) | Apache 2.0 | Yes \u2014 tool calling follows OpenAI function definition schema; benchmarked on Berkeley BFCL | 512K context window. Enterprise-focused; ISO 42001 certified (Granite 4.0 line). 30B uses hybrid Mamba-Transformer architecture for long-context efficiency; 3B/8B are dense. Sizes 3B\u201330B. |"
   ],
   "license_evidence": [
    "# Open-Weight Models for Agents",
    "Cross-vendor comparison table of major open-weight LLM families \u2014 license, tool-calling support, context window, and agent-builder notes \u2014 as of July 2026.",
    "## Comparison table",
    "Comparison table",
    "| Family | Latest open-weight release | License | Native tool/function-calling | Notable for agents |",
    "| **IBM Granite 4.1** | Granite 4.1 (3B, 8B, 30B, Apr 2026) | Apache 2.0 | Yes \u2014 tool calling follows OpenAI function definition schema; benchmarked on Berkeley BFCL | 512K context window. Enterprise-focused; ISO 42001 certified (Granite 4.0 line). 30B uses hybrid Mamba-Transformer architecture for long-context efficiency; 3B/8B are dense. Sizes 3B\u201330B. |"
   ],
   "significance_evidence": [
    "# Open-Weight Models for Agents",
    "Cross-vendor comparison table of major open-weight LLM families \u2014 license, tool-calling support, context window, and agent-builder notes \u2014 as of July 2026.",
    "## Comparison table",
    "Comparison table",
    "| Family | Latest open-weight release | License | Native tool/function-calling | Notable for agents |",
    "| **IBM Granite 4.1** | Granite 4.1 (3B, 8B, 30B, Apr 2026) | Apache 2.0 | Yes \u2014 tool calling follows OpenAI function definition schema; benchmarked on Berkeley BFCL | 512K context window. Enterprise-focused; ISO 42001 certified (Granite 4.0 line). 30B uses hybrid Mamba-Transformer architecture for long-context efficiency; 3B/8B are dense. Sizes 3B\u201330B. |"
   ]
  },
  {
   "url": "https://changegamer.ai/resources/open-weight-models-for-agents",
   "title": "Open-Weight Models for Agents",
   "published_at": null,
   "organization": "Z.ai",
   "model": "Z.ai GLM-5.2",
   "release_date": "2026-06",
   "access_type": "open-weight",
   "license": "MIT",
   "significance": "Frontier-class long-horizon coding/agentic model with a 1M context window",
   "organization_evidence": [
    "# Open-Weight Models for Agents",
    "Cross-vendor comparison table of major open-weight LLM families \u2014 license, tool-calling support, context window, and agent-builder notes \u2014 as of July 2026.",
    "## Comparison table",
    "Comparison table",
    "| Family | Latest open-weight release | License | Native tool/function-calling | Notable for agents |",
    "| **Z.ai GLM-5.2** | GLM-5.2 (744B total / ~40B active MoE, Jun 2026) | MIT | Yes \u2014 native tool-calling, function calling, structured output, and MCP support | 1M context (128K max output). Frontier-class long-horizon coding/agentic model (SWE-bench Pro 62.1); the highest-ranked open-weight model across several long-horizon coding benchmarks. Self-hosting needs a multi-GPU node (\u22488\u00d7H100/H200-class at FP8); official FP8 weights at huggingface.co/zai-org/GLM-5.2-FP8. |"
   ],
   "model_evidence": [
    "# Open-Weight Models for Agents",
    "Cross-vendor comparison table of major open-weight LLM families \u2014 license, tool-calling support, context window, and agent-builder notes \u2014 as of July 2026.",
    "## Comparison table",
    "Comparison table",
    "| Family | Latest open-weight release | License | Native tool/function-calling | Notable for agents |",
    "| **Z.ai GLM-5.2** | GLM-5.2 (744B total / ~40B active MoE, Jun 2026) | MIT | Yes \u2014 native tool-calling, function calling, structured output, and MCP support | 1M context (128K max output). Frontier-class long-horizon coding/agentic model (SWE-bench Pro 62.1); the highest-ranked open-weight model across several long-horizon coding benchmarks. Self-hosting needs a multi-GPU node (\u22488\u00d7H100/H200-class at FP8); official FP8 weights at huggingface.co/zai-org/GLM-5.2-FP8. |"
   ],
   "release_date_evidence": [
    "# Open-Weight Models for Agents",
    "Cross-vendor comparison table of major open-weight LLM families \u2014 license, tool-calling support, context window, and agent-builder notes \u2014 as of July 2026.",
    "## Comparison table",
    "Comparison table",
    "| Family | Latest open-weight release | License | Native tool/function-calling | Notable for agents |",
    "| **Z.ai GLM-5.2** | GLM-5.2 (744B total / ~40B active MoE, Jun 2026) | MIT | Yes \u2014 native tool-calling, function calling, structured output, and MCP support | 1M context (128K max output). Frontier-class long-horizon coding/agentic model (SWE-bench Pro 62.1); the highest-ranked open-weight model across several long-horizon coding benchmarks. Self-hosting needs a multi-GPU node (\u22488\u00d7H100/H200-class at FP8); official FP8 weights at huggingface.co/zai-org/GLM-5.2-FP8. |"
   ],
   "access_type_evidence": [
    "# Open-Weight Models for Agents",
    "Cross-vendor comparison table of major open-weight LLM families \u2014 license, tool-calling support, context window, and agent-builder notes \u2014 as of July 2026.",
    "## Comparison table",
    "Comparison table",
    "| Family | Latest open-weight release | License | Native tool/function-calling | Notable for agents |",
    "| **Z.ai GLM-5.2** | GLM-5.2 (744B total / ~40B active MoE, Jun 2026) | MIT | Yes \u2014 native tool-calling, function calling, structured output, and MCP support | 1M context (128K max output). Frontier-class long-horizon coding/agentic model (SWE-bench Pro 62.1); the highest-ranked open-weight model across several long-horizon coding benchmarks. Self-hosting needs a multi-GPU node (\u22488\u00d7H100/H200-class at FP8); official FP8 weights at huggingface.co/zai-org/GLM-5.2-FP8. |"
   ],
   "license_evidence": [
    "# Open-Weight Models for Agents",
    "Cross-vendor comparison table of major open-weight LLM families \u2014 license, tool-calling support, context window, and agent-builder notes \u2014 as of July 2026.",
    "## Comparison table",
    "Comparison table",
    "| Family | Latest open-weight release | License | Native tool/function-calling | Notable for agents |",
    "| **Z.ai GLM-5.2** | GLM-5.2 (744B total / ~40B active MoE, Jun 2026) | MIT | Yes \u2014 native tool-calling, function calling, structured output, and MCP support | 1M context (128K max output). Frontier-class long-horizon coding/agentic model (SWE-bench Pro 62.1); the highest-ranked open-weight model across several long-horizon coding benchmarks. Self-hosting needs a multi-GPU node (\u22488\u00d7H100/H200-class at FP8); official FP8 weights at huggingface.co/zai-org/GLM-5.2-FP8. |"
   ],
   "significance_evidence": [
    "# Open-Weight Models for Agents",
    "Cross-vendor comparison table of major open-weight LLM families \u2014 license, tool-calling support, context window, and agent-builder notes \u2014 as of July 2026.",
    "## Comparison table",
    "Comparison table",
    "| Family | Latest open-weight release | License | Native tool/function-calling | Notable for agents |",
    "| **Z.ai GLM-5.2** | GLM-5.2 (744B total / ~40B active MoE, Jun 2026) | MIT | Yes \u2014 native tool-calling, function calling, structured output, and MCP support | 1M context (128K max output). Frontier-class long-horizon coding/agentic model (SWE-bench Pro 62.1); the highest-ranked open-weight model across several long-horizon coding benchmarks. Self-hosting needs a multi-GPU node (\u22488\u00d7H100/H200-class at FP8); official FP8 weights at huggingface.co/zai-org/GLM-5.2-FP8. |"
   ]
  },
  {
   "url": "https://changegamer.ai/resources/open-weight-models-for-agents",
   "title": "Open-Weight Models for Agents",
   "published_at": null,
   "organization": "Moonshot AI",
   "model": "Kimi K2.7 Code",
   "release_date": "2026-06-12",
   "access_type": "open-weight",
   "license": "Modified MIT",
   "significance": "Purpose-built for long-horizon software-engineering workflows with ~30% fewer reasoning tokens than K2.6",
   "organization_evidence": [
    "# Open-Weight Models for Agents",
    "Cross-vendor comparison table of major open-weight LLM families \u2014 license, tool-calling support, context window, and agent-builder notes \u2014 as of July 2026.",
    "## Comparison table",
    "Comparison table",
    "| Family | Latest open-weight release | License | Native tool/function-calling | Notable for agents |",
    "| **Moonshot AI Kimi K2.7 Code** | Kimi K2.7 Code (~1T total / ~32B active MoE, Jun 12 2026) | Modified MIT (commercial use and internal deployment permitted; attribution required above a revenue/MAU threshold, consistent with the established Kimi K2 license family) | Yes \u2014 native agentic tool-calling; purpose-built for long-horizon software-engineering workflows | 256K context. The coding/agentic-specialized variant of the Kimi K2 line; Moonshot reports ~30% fewer reasoning tokens than K2.6 at higher coding-benchmark scores. Weights on Hugging Face (moonshotai/Kimi-K2.7-Code). |"
   ],
   "model_evidence": [
    "# Open-Weight Models for Agents",
    "Cross-vendor comparison table of major open-weight LLM families \u2014 license, tool-calling support, context window, and agent-builder notes \u2014 as of July 2026.",
    "## Comparison table",
    "Comparison table",
    "| Family | Latest open-weight release | License | Native tool/function-calling | Notable for agents |",
    "| **Moonshot AI Kimi K2.7 Code** | Kimi K2.7 Code (~1T total / ~32B active MoE, Jun 12 2026) | Modified MIT (commercial use and internal deployment permitted; attribution required above a revenue/MAU threshold, consistent with the established Kimi K2 license family) | Yes \u2014 native agentic tool-calling; purpose-built for long-horizon software-engineering workflows | 256K context. The coding/agentic-specialized variant of the Kimi K2 line; Moonshot reports ~30% fewer reasoning tokens than K2.6 at higher coding-benchmark scores. Weights on Hugging Face (moonshotai/Kimi-K2.7-Code). |"
   ],
   "release_date_evidence": [
    "# Open-Weight Models for Agents",
    "Cross-vendor comparison table of major open-weight LLM families \u2014 license, tool-calling support, context window, and agent-builder notes \u2014 as of July 2026.",
    "## Comparison table",
    "Comparison table",
    "| Family | Latest open-weight release | License | Native tool/function-calling | Notable for agents |",
    "| **Moonshot AI Kimi K2.7 Code** | Kimi K2.7 Code (~1T total / ~32B active MoE, Jun 12 2026) | Modified MIT (commercial use and internal deployment permitted; attribution required above a revenue/MAU threshold, consistent with the established Kimi K2 license family) | Yes \u2014 native agentic tool-calling; purpose-built for long-horizon software-engineering workflows | 256K context. The coding/agentic-specialized variant of the Kimi K2 line; Moonshot reports ~30% fewer reasoning tokens than K2.6 at higher coding-benchmark scores. Weights on Hugging Face (moonshotai/Kimi-K2.7-Code). |"
   ],
   "access_type_evidence": [
    "# Open-Weight Models for Agents",
    "Cross-vendor comparison table of major open-weight LLM families \u2014 license, tool-calling support, context window, and agent-builder notes \u2014 as of July 2026.",
    "## Comparison table",
    "Comparison table",
    "| Family | Latest open-weight release | License | Native tool/function-calling | Notable for agents |",
    "| **Moonshot AI Kimi K2.7 Code** | Kimi K2.7 Code (~1T total / ~32B active MoE, Jun 12 2026) | Modified MIT (commercial use and internal deployment permitted; attribution required above a revenue/MAU threshold, consistent with the established Kimi K2 license family) | Yes \u2014 native agentic tool-calling; purpose-built for long-horizon software-engineering workflows | 256K context. The coding/agentic-specialized variant of the Kimi K2 line; Moonshot reports ~30% fewer reasoning tokens than K2.6 at higher coding-benchmark scores. Weights on Hugging Face (moonshotai/Kimi-K2.7-Code). |"
   ],
   "license_evidence": [
    "# Open-Weight Models for Agents",
    "Cross-vendor comparison table of major open-weight LLM families \u2014 license, tool-calling support, context window, and agent-builder notes \u2014 as of July 2026.",
    "## Comparison table",
    "Comparison table",
    "| Family | Latest open-weight release | License | Native tool/function-calling | Notable for agents |",
    "| **Moonshot AI Kimi K2.7 Code** | Kimi K2.7 Code (~1T total / ~32B active MoE, Jun 12 2026) | Modified MIT (commercial use and internal deployment permitted; attribution required above a revenue/MAU threshold, consistent with the established Kimi K2 license family) | Yes \u2014 native agentic tool-calling; purpose-built for long-horizon software-engineering workflows | 256K context. The coding/agentic-specialized variant of the Kimi K2 line; Moonshot reports ~30% fewer reasoning tokens than K2.6 at higher coding-benchmark scores. Weights on Hugging Face (moonshotai/Kimi-K2.7-Code). |"
   ],
   "significance_evidence": [
    "# Open-Weight Models for Agents",
    "Cross-vendor comparison table of major open-weight LLM families \u2014 license, tool-calling support, context window, and agent-builder notes \u2014 as of July 2026.",
    "## Comparison table",
    "Comparison table",
    "| Family | Latest open-weight release | License | Native tool/function-calling | Notable for agents |",
    "| **Moonshot AI Kimi K2.7 Code** | Kimi K2.7 Code (~1T total / ~32B active MoE, Jun 12 2026) | Modified MIT (commercial use and internal deployment permitted; attribution required above a revenue/MAU threshold, consistent with the established Kimi K2 license family) | Yes \u2014 native agentic tool-calling; purpose-built for long-horizon software-engineering workflows | 256K context. The coding/agentic-specialized variant of the Kimi K2 line; Moonshot reports ~30% fewer reasoning tokens than K2.6 at higher coding-benchmark scores. Weights on Hugging Face (moonshotai/Kimi-K2.7-Code). |"
   ]
  },
  {
   "url": "https://changegamer.ai/resources/open-weight-models-for-agents",
   "title": "Open-Weight Models for Agents",
   "published_at": null,
   "organization": "NVIDIA",
   "model": "Nemotron 3 Ultra",
   "release_date": "2026-06-04",
   "access_type": "open",
   "license": "OpenMDW-1.1",
   "significance": "Tuned for long-running agent orchestration with a 1M context and hybrid Mamba-Transformer MoE architecture",
   "organization_evidence": [
    "# Open-Weight Models for Agents",
    "Cross-vendor comparison table of major open-weight LLM families \u2014 license, tool-calling support, context window, and agent-builder notes \u2014 as of July 2026.",
    "## Comparison table",
    "Comparison table",
    "| Family | Latest open-weight release | License | Native tool/function-calling | Notable for agents |",
    "| **NVIDIA Nemotron 3 Ultra** | Nemotron 3 Ultra (550B total / 55B active MoE, Jun 4 2026) | OpenMDW-1.1 (the Linux Foundation's open AI-model license; NVIDIA adopted it across the Cosmos/Isaac GR00T/Nemotron families; permissive \u2014 commercial and non-commercial use, modification, and redistribution allowed, patent-termination clause, no revenue/MAU cap) | Yes \u2014 tuned for long-running agent orchestration across hundreds of sequential tool calls | 1M context. Hybrid Mamba-Transformer MoE architecture with an NVFP4 pretraining/quantization recipe for compute-efficient long-horizon inference. Weights on Hugging Face (nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B). |"
   ],
   "model_evidence": [
    "# Open-Weight Models for Agents",
    "Cross-vendor comparison table of major open-weight LLM families \u2014 license, tool-calling support, context window, and agent-builder notes \u2014 as of July 2026.",
    "## Comparison table",
    "Comparison table",
    "| Family | Latest open-weight release | License | Native tool/function-calling | Notable for agents |",
    "| **NVIDIA Nemotron 3 Ultra** | Nemotron 3 Ultra (550B total / 55B active MoE, Jun 4 2026) | OpenMDW-1.1 (the Linux Foundation's open AI-model license; NVIDIA adopted it across the Cosmos/Isaac GR00T/Nemotron families; permissive \u2014 commercial and non-commercial use, modification, and redistribution allowed, patent-termination clause, no revenue/MAU cap) | Yes \u2014 tuned for long-running agent orchestration across hundreds of sequential tool calls | 1M context. Hybrid Mamba-Transformer MoE architecture with an NVFP4 pretraining/quantization recipe for compute-efficient long-horizon inference. Weights on Hugging Face (nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B). |"
   ],
   "release_date_evidence": [
    "# Open-Weight Models for Agents",
    "Cross-vendor comparison table of major open-weight LLM families \u2014 license, tool-calling support, context window, and agent-builder notes \u2014 as of July 2026.",
    "## Comparison table",
    "Comparison table",
    "| Family | Latest open-weight release | License | Native tool/function-calling | Notable for agents |",
    "| **NVIDIA Nemotron 3 Ultra** | Nemotron 3 Ultra (550B total / 55B active MoE, Jun 4 2026) | OpenMDW-1.1 (the Linux Foundation's open AI-model license; NVIDIA adopted it across the Cosmos/Isaac GR00T/Nemotron families; permissive \u2014 commercial and non-commercial use, modification, and redistribution allowed, patent-termination clause, no revenue/MAU cap) | Yes \u2014 tuned for long-running agent orchestration across hundreds of sequential tool calls | 1M context. Hybrid Mamba-Transformer MoE architecture with an NVFP4 pretraining/quantization recipe for compute-efficient long-horizon inference. Weights on Hugging Face (nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B). |"
   ],
   "access_type_evidence": [
    "# Open-Weight Models for Agents",
    "Cross-vendor comparison table of major open-weight LLM families \u2014 license, tool-calling support, context window, and agent-builder notes \u2014 as of July 2026.",
    "## Comparison table",
    "Comparison table",
    "| Family | Latest open-weight release | License | Native tool/function-calling | Notable for agents |",
    "| **NVIDIA Nemotron 3 Ultra** | Nemotron 3 Ultra (550B total / 55B active MoE, Jun 4 2026) | OpenMDW-1.1 (the Linux Foundation's open AI-model license; NVIDIA adopted it across the Cosmos/Isaac GR00T/Nemotron families; permissive \u2014 commercial and non-commercial use, modification, and redistribution allowed, patent-termination clause, no revenue/MAU cap) | Yes \u2014 tuned for long-running agent orchestration across hundreds of sequential tool calls | 1M context. Hybrid Mamba-Transformer MoE architecture with an NVFP4 pretraining/quantization recipe for compute-efficient long-horizon inference. Weights on Hugging Face (nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B). |"
   ],
   "license_evidence": [
    "# Open-Weight Models for Agents",
    "Cross-vendor comparison table of major open-weight LLM families \u2014 license, tool-calling support, context window, and agent-builder notes \u2014 as of July 2026.",
    "## Comparison table",
    "Comparison table",
    "| Family | Latest open-weight release | License | Native tool/function-calling | Notable for agents |",
    "| **NVIDIA Nemotron 3 Ultra** | Nemotron 3 Ultra (550B total / 55B active MoE, Jun 4 2026) | OpenMDW-1.1 (the Linux Foundation's open AI-model license; NVIDIA adopted it across the Cosmos/Isaac GR00T/Nemotron families; permissive \u2014 commercial and non-commercial use, modification, and redistribution allowed, patent-termination clause, no revenue/MAU cap) | Yes \u2014 tuned for long-running agent orchestration across hundreds of sequential tool calls | 1M context. Hybrid Mamba-Transformer MoE architecture with an NVFP4 pretraining/quantization recipe for compute-efficient long-horizon inference. Weights on Hugging Face (nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B). |"
   ],
   "significance_evidence": [
    "# Open-Weight Models for Agents",
    "Cross-vendor comparison table of major open-weight LLM families \u2014 license, tool-calling support, context window, and agent-builder notes \u2014 as of July 2026.",
    "## Comparison table",
    "Comparison table",
    "| Family | Latest open-weight release | License | Native tool/function-calling | Notable for agents |",
    "| **NVIDIA Nemotron 3 Ultra** | Nemotron 3 Ultra (550B total / 55B active MoE, Jun 4 2026) | OpenMDW-1.1 (the Linux Foundation's open AI-model license; NVIDIA adopted it across the Cosmos/Isaac GR00T/Nemotron families; permissive \u2014 commercial and non-commercial use, modification, and redistribution allowed, patent-termination clause, no revenue/MAU cap) | Yes \u2014 tuned for long-running agent orchestration across hundreds of sequential tool calls | 1M context. Hybrid Mamba-Transformer MoE architecture with an NVFP4 pretraining/quantization recipe for compute-efficient long-horizon inference. Weights on Hugging Face (nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B). |"
   ]
  },
  {
   "url": "https://huggingface.co/blog/ResterChed/kimi-k3-model-overview-mxfp4-quantization-open-wei",
   "title": "Kimi K3 Model Overview: 2.8T Parameters, MXFP4 Quantization, and What the Open Weights Mean for the Community",
   "published_at": "2026-07-17T14:56:58",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-16",
   "access_type": null,
   "license": null,
   "significance": "At 2.8 trillion parameters, it is the first open-source model to reach the 3-trillion-parameter class.",
   "organization_evidence": [
    "## *A structured overview for researchers and MLOps practitioners*",
    "Moonshot AI publicly released Kimi K3 on July 16, 2026, with full open-source weights promised by July 27.",
    "| **Developer** | Moonshot AI |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## *A structured overview for researchers and MLOps practitioners*",
    "Moonshot AI publicly released Kimi K3 on July 16, 2026, with full open-source weights promised by July 27.",
    "| **Release date** | July 16, 2026 (API); July 27, 2026 (weights) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## *A structured overview for researchers and MLOps practitioners*",
    "At 2.8 trillion parameters, it is the first open-source model to reach the 3-trillion-parameter class."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/gemma-4-e2b",
   "title": "Gemma 4 E2B - Intelligence, Performance & Price Analysis",
   "published_at": null,
   "organization": "Google",
   "model": null,
   "release_date": "2026-04-02",
   "access_type": null,
   "license": null,
   "significance": "5.1B total parameter reasoning model with a 128k context window supporting multimodal inputs",
   "organization_evidence": [
    "# Gemma 4 E2B (Reasoning) Intelligence, Performance & Price Analysis",
    "Gemma 4 E2B (Reasoning) was created by Google."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Gemma 4 E2B (Reasoning) Intelligence, Performance & Price Analysis",
    "Gemma 4 E2B (Reasoning) was released on April 2, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Gemma 4 E2B (Reasoning) Intelligence, Performance & Price Analysis",
    "Gemma 4 E2B (Reasoning) is above average in intelligence and well priced when comparing to other open weight models of similar size.",
    "The model supports text, image, speech, and video input, outputs text and image, and has a 128k tokens context window."
   ]
  },
  {
   "url": "https://developersdigest.tech/daily/2026-07-16",
   "title": "Inkling's 975B Open Weights, xAI Open-Sources Grok Build, and a DeepMind Departure",
   "published_at": "2026-07-16T00:00:00",
   "organization": "Thinking Machines",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "975 billion parameter Mixture-of-Experts model with 41B active parameters and fully open weights supporting a 1 million token context window",
   "organization_evidence": [
    "Briefing \u00b7 Thursday, July 16, 2026",
    "# Inkling's 975B Open Weights, xAI Open-Sources Grok Build, and a DeepMind Departure",
    "Thinking Machines ships Inkling, a 975B MoE with 41B active parameters and open weights",
    "Thinking Machines [released Inkling](https://thinkingmachines.ai/news/introducing-inkling/), a 975 billion parameter Mixture-of-Experts model with 41B active parameters and fully open weights."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Briefing \u00b7 Thursday, July 16, 2026",
    "# Inkling's 975B Open Weights, xAI Open-Sources Grok Build, and a DeepMind Departure",
    "Thinking Machines [released Inkling](https://thinkingmachines.ai/news/introducing-inkling/), a 975 billion parameter Mixture-of-Experts model with 41B active parameters and fully open weights."
   ]
  },
  {
   "url": "https://venturebeat.com/technology/thinking-machines-open-sources-first-multimodal-language-model-inkling-focused-on-low-cost-and-resistance-to-censorship",
   "title": "Thinking Machines open sources first multimodal language model, Inkling, focused on low cost and 'resistance to censorship'",
   "published_at": null,
   "organization": "Thinking Machines",
   "model": null,
   "release_date": "2026-00-00",
   "access_type": null,
   "license": null,
   "significance": "a 975 billion total parameter, natively multimodal, open-weights Mixture-of-Experts (MoE) system capable of reasoning across text, images, and audio",
   "organization_evidence": [
    "Today, Thinking Machines\u2014the highly capitalized American AI startup founded by former OpenAI CTO Mira Murati\u2014[released Inkling](https://thinkingmachines.ai/news/introducing-inkling/), its first major language model under an[ enterprise-friendly Apache 2.0 open source license](https://choosealicense.com/licenses/apache-2.0/), and it boasts high, if sub state-of-the-art, performance for open weights models on third-party benchmarks, specifically software engineering (77.6% on SWE-bench Verified, where it beats fellow U.S. open rival Nvidia Nemotron 3's 71.9%) and voice understanding (91.4% on VoiceBench compared to 94.4% for Gemini 3.1 Pro on high reasoning effort)."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "While Inkling is a formidable multimodal engine, it lands in a fiercely competitive 2026 open-weight landscape characterized by highly specialized MoE architectures."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Coming in at 975 billion total parameters, Inkling is a natively multimodal, open-weights Mixture-of-Experts (MoE) system capable of reasoning across text, images, and audio."
   ]
  },
  {
   "url": "https://robotsatlas.com/posts/tencent-hy3-open-moe-llm-apache-2",
   "title": "Tencent opens Hy3 \u2014 a 295B MoE model under Apache 2.0",
   "published_at": "2026-07-08T11:43:00",
   "organization": "Tencent",
   "model": null,
   "release_date": "2026-07-06",
   "access_type": null,
   "license": null,
   "significance": "A 295B MoE model with 21B active parameters, 256K-token context window, and 3.8B MTP layer that matches models two to five times its active size.",
   "organization_evidence": [
    "Artificial Intelligence",
    "Tencent has released Hy3, an open language model from the Hunyuan family built on a Mixture-of-Experts architecture, with 295 billion total parameters and 21 billion active on each query."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Artificial Intelligence",
    "The weights landed on Hugging Face on July 6, 2026 under an Apache 2.0 license, making Hy3 one of the most freely licensed large models out of China and a direct rival to Zhipu/Z.ai's open GLM family."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Artificial Intelligence",
    "Tencent has released Hy3, an open language model from the Hunyuan family built on a Mixture-of-Experts architecture, with 295 billion total parameters and 21 billion active on each query.",
    "Tencent's pitch rests on the claim that Hy3 matches models two to five times its active size."
   ]
  },
  {
   "url": "https://llm-releases.com/families/laguna",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Poolside",
   "model": null,
   "release_date": "2026-05-22",
   "access_type": null,
   "license": null,
   "significance": "A 118B-total / 8B-active Mixture-of-Experts with 1M-token context matching or beating models several times its size in coding and terminal benchmarks.",
   "organization_evidence": [
    "Model family timeline",
    "A source-backed timeline for the Laguna model family, collecting release dates, labs, access details, context windows, and major lifecycle changes.",
    "Poolside's open-weight agentic-coding model and a scale-up of the Laguna XS family (same pre-training data as XS 2.1): a 118B-total / 8B-active Mixture-of-Experts that activates only ~6.8% of its parameters per token, giving larger-model behavior while staying cheap to serve, with a 1M-token context in both thinking and no-thinking modes."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Model family timeline",
    "A source-backed timeline for the Laguna model family, collecting release dates, labs, access details, context windows, and major lifecycle changes.",
    "Trained in under nine weeks on 4,096 NVIDIA H200 GPUs (pre-training began 22 May 2026); first Poolside model with RL in FP8."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Model family timeline",
    "A source-backed timeline for the Laguna model family, collecting release dates, labs, access details, context windows, and major lifecycle changes.",
    "Poolside's open-weight agentic-coding model and a scale-up of the Laguna XS family (same pre-training data as XS 2.1): a 118B-total / 8B-active Mixture-of-Experts that activates only ~6.8% of its parameters per token, giving larger-model behavior while staying cheap to serve, with a 1M-token context in both thinking and no-thinking modes."
   ]
  },
  {
   "url": "https://llm-releases.com/families/laguna",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Poolside",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "A 33B-total / 3B-active Mixture-of-Experts built for agentic coding and long-horizon work on a local machine, served at 256K context.",
   "organization_evidence": [
    "Model family timeline",
    "A source-backed timeline for the Laguna model family, collecting release dates, labs, access details, context windows, and major lifecycle changes.",
    "Poolside's open-weight small coding model: a 33B-total / 3B-active Mixture-of-Experts built for agentic coding and long-horizon work on a local machine, served at 256K context."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Model family timeline",
    "A source-backed timeline for the Laguna model family, collecting release dates, labs, access details, context windows, and major lifecycle changes.",
    "Poolside's open-weight small coding model: a 33B-total / 3B-active Mixture-of-Experts built for agentic coding and long-horizon work on a local machine, served at 256K context."
   ]
  },
  {
   "url": "https://fastino.ai/blog/closing-the-gap-open-weight-vs-proprietary-frontier-language-models",
   "title": "Closing the Gap: Open-Weight vs. Proprietary Frontier Language Models",
   "published_at": "2026-07-20T00:00:00",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-06-16",
   "access_type": null,
   "license": null,
   "significance": "744-billion-parameter mixture-of-experts model with a 1-million-token context window",
   "organization_evidence": [
    "Research",
    "Z.ai shipped GLM-5.2 in mid-June under an MIT license, API live on June 16."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Research",
    "Z.ai shipped GLM-5.2 in mid-June under an MIT license, API live on June 16."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Research",
    "**GLM-5.2** is a 744-billion-parameter mixture-of-experts (MoE) model with about 40 billion active parameters per forward pass, a 1-million-token context window, and up to 128K output tokens."
   ]
  },
  {
   "url": "https://fastino.ai/blog/closing-the-gap-open-weight-vs-proprietary-frontier-language-models",
   "title": "Closing the Gap: Open-Weight vs. Proprietary Frontier Language Models",
   "published_at": "2026-07-20T00:00:00",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-27",
   "access_type": null,
   "license": null,
   "significance": "2.8-trillion-parameter MoE model, the first open model to reach the 3-trillion class",
   "organization_evidence": [
    "Research",
    "A month later, on July 16, Moonshot AI announced Kimi K3, with API access at launch and full weights arriving on Hugging Face by July 27."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Research",
    "A month later, on July 16, Moonshot AI announced Kimi K3, with API access at launch and full weights arriving on Hugging Face by July 27."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Research",
    "**Kimi K3** is a 2.8-trillion-parameter MoE model, the first open model to reach the 3-trillion class."
   ]
  },
  {
   "url": "https://edgen.tech/zh/news/post/zai-confirms-ox-alpha-as-glm-model-with-usage-2x-deepseek",
   "title": "Z.AI confirms Ox Alpha as GLM model with usage 2x DeepSeek",
   "published_at": "2026-08-26T12:16:39",
   "organization": "Z.AI",
   "model": null,
   "release_date": "2026-08-26",
   "access_type": null,
   "license": null,
   "significance": "Ox Alpha's usage more than doubled DeepSeek's on OpenRouter's leaderboard, making it the biggest launch in the marketplace's history, according to the company.",
   "organization_evidence": [
    "# Z.AI confirms Ox Alpha as GLM model with usage 2x DeepSeek",
    "The mystery AI model that swept to the top of OpenRouter's usage charts is a new GLM-family release from China's Z.AI, which confirmed the launch Wednesday and said it will release the model's weights tonight."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "[![Alex Nguyen](https://static.edwealth.ai/authors/alex-nguyen.png?w=32)\n## Alex Nguyen](https://www.edgen.tech/zh/author/news/alex-nguyen)\u00b7Aug 26 2026, 12:16",
    "# Z.AI confirms Ox Alpha as GLM model with usage 2x DeepSeek",
    "The mystery AI model that swept to the top of OpenRouter's usage charts is a new GLM-family release from China's Z.AI, which confirmed the launch Wednesday and said it will release the model's weights tonight."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Z.AI confirms Ox Alpha as GLM model with usage 2x DeepSeek",
    "Ox Alpha's usage more than doubled DeepSeek's on OpenRouter's leaderboard, making it the biggest launch in the marketplace's history, according to the company."
   ]
  },
  {
   "url": "https://ask-coreai.com/blog/ai-model-releases-2026-so-far-complete-list",
   "title": "AI Model Releases of 2026 So Far: The Complete List",
   "published_at": "2026-07-28T14:42:16",
   "organization": "MoonshotAI",
   "model": null,
   "release_date": "2026-07-16",
   "access_type": null,
   "license": null,
   "significance": "2.8T parameters, multimodal reasoning",
   "organization_evidence": [
    "# AI Model Releases of 2026 So Far: The Complete List",
    "## The 2026 AI Model Releases Timeline",
    "| Date (2026) | Model | Lab | Context | Notes |",
    "| Jul 16 | Kimi K3 | MoonshotAI | 1,048,576 | 2.8T params, multimodal reasoning |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# AI Model Releases of 2026 So Far: The Complete List",
    "## The 2026 AI Model Releases Timeline",
    "| Date (2026) | Model | Lab | Context | Notes |",
    "| Jul 16 | Kimi K3 | MoonshotAI | 1,048,576 | 2.8T params, multimodal reasoning |",
    "Kimi K3 shipped on July 16, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# AI Model Releases of 2026 So Far: The Complete List",
    "## The 2026 AI Model Releases Timeline",
    "| Date (2026) | Model | Lab | Context | Notes |",
    "| Jul 16 | Kimi K3 | MoonshotAI | 1,048,576 | 2.8T params, multimodal reasoning |"
   ]
  },
  {
   "url": "https://ask-coreai.com/blog/ai-model-releases-2026-so-far-complete-list",
   "title": "AI Model Releases of 2026 So Far: The Complete List",
   "published_at": "2026-07-28T14:42:16",
   "organization": "NVIDIA",
   "model": null,
   "release_date": "2026-06-04",
   "access_type": null,
   "license": null,
   "significance": "Open-weight heavyweight",
   "organization_evidence": [
    "# AI Model Releases of 2026 So Far: The Complete List",
    "## The 2026 AI Model Releases Timeline",
    "| Date (2026) | Model | Lab | Context | Notes |",
    "| Jun 4 | Nemotron 3 Ultra 550B | NVIDIA | 512,288 | Open-weight heavyweight |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# AI Model Releases of 2026 So Far: The Complete List",
    "## The 2026 AI Model Releases Timeline",
    "| Date (2026) | Model | Lab | Context | Notes |",
    "| Jun 4 | Nemotron 3 Ultra 550B | NVIDIA | 512,288 | Open-weight heavyweight |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# AI Model Releases of 2026 So Far: The Complete List",
    "## The 2026 AI Model Releases Timeline",
    "| Date (2026) | Model | Lab | Context | Notes |",
    "| Jun 4 | Nemotron 3 Ultra 550B | NVIDIA | 512,288 | Open-weight heavyweight |"
   ]
  },
  {
   "url": "https://byteiota.com/kimi-k3-open-weight-frontier-model",
   "title": "Kimi K3: The Open-Weight Model That Finally Hit Frontier",
   "published_at": "2026-08-23T08:38:27",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-27",
   "access_type": null,
   "license": null,
   "significance": "A 2.8-trillion-parameter Mixture-of-Experts model that became the first open-weight model to reach the top of the Artificial Analysis Intelligence Index.",
   "organization_evidence": [
    "On July 27, 2026, Moonshot AI closed the gap.",
    "Open-weight models have been chasing frontier performance for two years."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "On July 27, 2026, Moonshot AI closed the gap.",
    "Open-weight models have been chasing frontier performance for two years."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Kimi K3, a 2.8-trillion-parameter Mixture-of-Experts model, became the first open-weight model to reach the top of the [Artificial Analysis Intelligence Index](https://artificialanalysis.ai/leaderboards/models) \u2014 tied for first with the closed-source flagships.",
    "Open-weight models have been chasing frontier performance for two years."
   ]
  },
  {
   "url": "https://huggingface.co/blog/nvidia/nemotron-3-5-content-safety",
   "title": "Nemotron 3.5 Content Safety: Customizable Multimodal Safety for Global Enterprise AI",
   "published_at": "2026-06-04T18:57:45",
   "organization": "NVIDIA",
   "model": null,
   "release_date": "2026-03",
   "access_type": null,
   "license": null,
   "significance": "combined multimodal and multilingual capabilities for the first time in a single 4B-parameter model",
   "organization_evidence": [
    "The last two years have seen NVIDIA's content safety stack grow from a focused English text classifier into a family of specialized models\u2014each extending coverage to new modalities, languages, and inference modes.",
    "Getting Started"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "[Nemotron 3 Content Safety](https://huggingface.co/nvidia/Nemotron-3-Content-Safety), released in March 2026, combined multimodal and multilingual capabilities for the first time in a single 4B-parameter model.",
    "Getting Started"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "[Nemotron 3 Content Safety](https://huggingface.co/nvidia/Nemotron-3-Content-Safety), released in March 2026, combined multimodal and multilingual capabilities for the first time in a single 4B-parameter model.",
    "Getting Started"
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/inkling-small",
   "title": "Inkling Small - Intelligence, Performance & Price Analysis",
   "published_at": null,
   "organization": "Thinking Machines",
   "model": null,
   "release_date": "2026-07-30",
   "access_type": null,
   "license": null,
   "significance": "A 266B parameter (12B active) Mixture of Experts reasoning model supporting text, image, and speech input with a 1M token context window.",
   "organization_evidence": [
    "Inkling Small Intelligence, Performance & Price Analysis",
    "Inkling Small was created by Thinking Machines."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Inkling Small Intelligence, Performance & Price Analysis",
    "Inkling Small was released on July 30, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Inkling Small Intelligence, Performance & Price Analysis",
    "The model supports text, image, and speech input, outputs text, and has a 1M tokens context window.",
    "Inkling Small has 266 billion parameters (12 billion active)."
   ]
  },
  {
   "url": "https://bivashvlog.com/gemini-robotics-er-2-thinking-machines-inkling",
   "title": "Gemini Robotics ER 2 and Thinking Machines' Inkling Explained",
   "published_at": "2026-07-31T00:03:09",
   "organization": "Thinking Machines Lab",
   "model": "Inkling",
   "release_date": "2026-07-16",
   "access_type": "open-weights",
   "license": "Apache 2.0",
   "significance": "Mixture-of-Experts transformer with 975 billion total parameters and 41 billion active per forward pass, natively multimodal across text, image, and audio.",
   "organization_evidence": [
    "Gemini Robotics ER 2 and Thinking Machines\u2019 Inkling Explained. A complete breakdown of Google DeepMind\u2019s Gemini Robotics ER 2 and Thinking Machines\u2019 Inkling open-weights model: specs, benchmarks, and what\u2019s next.",
    "On July 16, Thinking Machines Lab, the startup founded by former OpenAI CTO Mira Murati, released its first open-weights model."
   ],
   "model_evidence": [
    "Gemini Robotics ER 2 and Thinking Machines\u2019 Inkling Explained. A complete breakdown of Google DeepMind\u2019s Gemini Robotics ER 2 and Thinking Machines\u2019 Inkling open-weights model: specs, benchmarks, and what\u2019s next.",
    "On July 16, Thinking Machines Lab, the startup founded by former OpenAI CTO Mira Murati, released its first open-weights model. It\u2019s called Inkling, and unlike a lot of \u201copen\u201d releases that quietly trail a generation behind the frontier, this one is upfront about exactly what it isn\u2019t."
   ],
   "release_date_evidence": [
    "| **Released** | July 16, 2026 | July 30, 2026 |",
    "At a Glance",
    "On July 16, Thinking Machines Lab, the startup founded by former OpenAI CTO Mira Murati, released its first open-weights model."
   ],
   "access_type_evidence": [
    "Gemini Robotics ER 2 and Thinking Machines\u2019 Inkling Explained. A complete breakdown of Google DeepMind\u2019s Gemini Robotics ER 2 and Thinking Machines\u2019 Inkling open-weights model: specs, benchmarks, and what\u2019s next.",
    "On July 16, Thinking Machines Lab, the startup founded by former OpenAI CTO Mira Murati, released its first open-weights model.",
    "| **What it is** | Open-weights multimodal foundation model | Embodied reasoning model for robots |",
    "At a Glance"
   ],
   "license_evidence": [
    "| **License / access** | Apache 2.0, full weights on Hugging Face, fine-tunable via Tinker | Public preview via Gemini API and AI Studio |",
    "At a Glance",
    "The full weights are on Hugging Face under an Apache 2.0 license, about as permissive as licensing gets: no usage restrictions baked into the license itself, free to modify and redistribute."
   ],
   "significance_evidence": [
    "Inkling is a Mixture-of-Experts transformer with 975 billion total parameters and 41 billion active per forward pass.",
    "The Core Model: What\u2019s Actually Inside Inkling"
   ]
  },
  {
   "url": "https://huggingface.co/blog/Svngoku/qwen38-vs-deepseek-v4-vs-opus-coding-benchmarks",
   "title": "Qwen3.8 vs DeepSeek-V4 vs Claude Opus \u2014 Coding Benchmark Comparison",
   "published_at": "2026-08-16T22:10:56",
   "organization": "Alibaba",
   "model": "Qwen3.8-27B",
   "release_date": "2026-08-14",
   "access_type": "open-weight",
   "license": "Apache 2.0",
   "significance": "surprise value play with SWE-bench Pro 61.7 and agentic-terminal 73.0, runnable on a consumer GPU",
   "organization_evidence": [
    "3. **Qwen3.8** (August) \u2014 Alibaba's biggest model yet (2.4T / 95B active), plus an open-weight 27B, closing the gap on agentic coding.",
    "## Context"
   ],
   "model_evidence": [
    "- **Qwen3.8-27B** (open-weight, Apache 2.0, Aug 14, 2026) is the surprise value play: **SWE-bench Pro 61.7** and agentic-terminal 73.0, beating the larger Qwen3.7-Plus on coding, runnable on a consumer GPU.",
    "## TL;DR",
    "3. **Qwen3.8** (August) \u2014 Alibaba's biggest model yet (2.4T / 95B active), plus an open-weight 27B, closing the gap on agentic coding."
   ],
   "release_date_evidence": [
    "- **Qwen3.8-27B** (open-weight, Apache 2.0, Aug 14, 2026) is the surprise value play: **SWE-bench Pro 61.7** and agentic-terminal 73.0, beating the larger Qwen3.7-Plus on coding, runnable on a consumer GPU.",
    "## TL;DR"
   ],
   "access_type_evidence": [
    "- **Qwen3.8-27B** (open-weight, Apache 2.0, Aug 14, 2026) is the surprise value play: **SWE-bench Pro 61.7** and agentic-terminal 73.0, beating the larger Qwen3.7-Plus on coding, runnable on a consumer GPU.",
    "## TL;DR",
    "3. **Qwen3.8** (August) \u2014 Alibaba's biggest model yet (2.4T / 95B active), plus an open-weight 27B, closing the gap on agentic coding."
   ],
   "license_evidence": [
    "- **Qwen3.8-27B** (open-weight, Apache 2.0, Aug 14, 2026) is the surprise value play: **SWE-bench Pro 61.7** and agentic-terminal 73.0, beating the larger Qwen3.7-Plus on coding, runnable on a consumer GPU.",
    "## TL;DR"
   ],
   "significance_evidence": [
    "- **Qwen3.8-27B** (open-weight, Apache 2.0, Aug 14, 2026) is the surprise value play: **SWE-bench Pro 61.7** and agentic-terminal 73.0, beating the larger Qwen3.7-Plus on coding, runnable on a consumer GPU.",
    "## TL;DR"
   ]
  },
  {
   "url": "https://edgen.tech/vi/news/post/zai-confirms-ox-alpha-as-glm-model-with-usage-2x-deepseek",
   "title": "Z.AI confirms Ox Alpha as GLM model with usage 2x DeepSeek",
   "published_at": "2026-08-26T12:16:39",
   "organization": "Z.AI",
   "model": null,
   "release_date": "2026-08-26",
   "access_type": null,
   "license": null,
   "significance": "reasoning model carrying a 1,048,600-token context window that more than doubled DeepSeek's usage on OpenRouter",
   "organization_evidence": [
    "# Z.AI confirms Ox Alpha as GLM model with usage 2x DeepSeek",
    "The mystery AI model that swept to the top of OpenRouter's usage charts is a new GLM-family release from China's Z.AI, which confirmed the launch Wednesday and said it will release the model's weights tonight."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Z.AI confirms Ox Alpha as GLM model with usage 2x DeepSeek",
    "[![Alex Nguyen](https://static.edwealth.ai/authors/alex-nguyen.png?w=32)\n## Alex Nguyen](https://www.edgen.tech/vi/author/news/alex-nguyen)\u00b7Aug 26 2026, 12:16",
    "The mystery AI model that swept to the top of OpenRouter's usage charts is a new GLM-family release from China's Z.AI, which confirmed the launch Wednesday and said it will release the model's weights tonight."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Z.AI confirms Ox Alpha as GLM model with usage 2x DeepSeek",
    "Ox Alpha's usage more than doubled DeepSeek's on OpenRouter's leaderboard, making it the biggest launch in the marketplace's history, according to the company."
   ]
  },
  {
   "url": "https://codersera.com/blog/open-source-llms-landscape-2026",
   "title": "Open-Source LLMs Landscape 2026: Qwen, Llama, DeepSeek, Kimi",
   "published_at": "2026-05-01T07:17:14",
   "organization": "Meta",
   "model": null,
   "release_date": "2026-08-10",
   "access_type": null,
   "license": null,
   "significance": "a 30B dense decoder-only multimodal model with a 128K context and built-in vision encoder",
   "organization_evidence": [
    "# The Open-Source LLM Landscape in 2026",
    "**August 13, 2026 update:** Meta reversed its closed-weights pivot \u2014 **Muse Glimmer** (30B dense, Apache 2.0, 128K context) shipped August 10 and runs on one 24\u201332 GB GPU or Mac, and Muse Spark 1.2\u2019s weights are announced but not yet released.",
    "On August 10, 2026, Meta Superintelligence Labs released **Muse Glimmer**: a 30B dense decoder-only multimodal model (built-in vision encoder) under **Apache 2.0**, with a 128K context, text + image input, training across 100+ languages, and a lineage distilled from the closed Muse Spark line."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# The Open-Source LLM Landscape in 2026",
    "**August 13, 2026 update:** Meta reversed its closed-weights pivot \u2014 **Muse Glimmer** (30B dense, Apache 2.0, 128K context) shipped August 10 and runs on one 24\u201332 GB GPU or Mac, and Muse Spark 1.2\u2019s weights are announced but not yet released.",
    "On August 10, 2026, Meta Superintelligence Labs released **Muse Glimmer**: a 30B dense decoder-only multimodal model (built-in vision encoder) under **Apache 2.0**, with a 128K context, text + image input, training across 100+ languages, and a lineage distilled from the closed Muse Spark line."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# The Open-Source LLM Landscape in 2026",
    "On August 10, 2026, Meta Superintelligence Labs released **Muse Glimmer**: a 30B dense decoder-only multimodal model (built-in vision encoder) under **Apache 2.0**, with a 128K context, text + image input, training across 100+ languages, and a lineage distilled from the closed Muse Spark line."
   ]
  },
  {
   "url": "https://codersera.com/blog/open-source-llms-landscape-2026",
   "title": "Open-Source LLMs Landscape 2026: Qwen, Llama, DeepSeek, Kimi",
   "published_at": "2026-05-01T07:17:14",
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-08-12",
   "access_type": null,
   "license": null,
   "significance": "the second-largest open model ever released at 2.4T total / 95B active, essentially matching the hosted Qwen3.8-Max on text intelligence",
   "organization_evidence": [
    "# The Open-Source LLM Landscape in 2026",
    "**And on August 12\u201314 Alibaba re-opened the Qwen top tier**: **Qwen3.8-2.4T-A95B** (the first-ever Qwen-Max-class open release, under a custom licence) and **Qwen3.8-27B** (Apache 2.0, native vision-language, 262K context, Artificial Analysis score 52 \u2014 level with GLM-5.2).",
    "On **August 12, 2026** Alibaba released the weights of **Qwen3.8-2.4T-A95B** \u2014 the first time a Qwen-Max-class flagship has ever been open-weighted."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# The Open-Source LLM Landscape in 2026",
    "**And on August 12\u201314 Alibaba re-opened the Qwen top tier**: **Qwen3.8-2.4T-A95B** (the first-ever Qwen-Max-class open release, under a custom licence) and **Qwen3.8-27B** (Apache 2.0, native vision-language, 262K context, Artificial Analysis score 52 \u2014 level with GLM-5.2).",
    "On **August 12, 2026** Alibaba released the weights of **Qwen3.8-2.4T-A95B** \u2014 the first time a Qwen-Max-class flagship has ever been open-weighted."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# The Open-Source LLM Landscape in 2026",
    "At 2.4T total / 95B active it is the **second-largest open model ever released** \u2014 Kimi K3 (2.8T) is larger \u2014 and per Artificial Analysis the open weights essentially match the hosted Qwen3.8-Max on text intelligence (Agentic Index 57.1 vs 58.4, with the open build actually ahead on Terminal-Bench 2.1 and GPQA)."
   ]
  },
  {
   "url": "https://codersera.com/blog/open-source-llms-landscape-2026",
   "title": "Open-Source LLMs Landscape 2026: Qwen, Llama, DeepSeek, Kimi",
   "published_at": "2026-05-01T07:17:14",
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-08-14",
   "access_type": null,
   "license": null,
   "significance": "a dense, native vision-language model with 262K context scoring 52 on Artificial Analysis at 27B parameters",
   "organization_evidence": [
    "# The Open-Source LLM Landscape in 2026",
    "**And on August 12\u201314 Alibaba re-opened the Qwen top tier**: **Qwen3.8-2.4T-A95B** (the first-ever Qwen-Max-class open release, under a custom licence) and **Qwen3.8-27B** (Apache 2.0, native vision-language, 262K context, Artificial Analysis score 52 \u2014 level with GLM-5.2).",
    "Two days later, on **August 14**, came the release most people can actually run: **Qwen3.8-27B** \u2014 dense, **Apache 2.0**, native vision-language (image and hour-scale video input), 262K context, ~17GB at Q4_K_M on one GPU."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# The Open-Source LLM Landscape in 2026",
    "**And on August 12\u201314 Alibaba re-opened the Qwen top tier**: **Qwen3.8-2.4T-A95B** (the first-ever Qwen-Max-class open release, under a custom licence) and **Qwen3.8-27B** (Apache 2.0, native vision-language, 262K context, Artificial Analysis score 52 \u2014 level with GLM-5.2).",
    "Two days later, on **August 14**, came the release most people can actually run: **Qwen3.8-27B** \u2014 dense, **Apache 2.0**, native vision-language (image and hour-scale video input), 262K context, ~17GB at Q4_K_M on one GPU."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# The Open-Source LLM Landscape in 2026",
    "Two days later, on **August 14**, came the release most people can actually run: **Qwen3.8-27B** \u2014 dense, **Apache 2.0**, native vision-language (image and hour-scale video input), 262K context, ~17GB at Q4_K_M on one GPU.",
    "Artificial Analysis independently scores it **52** on its Intelligence Index \u2014 up from 38 for the architecturally identical Qwen3.6-27B, level with GLM-5.2 and DeepSeek V4 Flash, and ahead of every open model in the 40B\u2013150B range."
   ]
  },
  {
   "url": "https://ainchina.com/blog/moonshot-kimi-k3-2-8-trillion-open-ai-china-2026",
   "title": "Moonshot's Kimi K3: How a 2.8-Trillion-Parameter Open-Weight Model Shook Global Markets",
   "published_at": "2026-07-30T00:00:00",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-27",
   "access_type": null,
   "license": null,
   "significance": "2.8-trillion-parameter open-weight model with 1 million token context window",
   "organization_evidence": [
    "# Moonshot's Kimi K3: How a 2.8-Trillion-Parameter Open-Weight Model Shook Global Markets",
    "At 11:00 PM Beijing time on July 27, 2026, engineers at Moonshot AI pressed a button that would reverberate through Silicon Valley within hours."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "AI Models16 min read",
    "# Moonshot's Kimi K3: How a 2.8-Trillion-Parameter Open-Weight Model Shook Global Markets",
    "At 11:00 PM Beijing time on July 27, 2026, engineers at Moonshot AI pressed a button that would reverberate through Silicon Valley within hours.",
    "| K3 weights released | July 27, 2026 | Full open weights on Hugging Face |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "AI Models16 min read",
    "# Moonshot's Kimi K3: How a 2.8-Trillion-Parameter Open-Weight Model Shook Global Markets",
    "At 11:00 PM Beijing time on July 27, 2026, engineers at Moonshot AI pressed a button that would reverberate through Silicon Valley within hours.",
    "The full weights for Kimi K3 \u2014 a 2.8-trillion-parameter AI model \u2014 went live on Hugging Face."
   ]
  },
  {
   "url": "https://huggingface.co/blog/huggingface/one-year-since-the-deepseek-moment-blog-2",
   "title": "Architectural Choices in China's Open-Source AI Ecosystem: Building Beyond DeepSeek",
   "published_at": "2026-01-27T15:01:45",
   "organization": "Zhipu",
   "model": null,
   "release_date": "2026",
   "access_type": null,
   "license": null,
   "significance": "trained entirely on domestic chips",
   "organization_evidence": [
    "In this second piece we turn our focus from models to the architectural and hardware choices Chinese companies have made as openness becomes the norm.",
    "At the beginning of 2026, Zhipu's [GLM-Image ](https://huggingface.co/zai-org/GLM-Image)and China Telecom's latest open model - [TeleChat3](https://huggingface.co/Tele-AI/TeleChat3-36B-Thinking), were both announced as being trained entirely on domestic chips."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "In this second piece we turn our focus from models to the architectural and hardware choices Chinese companies have made as openness becomes the norm.",
    "At the beginning of 2026, Zhipu's [GLM-Image ](https://huggingface.co/zai-org/GLM-Image)and China Telecom's latest open model - [TeleChat3](https://huggingface.co/Tele-AI/TeleChat3-36B-Thinking), were both announced as being trained entirely on domestic chips."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "In this second piece we turn our focus from models to the architectural and hardware choices Chinese companies have made as openness becomes the norm.",
    "At the beginning of 2026, Zhipu's [GLM-Image ](https://huggingface.co/zai-org/GLM-Image)and China Telecom's latest open model - [TeleChat3](https://huggingface.co/Tele-AI/TeleChat3-36B-Thinking), were both announced as being trained entirely on domestic chips."
   ]
  },
  {
   "url": "https://huggingface.co/blog/huggingface/one-year-since-the-deepseek-moment-blog-2",
   "title": "Architectural Choices in China's Open-Source AI Ecosystem: Building Beyond DeepSeek",
   "published_at": "2026-01-27T15:01:45",
   "organization": "China Telecom",
   "model": null,
   "release_date": "2026",
   "access_type": null,
   "license": null,
   "significance": "latest open model trained entirely on domestic chips",
   "organization_evidence": [
    "In this second piece we turn our focus from models to the architectural and hardware choices Chinese companies have made as openness becomes the norm.",
    "At the beginning of 2026, Zhipu's [GLM-Image ](https://huggingface.co/zai-org/GLM-Image)and China Telecom's latest open model - [TeleChat3](https://huggingface.co/Tele-AI/TeleChat3-36B-Thinking), were both announced as being trained entirely on domestic chips."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "In this second piece we turn our focus from models to the architectural and hardware choices Chinese companies have made as openness becomes the norm.",
    "At the beginning of 2026, Zhipu's [GLM-Image ](https://huggingface.co/zai-org/GLM-Image)and China Telecom's latest open model - [TeleChat3](https://huggingface.co/Tele-AI/TeleChat3-36B-Thinking), were both announced as being trained entirely on domestic chips."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "In this second piece we turn our focus from models to the architectural and hardware choices Chinese companies have made as openness becomes the norm.",
    "At the beginning of 2026, Zhipu's [GLM-Image ](https://huggingface.co/zai-org/GLM-Image)and China Telecom's latest open model - [TeleChat3](https://huggingface.co/Tele-AI/TeleChat3-36B-Thinking), were both announced as being trained entirely on domestic chips."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/hy3",
   "title": "Hy3 - Intelligence, Performance & Price Analysis | Artificial Analysis",
   "published_at": null,
   "organization": "Tencent",
   "model": null,
   "release_date": "2026-07-06",
   "access_type": null,
   "license": null,
   "significance": "A 299B parameter reasoning Mixture of Experts model with a 256k context window and competitive intelligence index scores.",
   "organization_evidence": [
    "![Hy3 logo](/img/logos/tencent_small.svg)",
    "Hy3 Intelligence, Performance & Price Analysis",
    "Hy3 was created by Tencent."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "\u2022\nReleased July 2026",
    "Hy3 Intelligence, Performance & Price Analysis",
    "Hy3 was released on July 6, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Hy3 is amongst the leading models in intelligence and reasonably priced when comparing to other open weight models of similar size.",
    "Hy3 Intelligence, Performance & Price Analysis",
    "| Total parameters | 299B |",
    "| Active parameters | 21B Number of parameters active per token during inference |",
    "Yes, Hy3 is a reasoning model."
   ]
  },
  {
   "url": "https://gadgetsnow.indiatimes.com/tech-news/harvey-builds-its-own-legal-ai-model-on-chinese-open-weights/articleshow/133352879.cms?frmapp=yes",
   "title": "Harvey Builds Its Own Legal AI Model On Chinese Open Weights",
   "published_at": "2026-08-19T16:00:00",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-27",
   "access_type": null,
   "license": null,
   "significance": "2.8-trillion-parameter open-weight release",
   "organization_evidence": [
    "Harvey Builds Its Own Legal AI Model On Chinese Open Weights",
    "Reporting from the supply chain places Tenet's base model as Kimi K3, Moonshot AI's 2.8-trillion-parameter open-weight release from July 2026."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Harvey Builds Its Own Legal AI Model On Chinese Open Weights",
    "Kimi K3 shipped as open weights on 27 July, which means the model file itself is downloadable and runnable on hardware its user controls."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Harvey Builds Its Own Legal AI Model On Chinese Open Weights",
    "Reporting from the supply chain places Tenet's base model as Kimi K3, Moonshot AI's 2.8-trillion-parameter open-weight release from July 2026."
   ]
  },
  {
   "url": "https://edgen.tech/tr/news/post/zai-confirms-ox-alpha-as-glm-model-with-usage-2x-deepseek",
   "title": "Z.AI confirms Ox Alpha as GLM model with usage 2x DeepSeek",
   "published_at": "2026-08-26T12:16:39",
   "organization": "Z.AI",
   "model": null,
   "release_date": "2026-08-26",
   "access_type": null,
   "license": null,
   "significance": "Ox Alpha's usage more than doubled DeepSeek's on OpenRouter's leaderboard, making it the biggest launch in the marketplace's history",
   "organization_evidence": [
    "Z.AI confirms Ox Alpha as GLM model with usage 2x DeepSeek",
    "Z.AI confirmed Ox Alpha is a new GLM-family model, releasing its weights tonight.",
    "The mystery AI model that swept to the top of OpenRouter's usage charts is a new GLM-family release from China's Z.AI, which confirmed the launch Wednesday and said it will release the model's weights tonight."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "[![Alex Nguyen](https://static.edwealth.ai/authors/alex-nguyen.png?w=32)\n## Alex Nguyen](https://www.edgen.tech/tr/author/news/alex-nguyen)\u00b7Aug 26 2026, 12:16",
    "Z.AI confirms Ox Alpha as GLM model with usage 2x DeepSeek",
    "The mystery AI model that swept to the top of OpenRouter's usage charts is a new GLM-family release from China's Z.AI, which confirmed the launch Wednesday and said it will release the model's weights tonight."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Z.AI confirms Ox Alpha as GLM model with usage 2x DeepSeek",
    "Ox Alpha's usage more than doubled DeepSeek's on OpenRouter's leaderboard, making it the biggest launch in the marketplace's history, according to the company."
   ]
  },
  {
   "url": "https://tech-insider.org/kimi-k3-vs-qwen3-8-max-vs-glm-5-2-2026",
   "title": "Kimi K3 vs Qwen3.8-Max vs GLM-5.2 Compared [2026]",
   "published_at": "2026-08-21T12:13:10",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-16",
   "access_type": null,
   "license": null,
   "significance": "A 2.8-trillion-parameter Mixture-of-Experts model posting the highest GPQA Diamond score of any open model tracked at 93.5%.",
   "organization_evidence": [
    "# Kimi K3 vs Qwen3.8-Max vs GLM-5.2: $10.60 Gap [2026]",
    "Moonshot AI released **Kimi K3** on July 16, 2026, with full weights following on July 27."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Kimi K3 vs Qwen3.8-Max vs GLM-5.2: $10.60 Gap [2026]",
    "Moonshot AI released **Kimi K3** on July 16, 2026, with full weights following on July 27."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Kimi K3 vs Qwen3.8-Max vs GLM-5.2: $10.60 Gap [2026]",
    "## Kimi K3: Moonshot AI\u2019s 2.8-Trillion-Parameter Bet",
    "On independent benchmark aggregation, K3 currently posts the highest GPQA Diamond score of any open model tracked, at 93.5%, edging out both Qwen3.8-Max and GLM-5.2 on that specific graduate-level reasoning test."
   ]
  },
  {
   "url": "https://tech-insider.org/kimi-k3-vs-qwen3-8-max-vs-glm-5-2-2026",
   "title": "Kimi K3 vs Qwen3.8-Max vs GLM-5.2 Compared [2026]",
   "published_at": "2026-08-21T12:13:10",
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-08-03",
   "access_type": null,
   "license": null,
   "significance": "A 2.4-trillion-parameter Mixture-of-Experts model, the first Max-tier model Alibaba ever open-sourced.",
   "organization_evidence": [
    "# Kimi K3 vs Qwen3.8-Max vs GLM-5.2: $10.60 Gap [2026]",
    "Alibaba followed with **Qwen3.8-Max** on August 3, 2026, its first \u201cMax-class\u201d model ever released with open weights."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Kimi K3 vs Qwen3.8-Max vs GLM-5.2: $10.60 Gap [2026]",
    "Alibaba followed with **Qwen3.8-Max** on August 3, 2026, its first \u201cMax-class\u201d model ever released with open weights."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Kimi K3 vs Qwen3.8-Max vs GLM-5.2: $10.60 Gap [2026]",
    "## Qwen3.8-Max: Alibaba\u2019s First Open-Sourced Max-Class Model",
    "The architecture is a 2.4-trillion-parameter Mixture-of-Experts model with a 1-million-token context window, broken down as roughly 991,000 tokens of maximum input, 131,000 tokens of maximum output, and up to 262,000 tokens reserved for internal reasoning budget when the model runs in thinking mode."
   ]
  },
  {
   "url": "https://tech-insider.org/kimi-k3-vs-qwen3-8-max-vs-glm-5-2-2026",
   "title": "Kimi K3 vs Qwen3.8-Max vs GLM-5.2 Compared [2026]",
   "published_at": "2026-08-21T12:13:10",
   "organization": "Zhipu AI",
   "model": null,
   "release_date": "2026-06-17",
   "access_type": null,
   "license": null,
   "significance": "A contender matching a 1-million-token context window and ranked as the strongest open-source model in July 2026.",
   "organization_evidence": [
    "# Kimi K3 vs Qwen3.8-Max vs GLM-5.2: $10.60 Gap [2026]",
    "Zhipu AI (operating internationally as Z.ai) had already shipped **GLM-5.2** on June 17, 2026."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Kimi K3 vs Qwen3.8-Max vs GLM-5.2: $10.60 Gap [2026]",
    "Zhipu AI (operating internationally as Z.ai) had already shipped **GLM-5.2** on June 17, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Kimi K3 vs Qwen3.8-Max vs GLM-5.2: $10.60 Gap [2026]",
    "## GLM-5.2: Zhipu AI\u2019s MIT-Licensed Contender",
    "GLM-5.2 matches the 1-million-token context window of its rivals."
   ]
  },
  {
   "url": "https://venturebeat.com/technology/xiaomis-new-open-source-agentic-ai-coding-harness-mimo-code-beats-claude-code-at-ultra-long-200-step-tasks",
   "title": "Xiaomi's new open source, agentic AI coding harness MiMo Code beats Claude Code at ultra-long, 200+ step tasks",
   "published_at": "2026-06-11T23:17:06",
   "organization": "Xiaomi",
   "model": null,
   "release_date": "2026-06-10",
   "access_type": null,
   "license": null,
   "significance": "A terminal-native AI coding assistant that outperforms Anthropic's Claude Code on key agentic coding benchmarks, especially on long-horizon, multi-step tasks.",
   "organization_evidence": [
    "Xiaomi's MiMo AI team has [open-sourced MiMo Code V0.1.0](https://mimo.xiaomi.com/blog/mimo-code-long-horizon), a terminal-native AI coding assistant that the Chinese electronics giant says outperforms Anthropic's Claude Code on key agentic coding benchmarks, especially on long-horizon, multi-step tasks (200+ steps) \u2014 at least, according to its own internal beta release and survey of 576 developers."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The release was announced June 10, 2026 in a post on the social network X from the official [@XiaomiMiMo account](https://x.com/XiaomiMiMo/status/2064799879352959085), which described the tool as \"more than an AI coding assistant in your terminal \u2014 it's the smartest coding partner you'll ever work with.\"",
    "Xiaomi's MiMo AI team has [open-sourced MiMo Code V0.1.0](https://mimo.xiaomi.com/blog/mimo-code-long-horizon), a terminal-native AI coding assistant that the Chinese electronics giant says outperforms Anthropic's Claude Code on key agentic coding benchmarks, especially on long-horizon, multi-step tasks (200+ steps) \u2014 at least, according to its own internal beta release and survey of 576 developers."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Xiaomi's MiMo AI team has [open-sourced MiMo Code V0.1.0](https://mimo.xiaomi.com/blog/mimo-code-long-horizon), a terminal-native AI coding assistant that the Chinese electronics giant says outperforms Anthropic's Claude Code on key agentic coding benchmarks, especially on long-horizon, multi-step tasks (200+ steps) \u2014 at least, according to its own internal beta release and survey of 576 developers."
   ]
  },
  {
   "url": "https://venturebeat.com/technology/xiaomis-new-open-source-agentic-ai-coding-harness-mimo-code-beats-claude-code-at-ultra-long-200-step-tasks",
   "title": "Xiaomi's new open source, agentic AI coding harness MiMo Code beats Claude Code at ultra-long, 200+ step tasks",
   "published_at": "2026-06-11T23:17:06",
   "organization": "Xiaomi",
   "model": null,
   "release_date": "2026-04-01",
   "access_type": null,
   "license": null,
   "significance": "A natively multimodal model, sparse mixture-of-experts design with 310 billion total parameters and a 1 million token context window.",
   "organization_evidence": [
    "As [VentureBeat reported when the MiMo-V2.5 family launched in April,](https://venturebeat.com/technology/open-source-xiaomi-mimo-v2-5-and-v2-5-pro-are-among-the-most-efficient-and-affordable-at-agentic-claw-tasks) the models are MIT-licensed and among the most efficient and affordable available for agentic tasks."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "As [VentureBeat reported when the MiMo-V2.5 family launched in April,](https://venturebeat.com/technology/open-source-xiaomi-mimo-v2-5-and-v2-5-pro-are-among-the-most-efficient-and-affordable-at-agentic-claw-tasks) the models are MIT-licensed and among the most efficient and affordable available for agentic tasks."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "As [VentureBeat reported when the MiMo-V2.5 family launched in April,](https://venturebeat.com/technology/open-source-xiaomi-mimo-v2-5-and-v2-5-pro-are-among-the-most-efficient-and-affordable-at-agentic-claw-tasks) the models are MIT-licensed and among the most efficient and affordable available for agentic tasks."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/kimi-k3",
   "title": "Kimi K3 (max) - Intelligence, Performance & Price Analysis | Artificial Analysis",
   "published_at": null,
   "organization": "Kimi",
   "model": null,
   "release_date": "2026-07-16",
   "access_type": null,
   "license": null,
   "significance": "A Mixture of Experts (MoE) reasoning model with 2.8 trillion total parameters (104 billion active), supporting text and image input and a 1M token context window.",
   "organization_evidence": [
    "Kimi K3 (max) Intelligence, Performance & Price Analysis",
    "Kimi K3 (max) was created by Kimi."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Released July 2026",
    "Kimi K3 (max) Intelligence, Performance & Price Analysis",
    "Kimi K3 (max) was released on July 16, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Kimi K3 (max) Intelligence, Performance & Price Analysis",
    "Kimi K3 (max) is amongst the leading models in intelligence, but particularly expensive when comparing to other open weight models of similar size.",
    "Kimi K3 (max) has a context window of 1.0M tokens.",
    "Yes, Kimi K3 (max) is a reasoning model.",
    "Kimi K3 (max) is a Mixture of Experts (MoE) model with 2.8 trillion total parameters, but only 104 billion active parameters are used during inference."
   ]
  },
  {
   "url": "https://pondero.ai/news/2026-07-16-inkling-thinking-machines-lab",
   "title": "Thinking Machines Lab releases Inkling, a 975B open-weight MoE with 1M-token context and native audio",
   "published_at": "2026-07-16T00:00:00",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "A 975-billion-parameter mixture-of-experts model with 41B active parameters per forward pass, a 1-million-token context window, and native audio understanding.",
   "organization_evidence": [
    "## Thinking Machines Lab releases Inkling, a 975B open-weight MoE with 1M-token context and native audio",
    "Mira Murati's Thinking Machines Lab released Inkling on July 15, 2026: a 975-billion-parameter mixture-of-experts model with 41B active parameters per forward pass, a 1-million-token context window, and native audio understanding."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## Thinking Machines Lab releases Inkling, a 975B open-weight MoE with 1M-token context and native audio",
    "Mira Murati's Thinking Machines Lab released Inkling on July 15, 2026: a 975-billion-parameter mixture-of-experts model with 41B active parameters per forward pass, a 1-million-token context window, and native audio understanding."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## Thinking Machines Lab releases Inkling, a 975B open-weight MoE with 1M-token context and native audio",
    "Mira Murati's Thinking Machines Lab released Inkling on July 15, 2026: a 975-billion-parameter mixture-of-experts model with 41B active parameters per forward pass, a 1-million-token context window, and native audio understanding."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/glm-5-2?aa-briefcase=aa-briefcase-rubric-score",
   "title": "GLM-5.2 (max) - Intelligence, Performance & Price Analysis | Artificial Analysis",
   "published_at": null,
   "organization": "Z AI",
   "model": null,
   "release_date": "2026-06-16",
   "access_type": null,
   "license": null,
   "significance": "amongst the leading models in intelligence with a 1M tokens context window and 753B parameters",
   "organization_evidence": [
    "GLM-5.2 (max) Intelligence, Performance & Price Analysis",
    "GLM-5.2 (max) was created by Z AI."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "GLM-5.2 (max) Intelligence, Performance & Price Analysis",
    "GLM-5.2 (max) was released on June 16, 2026.",
    "Released June 2026"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "GLM-5.2 (max) Intelligence, Performance & Price Analysis",
    "GLM-5.2 (max) is amongst the leading models in intelligence, but particularly expensive when comparing to other open weight models of similar size.",
    "The model supports text input, outputs text, and has a 1M tokens context window.",
    "GLM-5.2 (max) has 753 billion parameters (40 billion active)."
   ]
  },
  {
   "url": "https://micronomicon.com/ai-tooling/revealing-china-s-ai-launch-strategy-four-frontier-models-in-eight-weeks",
   "title": "Revealing China\u2019s AI Launch Strategy: Four Frontier Models In Eight Weeks - Micronomicon",
   "published_at": "2026-07-15T06:14:28",
   "organization": "DeepSeek",
   "model": "DeepSeek V4",
   "release_date": "2026-04-24",
   "access_type": "open-weight",
   "license": "MIT",
   "significance": "1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor",
   "organization_evidence": [
    "This rapid output marks a transition from a single-lab focus to a broader, multi-lab approach involving organizations such as DeepSeek, Z.ai, Moonshot, and Alibaba, each with distinct strategic focuses.",
    "Revealing China\u2019s AI Launch Strategy: Four Frontier Models In Eight Weeks"
   ],
   "model_evidence": [
    "Between April 24 and mid-June 2026, Chinese labs launched four major open-weight models: DeepSeek V4 on April 24, MiniMax M3 on June 1, and Kimi K2.7-Code and GLM-5.2 within days of each other in mid-June.",
    "Revealing China\u2019s AI Launch Strategy: Four Frontier Models In Eight Weeks"
   ],
   "release_date_evidence": [
    "Between April 24 and mid-June 2026, Chinese labs launched four major open-weight models: DeepSeek V4 on April 24, MiniMax M3 on June 1, and Kimi K2.7-Code and GLM-5.2 within days of each other in mid-June.",
    "Revealing China\u2019s AI Launch Strategy: Four Frontier Models In Eight Weeks"
   ],
   "access_type_evidence": [
    "Between April 24 and mid-June 2026, Chinese labs launched four major open-weight models: DeepSeek V4 on April 24, MiniMax M3 on June 1, and Kimi K2.7-Code and GLM-5.2 within days of each other in mid-June.",
    "Revealing China\u2019s AI Launch Strategy: Four Frontier Models In Eight Weeks"
   ],
   "license_evidence": [
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor",
    "### The production line \u2014 spring 2026"
   ],
   "significance_evidence": [
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor",
    "### The production line \u2014 spring 2026"
   ]
  },
  {
   "url": "https://micronomicon.com/ai-tooling/revealing-china-s-ai-launch-strategy-four-frontier-models-in-eight-weeks",
   "title": "Revealing China\u2019s AI Launch Strategy: Four Frontier Models In Eight Weeks - Micronomicon",
   "published_at": "2026-07-15T06:14:28",
   "organization": "MiniMax",
   "model": "MiniMax M3",
   "release_date": "2026-06-01",
   "access_type": "open-weight",
   "license": "modified-MIT",
   "significance": "cheap 1M-token context, native multimodal, modified-MIT",
   "organization_evidence": [
    "Between April 24 and mid-June 2026, Chinese labs launched four major open-weight models: DeepSeek V4 on April 24, MiniMax M3 on June 1, and Kimi K2.7-Code and GLM-5.2 within days of each other in mid-June.",
    "Revealing China\u2019s AI Launch Strategy: Four Frontier Models In Eight Weeks"
   ],
   "model_evidence": [
    "Between April 24 and mid-June 2026, Chinese labs launched four major open-weight models: DeepSeek V4 on April 24, MiniMax M3 on June 1, and Kimi K2.7-Code and GLM-5.2 within days of each other in mid-June.",
    "Revealing China\u2019s AI Launch Strategy: Four Frontier Models In Eight Weeks"
   ],
   "release_date_evidence": [
    "Between April 24 and mid-June 2026, Chinese labs launched four major open-weight models: DeepSeek V4 on April 24, MiniMax M3 on June 1, and Kimi K2.7-Code and GLM-5.2 within days of each other in mid-June.",
    "Revealing China\u2019s AI Launch Strategy: Four Frontier Models In Eight Weeks"
   ],
   "access_type_evidence": [
    "Between April 24 and mid-June 2026, Chinese labs launched four major open-weight models: DeepSeek V4 on April 24, MiniMax M3 on June 1, and Kimi K2.7-Code and GLM-5.2 within days of each other in mid-June.",
    "Revealing China\u2019s AI Launch Strategy: Four Frontier Models In Eight Weeks"
   ],
   "license_evidence": [
    "**MiniMax M3**cheap 1M-token context, native multimodal, modified-MIT",
    "### The production line \u2014 spring 2026"
   ],
   "significance_evidence": [
    "**MiniMax M3**cheap 1M-token context, native multimodal, modified-MIT",
    "### The production line \u2014 spring 2026"
   ]
  },
  {
   "url": "https://micronomicon.com/ai-tooling/revealing-china-s-ai-launch-strategy-four-frontier-models-in-eight-weeks",
   "title": "Revealing China\u2019s AI Launch Strategy: Four Frontier Models In Eight Weeks - Micronomicon",
   "published_at": "2026-07-15T06:14:28",
   "organization": "Moonshot",
   "model": "Kimi K2.7-Code",
   "release_date": "2026-06-13",
   "access_type": "open-weight",
   "license": null,
   "significance": "agent-run specialist, ~30% fewer thinking tokens than K2.6",
   "organization_evidence": [
    "This rapid output marks a transition from a single-lab focus to a broader, multi-lab approach involving organizations such as DeepSeek, Z.ai, Moonshot, and Alibaba, each with distinct strategic focuses.",
    "Revealing China\u2019s AI Launch Strategy: Four Frontier Models In Eight Weeks"
   ],
   "model_evidence": [
    "Between April 24 and mid-June 2026, Chinese labs launched four major open-weight models: DeepSeek V4 on April 24, MiniMax M3 on June 1, and Kimi K2.7-Code and GLM-5.2 within days of each other in mid-June.",
    "Revealing China\u2019s AI Launch Strategy: Four Frontier Models In Eight Weeks"
   ],
   "release_date_evidence": [
    "Between April 24 and mid-June 2026, Chinese labs launched four major open-weight models: DeepSeek V4 on April 24, MiniMax M3 on June 1, and Kimi K2.7-Code and GLM-5.2 within days of each other in mid-June.",
    "Revealing China\u2019s AI Launch Strategy: Four Frontier Models In Eight Weeks"
   ],
   "access_type_evidence": [
    "Between April 24 and mid-June 2026, Chinese labs launched four major open-weight models: DeepSeek V4 on April 24, MiniMax M3 on June 1, and Kimi K2.7-Code and GLM-5.2 within days of each other in mid-June.",
    "Revealing China\u2019s AI Launch Strategy: Four Frontier Models In Eight Weeks"
   ],
   "license_evidence": null,
   "significance_evidence": [
    "**Kimi K2.7-Code (Moonshot)**agent-run specialist, ~30% fewer thinking tokens than K2.6",
    "### The production line \u2014 spring 2026"
   ]
  },
  {
   "url": "https://micronomicon.com/ai-tooling/revealing-china-s-ai-launch-strategy-four-frontier-models-in-eight-weeks",
   "title": "Revealing China\u2019s AI Launch Strategy: Four Frontier Models In Eight Weeks - Micronomicon",
   "published_at": "2026-07-15T06:14:28",
   "organization": "Z.ai",
   "model": "GLM-5.2",
   "release_date": "2026-06-13",
   "access_type": "open-weight",
   "license": "MIT",
   "significance": "753B MoE, MIT, top open-weight on Artificial Analysis index",
   "organization_evidence": [
    "This rapid output marks a transition from a single-lab focus to a broader, multi-lab approach involving organizations such as DeepSeek, Z.ai, Moonshot, and Alibaba, each with distinct strategic focuses.",
    "Revealing China\u2019s AI Launch Strategy: Four Frontier Models In Eight Weeks"
   ],
   "model_evidence": [
    "Between April 24 and mid-June 2026, Chinese labs launched four major open-weight models: DeepSeek V4 on April 24, MiniMax M3 on June 1, and Kimi K2.7-Code and GLM-5.2 within days of each other in mid-June.",
    "Revealing China\u2019s AI Launch Strategy: Four Frontier Models In Eight Weeks"
   ],
   "release_date_evidence": [
    "Between April 24 and mid-June 2026, Chinese labs launched four major open-weight models: DeepSeek V4 on April 24, MiniMax M3 on June 1, and Kimi K2.7-Code and GLM-5.2 within days of each other in mid-June.",
    "Revealing China\u2019s AI Launch Strategy: Four Frontier Models In Eight Weeks"
   ],
   "access_type_evidence": [
    "Between April 24 and mid-June 2026, Chinese labs launched four major open-weight models: DeepSeek V4 on April 24, MiniMax M3 on June 1, and Kimi K2.7-Code and GLM-5.2 within days of each other in mid-June.",
    "Revealing China\u2019s AI Launch Strategy: Four Frontier Models In Eight Weeks"
   ],
   "license_evidence": [
    "**GLM-5.2 (Z.ai)**753B MoE, MIT, top open-weight on Artificial Analysis index",
    "### The production line \u2014 spring 2026"
   ],
   "significance_evidence": [
    "**GLM-5.2 (Z.ai)**753B MoE, MIT, top open-weight on Artificial Analysis index",
    "### The production line \u2014 spring 2026"
   ]
  },
  {
   "url": "https://edgen.tech/ko/news/post/zai-confirms-ox-alpha-as-glm-model-with-usage-2x-deepseek",
   "title": "Z.AI confirms Ox Alpha as GLM model with usage 2x DeepSeek",
   "published_at": "2026-08-26T12:16:39",
   "organization": "Z.AI",
   "model": null,
   "release_date": "2026-08-26",
   "access_type": null,
   "license": null,
   "significance": "Ox Alpha's OpenRouter usage more than doubled DeepSeek's, the marketplace's biggest launch.",
   "organization_evidence": [
    "# Z.AI confirms Ox Alpha as GLM model with usage 2x DeepSeek",
    "The anonymous AI model that topped OpenRouter's usage charts within days is a new GLM-family release from China's Z.AI, which confirmed the launch and will open-source its weights."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Z.AI confirms Ox Alpha as GLM model with usage 2x DeepSeek",
    "[![Alex Nguyen](https://static.edwealth.ai/authors/alex-nguyen.png?w=32)\n## Alex Nguyen](https://www.edgen.tech/ko/author/news/alex-nguyen)\u00b7Aug 26 2026, 05:16",
    "The mystery AI model that swept to the top of OpenRouter's usage charts is a new GLM-family release from China's Z.AI, which confirmed the launch Wednesday and said it will release the model's weights tonight."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Z.AI confirms Ox Alpha as GLM model with usage 2x DeepSeek",
    "Ox Alpha's OpenRouter usage more than doubled DeepSeek's, the marketplace's biggest launch."
   ]
  },
  {
   "url": "https://outlookbusiness.com/deeptech/artificial-intelligence/openai-unveils-open-weight-models-that-run-on-single-nvidia-gpu",
   "title": "OpenAI Unveils Open-Weight Models that Run on Single Nvidia GPU",
   "published_at": "2025-08-06T07:26:17",
   "organization": "OpenAI",
   "model": null,
   "release_date": "2026-05-01",
   "access_type": null,
   "license": null,
   "significance": "capable of advanced reasoning on a single GPU or 16 GB RAM laptop",
   "organization_evidence": [
    "# OpenAI Unveils Open-Weight Models that Run on Single Nvidia GPU",
    "OpenAI releases two Apache 2.0\u2013licensed open-weight models, gpt-oss-120b and gpt-oss-20b, capable of advanced reasoning on a single GPU or 16 GB RAM laptop."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# OpenAI Unveils Open-Weight Models that Run on Single Nvidia GPU",
    "1 May 2026",
    "Get the latest issue of Outlook Business"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# OpenAI Unveils Open-Weight Models that Run on Single Nvidia GPU",
    "OpenAI releases two Apache 2.0\u2013licensed open-weight models, gpt-oss-120b and gpt-oss-20b, capable of advanced reasoning on a single GPU or 16 GB RAM laptop."
   ]
  },
  {
   "url": "https://thenextweb.com/news/poolside-laguna-s-2-1-open-weight-coding-model",
   "title": "Poolside releases Laguna S 2.1, the open-weight coding model pitched as the West\u2019s answer to DeepSeek and Qwen",
   "published_at": "2026-07-21T19:49:20",
   "organization": "Poolside",
   "model": null,
   "release_date": "2026-07-21",
   "access_type": null,
   "license": null,
   "significance": "118-billion-parameter open-weight model built for agentic coding that matches or exceeds models several times its size",
   "organization_evidence": [
    "# Poolside releases Laguna S 2.1, the open-weight coding model pitched as the West\u2019s answer to DeepSeek and Qwen",
    "Poolside releases Laguna S 2.1, the open-weight coding model pitched as the West\u2019s answer to DeepSeek and Qwen"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Poolside releases Laguna S 2.1, the open-weight coding model pitched as the West\u2019s answer to DeepSeek and Qwen",
    "[Poolside](https://poolside.ai/blog/introducing-laguna-s-2-1) has released Laguna S 2.1, a[ 118-billion-parameter open-weight model](https://www.globenewswire.com/news-release/2026/07/21/3330818/0/en/Poolside-releases-Laguna-S-2-1-the-West-s-most-capable-open-weight-model.html) built for agentic coding that the San Francisco startup says matches or exceeds models several times its size."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Poolside releases Laguna S 2.1, the open-weight coding model pitched as the West\u2019s answer to DeepSeek and Qwen",
    "[Poolside](https://poolside.ai/blog/introducing-laguna-s-2-1) has released Laguna S 2.1, a[ 118-billion-parameter open-weight model](https://www.globenewswire.com/news-release/2026/07/21/3330818/0/en/Poolside-releases-Laguna-S-2-1-the-West-s-most-capable-open-weight-model.html) built for agentic coding that the San Francisco startup says matches or exceeds models several times its size."
   ]
  },
  {
   "url": "https://huggingface.co/blog/balaatdell/under-the-hood-trinity-large-thinking-architecture",
   "title": "Under The Hood : Trinity-Large-Thinking Disected",
   "published_at": "2026-05-01T00:42:52",
   "organization": "Arcee AI",
   "model": null,
   "release_date": "2026-04-01",
   "access_type": null,
   "license": null,
   "significance": "a 398-billion parameter sparse Mixture-of-Experts model that activates ~13 billion parameters per token, uses a hybrid sliding-window and full-attention design across 60 layers",
   "organization_evidence": [
    "Arcee AI's Trinity-Large-Thinking is a 398-billion parameter sparse Mixture-of-Experts model that activates ~13 billion parameters per token, uses a hybrid sliding-window and full-attention design across 60 layers, and ships under Apache 2.0."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "It was released on April 1, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Arcee AI's Trinity-Large-Thinking is a 398-billion parameter sparse Mixture-of-Experts model that activates ~13 billion parameters per token, uses a hybrid sliding-window and full-attention design across 60 layers, and ships under Apache 2.0."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/qwen3-6-27b",
   "title": "Qwen3.6 27B - Intelligence, Performance & Price Analysis | Artificial Analysis",
   "published_at": null,
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-04-22",
   "access_type": null,
   "license": null,
   "significance": "amongst the leading models in intelligence",
   "organization_evidence": [
    "Qwen3.6 27B (Reasoning) Intelligence, Performance & Price Analysis",
    "Qwen3.6 27B (Reasoning) was created by Alibaba."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Qwen3.6 27B (Reasoning) Intelligence, Performance & Price Analysis",
    "Qwen3.6 27B (Reasoning) was released on April 22, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Qwen3.6 27B (Reasoning) Intelligence, Performance & Price Analysis",
    "Qwen3.6 27B (Reasoning) is amongst the leading models in intelligence, but particularly expensive when comparing to other open weight models of similar size."
   ]
  },
  {
   "url": "https://cybermediacreations.com/china-s-ai-strategy-in-action-four-frontier-class-models-released-in-eight-weeks",
   "title": "China\u2019s AI Strategy In Action: Four Frontier-Class Models Released In Eight Weeks - Cyber Media Creations",
   "published_at": "2026-07-15T06:13:53",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor",
   "organization_evidence": [
    "### The production line \u2014 spring 2026"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The first, **DeepSeek V4**, was released on April 24, and quickly rose to the top of Chinese benchmarks, scoring 87 in the BenchLM rankings, just six points below the leading proprietary models.",
    "### The production line \u2014 spring 2026",
    "APR 24"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "This model features 1.6 trillion total parameters but activates only 49 billion per pass, making it one of the most cost-efficient frontier models available.",
    "### The production line \u2014 spring 2026",
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor"
   ]
  },
  {
   "url": "https://cybermediacreations.com/china-s-ai-strategy-in-action-four-frontier-class-models-released-in-eight-weeks",
   "title": "China\u2019s AI Strategy In Action: Four Frontier-Class Models Released In Eight Weeks - Cyber Media Creations",
   "published_at": "2026-07-15T06:13:53",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "cheap 1M-token context, native multimodal, modified-MIT",
   "organization_evidence": [
    "Following this, **MiniMax M3** was launched on June 1, and then **Kimi K2.7-Code** and **GLM-5.2** appeared within days of each other in mid-June.",
    "### The production line \u2014 spring 2026"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Following this, **MiniMax M3** was launched on June 1, and then **Kimi K2.7-Code** and **GLM-5.2** appeared within days of each other in mid-June.",
    "### The production line \u2014 spring 2026",
    "JUN 01"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "These models are notable for their licensing terms\u2014most are available under MIT-class licenses\u2014and their relatively low pricing, especially when hosted on Chinese infrastructure.",
    "### The production line \u2014 spring 2026",
    "**MiniMax M3**cheap 1M-token context, native multimodal, modified-MIT"
   ]
  },
  {
   "url": "https://cybermediacreations.com/china-s-ai-strategy-in-action-four-frontier-class-models-released-in-eight-weeks",
   "title": "China\u2019s AI Strategy In Action: Four Frontier-Class Models Released In Eight Weeks - Cyber Media Creations",
   "published_at": "2026-07-15T06:13:53",
   "organization": "Moonshot",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "agent-run specialist, ~30% fewer thinking tokens than K2.6",
   "organization_evidence": [
    "### The production line \u2014 spring 2026"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Following this, **MiniMax M3** was launched on June 1, and then **Kimi K2.7-Code** and **GLM-5.2** appeared within days of each other in mid-June.",
    "### The production line \u2014 spring 2026",
    "JUN 13"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The first, **DeepSeek V4**, was released on April 24, and quickly rose to the top of Chinese benchmarks, scoring 87 in the BenchLM rankings, just six points below the leading proprietary models.",
    "### The production line \u2014 spring 2026",
    "**Kimi K2.7-Code (Moonshot)**agent-run specialist, ~30% fewer thinking tokens than K2.6"
   ]
  },
  {
   "url": "https://cybermediacreations.com/china-s-ai-strategy-in-action-four-frontier-class-models-released-in-eight-weeks",
   "title": "China\u2019s AI Strategy In Action: Four Frontier-Class Models Released In Eight Weeks - Cyber Media Creations",
   "published_at": "2026-07-15T06:13:53",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-06-16",
   "access_type": null,
   "license": null,
   "significance": "753B MoE, MIT, top open-weight on Artificial Analysis index",
   "organization_evidence": [
    "### The production line \u2014 spring 2026"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Following this, **MiniMax M3** was launched on June 1, and then **Kimi K2.7-Code** and **GLM-5.2** appeared within days of each other in mid-June.",
    "### The production line \u2014 spring 2026",
    "JUN 13\u201316"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The first, **DeepSeek V4**, was released on April 24, and quickly rose to the top of Chinese benchmarks, scoring 87 in the BenchLM rankings, just six points below the leading proprietary models.",
    "### The production line \u2014 spring 2026",
    "**GLM-5.2 (Z.ai)**753B MoE, MIT, top open-weight on Artificial Analysis index"
   ]
  },
  {
   "url": "https://theearlierstuff.com/general/ai-breakthrough-what-the-weights-first-approach-reveals-about-machine-thinking",
   "title": "AI Breakthrough: What The Weights First Approach Reveals About Machine Thinking - The Earlier Stuff",
   "published_at": "2026-07-17T04:46:42",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "A Mixture-of-Experts model with 975 billion total parameters and 41 billion active parameters, natively multimodal supporting text, images and audio with a 1-million-token context window and configurable thinking-effort setting.",
   "organization_evidence": [
    "TL;DR",
    "Thinking Machines Lab released the full weights for its first foundation model, Inkling, before offering a closed API.",
    "**Thinking Machines Lab**, founded by former OpenAI technology chief **Mira Murati**, released the full weights for its first foundation model, **Inkling**, on July 15 under an Apache 2.0 license."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "TL;DR",
    "Thinking Machines Lab released the full weights for its first foundation model, Inkling, before offering a closed API.",
    "**Thinking Machines Lab**, founded by former OpenAI technology chief **Mira Murati**, released the full weights for its first foundation model, **Inkling**, on July 15 under an Apache 2.0 license.",
    "Thinking Machines Lab released Inkling\u2019s full model weights under Apache 2.0 on July 15, 2026, making ownership the starting point rather than a later option."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "TL;DR",
    "Thinking Machines Lab released the full weights for its first foundation model, Inkling, before offering a closed API.",
    "Inkling is a **Mixture-of-Experts model** with 975 billion total parameters and 41 billion active parameters, according to materials cited in the supplied reporting.",
    "It accepts **text, images and audio**, produces text, and supports a **1-million-token context window**."
   ]
  },
  {
   "url": "https://digitalapplied.com/blog/open-weight-model-wave-july-2026-momentum-tracker",
   "title": "July 2026 Open-Weight Wave: K3, Inkling and the New Gap",
   "published_at": "2026-07-16T23:00:00",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "Thinking Machines Lab's first foundational model \u2014 natively multimodal in, text out, trained on 45T tokens, weights on Hugging Face at launch.",
   "organization_evidence": [
    "Thinking Machines Lab's first foundational model \u2014 natively multimodal in, text out, trained on 45T tokens, weights on Hugging Face at launch.",
    "Jul 15 \u00b7 Shipped",
    "## 03 \u2014 InklingInkling: the US *re-entry*."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Jul 15 \u00b7 Shipped",
    "## 03 \u2014 InklingInkling: the US *re-entry*."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Thinking Machines Lab's first foundational model \u2014 natively multimodal in, text out, trained on 45T tokens, weights on Hugging Face at launch.",
    "Jul 15 \u00b7 Shipped",
    "## 03 \u2014 InklingInkling: the US *re-entry*."
   ]
  },
  {
   "url": "https://thebriefscript.com/open-source-ai-models-2026",
   "title": "Open Source AI Models 2026: The Complete Release Tracker",
   "published_at": "2026-07-11T19:41:39",
   "organization": "DeepSeek",
   "model": "DeepSeek V4",
   "release_date": "2026-04-24",
   "access_type": "open-source",
   "license": "MIT",
   "significance": "1M token context, trained on 32T-plus tokens",
   "organization_evidence": [
    "Open Source AI Model Releases: The 2026 Tracker",
    "Table 1: Major 2026 Open-Weight Model Releases and Licensing",
    "| **Model** | **Lab** | **Release window** | **License** | **Notable spec** |",
    "| DeepSeek V4 | DeepSeek | Public preview, April 24, 2026 | MIT | 1M token context, trained on 32T-plus tokens |"
   ],
   "model_evidence": [
    "Open Source AI Model Releases: The 2026 Tracker",
    "Table 1: Major 2026 Open-Weight Model Releases and Licensing",
    "| **Model** | **Lab** | **Release window** | **License** | **Notable spec** |",
    "| DeepSeek V4 | DeepSeek | Public preview, April 24, 2026 | MIT | 1M token context, trained on 32T-plus tokens |"
   ],
   "release_date_evidence": [
    "Open Source AI Model Releases: The 2026 Tracker",
    "Table 1: Major 2026 Open-Weight Model Releases and Licensing",
    "| **Model** | **Lab** | **Release window** | **License** | **Notable spec** |",
    "| DeepSeek V4 | DeepSeek | Public preview, April 24, 2026 | MIT | 1M token context, trained on 32T-plus tokens |"
   ],
   "access_type_evidence": [
    "Open Source AI Model Releases: The 2026 Tracker",
    "DeepSeek and Alibaba's Qwen have captured a growing majority of open-source downloads since.",
    "Table 1: Major 2026 Open-Weight Model Releases and Licensing",
    "| **Model** | **Lab** | **Release window** | **License** | **Notable spec** |",
    "| DeepSeek V4 | DeepSeek | Public preview, April 24, 2026 | MIT | 1M token context, trained on 32T-plus tokens |"
   ],
   "license_evidence": [
    "Open Source AI Model Releases: The 2026 Tracker",
    "Table 1: Major 2026 Open-Weight Model Releases and Licensing",
    "| **Model** | **Lab** | **Release window** | **License** | **Notable spec** |",
    "| DeepSeek V4 | DeepSeek | Public preview, April 24, 2026 | MIT | 1M token context, trained on 32T-plus tokens |"
   ],
   "significance_evidence": [
    "Open Source AI Model Releases: The 2026 Tracker",
    "Table 1: Major 2026 Open-Weight Model Releases and Licensing",
    "| **Model** | **Lab** | **Release window** | **License** | **Notable spec** |",
    "| DeepSeek V4 | DeepSeek | Public preview, April 24, 2026 | MIT | 1M token context, trained on 32T-plus tokens |"
   ]
  },
  {
   "url": "https://thebriefscript.com/open-source-ai-models-2026",
   "title": "Open Source AI Models 2026: The Complete Release Tracker",
   "published_at": "2026-07-11T19:41:39",
   "organization": "Google",
   "model": "Gemma 4",
   "release_date": "2026-04",
   "access_type": "open source",
   "license": "Apache 2.0",
   "significance": "Native audio support, 256K context",
   "organization_evidence": [
    "Open Source AI Model Releases: The 2026 Tracker",
    "Table 1: Major 2026 Open-Weight Model Releases and Licensing",
    "| **Model** | **Lab** | **Release window** | **License** | **Notable spec** |",
    "| Gemma 4 | Google | April 2026 | Apache 2.0 | Native audio support, 256K context |"
   ],
   "model_evidence": [
    "Open Source AI Model Releases: The 2026 Tracker",
    "Table 1: Major 2026 Open-Weight Model Releases and Licensing",
    "| **Model** | **Lab** | **Release window** | **License** | **Notable spec** |",
    "| Gemma 4 | Google | April 2026 | Apache 2.0 | Native audio support, 256K context |"
   ],
   "release_date_evidence": [
    "Open Source AI Model Releases: The 2026 Tracker",
    "Table 1: Major 2026 Open-Weight Model Releases and Licensing",
    "| **Model** | **Lab** | **Release window** | **License** | **Notable spec** |",
    "| Gemma 4 | Google | April 2026 | Apache 2.0 | Native audio support, 256K context |"
   ],
   "access_type_evidence": [
    "Open Source AI Model Releases: The 2026 Tracker",
    "Table 1: Major 2026 Open-Weight Model Releases and Licensing",
    "| **Model** | **Lab** | **Release window** | **License** | **Notable spec** |",
    "| Gemma 4 | Google | April 2026 | Apache 2.0 | Native audio support, 256K context |"
   ],
   "license_evidence": [
    "Open Source AI Model Releases: The 2026 Tracker",
    "Table 1: Major 2026 Open-Weight Model Releases and Licensing",
    "| **Model** | **Lab** | **Release window** | **License** | **Notable spec** |",
    "| Gemma 4 | Google | April 2026 | Apache 2.0 | Native audio support, 256K context |"
   ],
   "significance_evidence": [
    "Open Source AI Model Releases: The 2026 Tracker",
    "Table 1: Major 2026 Open-Weight Model Releases and Licensing",
    "| **Model** | **Lab** | **Release window** | **License** | **Notable spec** |",
    "| Gemma 4 | Google | April 2026 | Apache 2.0 | Native audio support, 256K context |"
   ]
  },
  {
   "url": "https://thebriefscript.com/open-source-ai-models-2026",
   "title": "Open Source AI Models 2026: The Complete Release Tracker",
   "published_at": "2026-07-11T19:41:39",
   "organization": "Alibaba",
   "model": "Qwen 3.5 and successors",
   "release_date": "2026",
   "access_type": "open-source",
   "license": "Apache 2.0, open line",
   "significance": "Reported over 50% of global open-source downloads",
   "organization_evidence": [
    "Open Source AI Model Releases: The 2026 Tracker",
    "Table 1: Major 2026 Open-Weight Model Releases and Licensing",
    "| **Model** | **Lab** | **Release window** | **License** | **Notable spec** |",
    "| Qwen 3.5 and successors | Alibaba | Rolling releases through 2026 | Apache 2.0, open line | Reported over 50% of global open-source downloads |"
   ],
   "model_evidence": [
    "Open Source AI Model Releases: The 2026 Tracker",
    "Table 1: Major 2026 Open-Weight Model Releases and Licensing",
    "| **Model** | **Lab** | **Release window** | **License** | **Notable spec** |",
    "| Qwen 3.5 and successors | Alibaba | Rolling releases through 2026 | Apache 2.0, open line | Reported over 50% of global open-source downloads |"
   ],
   "release_date_evidence": [
    "Open Source AI Model Releases: The 2026 Tracker",
    "Table 1: Major 2026 Open-Weight Model Releases and Licensing",
    "| **Model** | **Lab** | **Release window** | **License** | **Notable spec** |",
    "| Qwen 3.5 and successors | Alibaba | Rolling releases through 2026 | Apache 2.0, open line | Reported over 50% of global open-source downloads |"
   ],
   "access_type_evidence": [
    "Open Source AI Model Releases: The 2026 Tracker",
    "Table 1: Major 2026 Open-Weight Model Releases and Licensing",
    "| **Model** | **Lab** | **Release window** | **License** | **Notable spec** |",
    "| Qwen 3.5 and successors | Alibaba | Rolling releases through 2026 | Apache 2.0, open line | Reported over 50% of global open-source downloads |"
   ],
   "license_evidence": [
    "Open Source AI Model Releases: The 2026 Tracker",
    "Table 1: Major 2026 Open-Weight Model Releases and Licensing",
    "| **Model** | **Lab** | **Release window** | **License** | **Notable spec** |",
    "| Qwen 3.5 and successors | Alibaba | Rolling releases through 2026 | Apache 2.0, open line | Reported over 50% of global open-source downloads |"
   ],
   "significance_evidence": [
    "Open Source AI Model Releases: The 2026 Tracker",
    "Table 1: Major 2026 Open-Weight Model Releases and Licensing",
    "| **Model** | **Lab** | **Release window** | **License** | **Notable spec** |",
    "| Qwen 3.5 and successors | Alibaba | Rolling releases through 2026 | Apache 2.0, open line | Reported over 50% of global open-source downloads |"
   ]
  },
  {
   "url": "https://thebriefscript.com/open-source-ai-models-2026",
   "title": "Open Source AI Models 2026: The Complete Release Tracker",
   "published_at": "2026-07-11T19:41:39",
   "organization": "Mistral",
   "model": "Mistral Large 3",
   "release_date": "2026",
   "access_type": "open-weight",
   "license": "Apache 2.0",
   "significance": "Sparse MoE, European data residency",
   "organization_evidence": [
    "Open Source AI Model Releases: The 2026 Tracker",
    "Table 1: Major 2026 Open-Weight Model Releases and Licensing",
    "| **Model** | **Lab** | **Release window** | **License** | **Notable spec** |",
    "| Mistral Large 3 | Mistral | 2026 | Apache 2.0 | Sparse MoE, European data residency |"
   ],
   "model_evidence": [
    "Open Source AI Model Releases: The 2026 Tracker",
    "Table 1: Major 2026 Open-Weight Model Releases and Licensing",
    "| **Model** | **Lab** | **Release window** | **License** | **Notable spec** |",
    "| Mistral Large 3 | Mistral | 2026 | Apache 2.0 | Sparse MoE, European data residency |"
   ],
   "release_date_evidence": [
    "Open Source AI Model Releases: The 2026 Tracker",
    "Table 1: Major 2026 Open-Weight Model Releases and Licensing",
    "| **Model** | **Lab** | **Release window** | **License** | **Notable spec** |",
    "| Mistral Large 3 | Mistral | 2026 | Apache 2.0 | Sparse MoE, European data residency |"
   ],
   "access_type_evidence": [
    "Open Source AI Model Releases: The 2026 Tracker",
    "Table 1: Major 2026 Open-Weight Model Releases and Licensing",
    "| **Model** | **Lab** | **Release window** | **License** | **Notable spec** |",
    "| Mistral Large 3 | Mistral | 2026 | Apache 2.0 | Sparse MoE, European data residency |"
   ],
   "license_evidence": [
    "Open Source AI Model Releases: The 2026 Tracker",
    "Table 1: Major 2026 Open-Weight Model Releases and Licensing",
    "| **Model** | **Lab** | **Release window** | **License** | **Notable spec** |",
    "| Mistral Large 3 | Mistral | 2026 | Apache 2.0 | Sparse MoE, European data residency |"
   ],
   "significance_evidence": [
    "Open Source AI Model Releases: The 2026 Tracker",
    "Table 1: Major 2026 Open-Weight Model Releases and Licensing",
    "| **Model** | **Lab** | **Release window** | **License** | **Notable spec** |",
    "| Mistral Large 3 | Mistral | 2026 | Apache 2.0 | Sparse MoE, European data residency |"
   ]
  },
  {
   "url": "https://ainewsbank.com/posts/thinking-machines-inkling-first-model-apache-2-weights",
   "title": "Thinking Machines Releases Inkling, Its First Model, With Apache 2.0 Weights on Day One",
   "published_at": "2026-07-15T00:00:00",
   "organization": "Thinking Machines",
   "model": "Inkling",
   "release_date": "2026-07-15",
   "access_type": "open weight",
   "license": "Apache 2.0",
   "significance": "a 975B-total/41B-active mixture-of-experts model with full weights on Hugging Face under Apache 2.0 from day one",
   "organization_evidence": [
    "Changes confirmed high confidence",
    "Thinking Machines Releases Inkling, Its First Model, With Apache 2.0 Weights on Day One",
    "Thinking Machines Lab, Mira Murati's startup, released its first model on July 15, 2026: Inkling, a 975B-total/41B-active mixture-of-experts model with full weights on Hugging Face under Apache 2.0 from day one \u2014 no gating, no waitlist \u2014 per [Vectrel's analysis citing VentureBeat](https://www.vectrel.ai/blog/thinking-machines-inkling-kimi-k3-open-weight-ownership-strategy) and [NYU Shanghai's RITS write-up](https://rits.shanghai.nyu.edu/ai/thinking-machines-releases-inkling-its-first-open-weight-model/)."
   ],
   "model_evidence": [
    "Changes confirmed high confidence",
    "Thinking Machines Releases Inkling, Its First Model, With Apache 2.0 Weights on Day One",
    "Thinking Machines Lab, Mira Murati's startup, released its first model on July 15, 2026: Inkling, a 975B-total/41B-active mixture-of-experts model with full weights on Hugging Face under Apache 2.0 from day one \u2014 no gating, no waitlist \u2014 per [Vectrel's analysis citing VentureBeat](https://www.vectrel.ai/blog/thinking-machines-inkling-kimi-k3-open-weight-ownership-strategy) and [NYU Shanghai's RITS write-up](https://rits.shanghai.nyu.edu/ai/thinking-machines-releases-inkling-its-first-open-weight-model/)."
   ],
   "release_date_evidence": [
    "Changes confirmed high confidence",
    "Thinking Machines Releases Inkling, Its First Model, With Apache 2.0 Weights on Day One",
    "Thinking Machines Lab, Mira Murati's startup, released its first model on July 15, 2026: Inkling, a 975B-total/41B-active mixture-of-experts model with full weights on Hugging Face under Apache 2.0 from day one \u2014 no gating, no waitlist \u2014 per [Vectrel's analysis citing VentureBeat](https://www.vectrel.ai/blog/thinking-machines-inkling-kimi-k3-open-weight-ownership-strategy) and [NYU Shanghai's RITS write-up](https://rits.shanghai.nyu.edu/ai/thinking-machines-releases-inkling-its-first-open-weight-model/)."
   ],
   "access_type_evidence": [
    "Changes confirmed high confidence",
    "Thinking Machines Releases Inkling, Its First Model, With Apache 2.0 Weights on Day One",
    "Inkling arrives as the first credible US-lab open-weight release at near-frontier scale \u2014 a counterpoint to July's Chinese open-weight wave (LongCat-2.0, Hy3, Kimi K3) and to Meta's pivot to closed paid APIs with Muse Spark 1.1."
   ],
   "license_evidence": [
    "Changes confirmed high confidence",
    "Thinking Machines Releases Inkling, Its First Model, With Apache 2.0 Weights on Day One",
    "Thinking Machines Lab, Mira Murati's startup, released its first model on July 15, 2026: Inkling, a 975B-total/41B-active mixture-of-experts model with full weights on Hugging Face under Apache 2.0 from day one \u2014 no gating, no waitlist \u2014 per [Vectrel's analysis citing VentureBeat](https://www.vectrel.ai/blog/thinking-machines-inkling-kimi-k3-open-weight-ownership-strategy) and [NYU Shanghai's RITS write-up](https://rits.shanghai.nyu.edu/ai/thinking-machines-releases-inkling-its-first-open-weight-model/)."
   ],
   "significance_evidence": [
    "Changes confirmed high confidence",
    "Thinking Machines Releases Inkling, Its First Model, With Apache 2.0 Weights on Day One",
    "Thinking Machines Lab, Mira Murati's startup, released its first model on July 15, 2026: Inkling, a 975B-total/41B-active mixture-of-experts model with full weights on Hugging Face under Apache 2.0 from day one \u2014 no gating, no waitlist \u2014 per [Vectrel's analysis citing VentureBeat](https://www.vectrel.ai/blog/thinking-machines-inkling-kimi-k3-open-weight-ownership-strategy) and [NYU Shanghai's RITS write-up](https://rits.shanghai.nyu.edu/ai/thinking-machines-releases-inkling-its-first-open-weight-model/)."
   ]
  },
  {
   "url": "https://thequantumdispatch.com/articles/thinking-machines-inkling-open-weights-model",
   "title": "Thinking Machines Inkling: A 975B Open-Weights Model",
   "published_at": "2026-07-16T22:24:54",
   "organization": "Thinking Machines Lab",
   "model": "Inkling",
   "release_date": "2026-07-15",
   "access_type": "open-weights",
   "license": null,
   "significance": "975B total parameters with 41B active per token, natively multimodal, and a 1M-token context window",
   "organization_evidence": [
    "# Thinking Machines Inkling: A 975B Open-Weights Model",
    "Thinking Machines Lab released Inkling, a 975B-parameter open-weights multimodal model with 41B active per token and a 1M-token context window."
   ],
   "model_evidence": [
    "# Thinking Machines Inkling: A 975B Open-Weights Model",
    "Thinking Machines Lab released Inkling, a 975B-parameter open-weights multimodal model with 41B active per token and a 1M-token context window."
   ],
   "release_date_evidence": [
    "# Thinking Machines Inkling: A 975B Open-Weights Model"
   ],
   "access_type_evidence": [
    "# Thinking Machines Inkling: A 975B Open-Weights Model",
    "Thinking Machines Lab released Inkling, a 975B-parameter open-weights multimodal model with 41B active per token and a 1M-token context window."
   ],
   "license_evidence": null,
   "significance_evidence": [
    "# Thinking Machines Inkling: A 975B Open-Weights Model",
    "Thinking Machines Lab released Inkling, a 975B-parameter open-weights multimodal model with 41B active per token and a 1M-token context window."
   ]
  },
  {
   "url": "https://thequantumdispatch.com/articles/thinking-machines-inkling-open-weights-model",
   "title": "Thinking Machines Inkling: A 975B Open-Weights Model",
   "published_at": "2026-07-16T22:24:54",
   "organization": "NVIDIA",
   "model": "Alpamayo 2 Super",
   "release_date": "2026-08-04",
   "access_type": "open reasoning model",
   "license": "OpenMDW license",
   "significance": "34B open reasoning model for robotaxis with 360-degree perception",
   "organization_evidence": [
    "### NVIDIA Alpamayo 2 Super Ships an Open 34B AV Model",
    "NVIDIA released Alpamayo 2 Super for commercial use \u2014 a 34B open reasoning model for robotaxis with 360-degree perception and a permissive OpenMDW license."
   ],
   "model_evidence": [
    "### NVIDIA Alpamayo 2 Super Ships an Open 34B AV Model",
    "NVIDIA released Alpamayo 2 Super for commercial use \u2014 a 34B open reasoning model for robotaxis with 360-degree perception and a permissive OpenMDW license."
   ],
   "release_date_evidence": [
    "### NVIDIA Alpamayo 2 Super Ships an Open 34B AV Model",
    "Dr. Nova Chen\u2605Aug 4, 2026\u26056 min read"
   ],
   "access_type_evidence": [
    "### NVIDIA Alpamayo 2 Super Ships an Open 34B AV Model",
    "NVIDIA released Alpamayo 2 Super for commercial use \u2014 a 34B open reasoning model for robotaxis with 360-degree perception and a permissive OpenMDW license."
   ],
   "license_evidence": [
    "### NVIDIA Alpamayo 2 Super Ships an Open 34B AV Model",
    "NVIDIA released Alpamayo 2 Super for commercial use \u2014 a 34B open reasoning model for robotaxis with 360-degree perception and a permissive OpenMDW license."
   ],
   "significance_evidence": [
    "### NVIDIA Alpamayo 2 Super Ships an Open 34B AV Model",
    "NVIDIA released Alpamayo 2 Super for commercial use \u2014 a 34B open reasoning model for robotaxis with 360-degree perception and a permissive OpenMDW license."
   ]
  },
  {
   "url": "https://huggingface.co/blog/JetBrains/mellum2-launch",
   "title": "Introducing Mellum2: A 12B Mixture-of-Experts Model by JetBrains",
   "published_at": "2026-06-01T15:45:17",
   "organization": "JetBrains",
   "model": "Mellum2",
   "release_date": "2026-05",
   "access_type": "open",
   "license": "Apache 2.0",
   "significance": "12B-parameter Mixture-of-Experts model with 2.5B active parameters per token, optimized for low-latency text and code workloads",
   "organization_evidence": [
    "Download the model on Hugging Face: [https://huggingface.co/collections/JetBrains/mellum-2](https://huggingface.co/collections/JetBrains/mellum-2)",
    "[![Mellum 2 Evals](https://huggingface.co/JetBrains/Mellum2-12B-A2.5B-Thinking/resolve/main/mellum_evals_grid_1700.jpg)](https://huggingface.co/JetBrains/Mellum2-12B-A2.5B-Thinking/resolve/main/mellum_evals_grid_1700.jpg)"
   ],
   "model_evidence": [
    "- Mellum2 is a 12B-parameter Mixture-of-Experts model trained from scratch on natural language and code.",
    "Today we\u2019re releasing Mellum2, an open Mixture-of-Experts model optimized for low-latency text-and-code workloads."
   ],
   "release_date_evidence": [
    "- For architecture details, training setup, benchmarks, and evaluation methodology, read the full technical report: [https://arxiv.org/pdf/2605.31268](https://arxiv.org/pdf/2605.31268)",
    "Today we\u2019re releasing Mellum2, an open Mixture-of-Experts model optimized for low-latency text-and-code workloads."
   ],
   "access_type_evidence": [
    "Today we\u2019re releasing Mellum2, an open Mixture-of-Experts model optimized for low-latency text-and-code workloads.",
    "Because Mellum2 is open and efficient to serve, it can be deployed in self-hosted environments involving proprietary code or internal data."
   ],
   "license_evidence": [
    "- It is released under the Apache 2.0 license.",
    "| Model | Total parameters | Active parameters per token | Modality | License |",
    "| Mellum2 | 12B | 2.5B | Text and code | Apache 2.0 |"
   ],
   "significance_evidence": [
    "- Mellum2 is a 12B-parameter Mixture-of-Experts model trained from scratch on natural language and code.",
    "- The model activates only 2.5B parameters per token, making it efficient for high-throughput, low-latency inference. Mellum2 is can be used for routing, RAG, summarization, sub-agents, high-throughput coding features, and private deployments.",
    "Today we\u2019re releasing Mellum2, an open Mixture-of-Experts model optimized for low-latency text-and-code workloads."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/granite-4-1-3b",
   "title": "Granite 4.1 3B - Intelligence, Performance & Price Analysis | Artificial Analysis",
   "published_at": null,
   "organization": "IBM",
   "model": "Granite 4.1 3B",
   "release_date": "2026-04-29",
   "access_type": "open weights",
   "license": "Apache 2.0",
   "significance": "Granite 4.1 3B is amongst the leading models in intelligence and well priced when comparing to other open weight non-reasoning models of similar size.",
   "organization_evidence": [
    "Granite 4.1 3B Intelligence, Performance & Price Analysis",
    "Granite 4.1 3B was created by IBM."
   ],
   "model_evidence": [
    "Granite 4.1 3B Intelligence, Performance & Price Analysis",
    "Granite 4.1 3B is amongst the leading models in intelligence and well priced when comparing to other open weight non-reasoning models of similar size."
   ],
   "release_date_evidence": [
    "Granite 4.1 3B Intelligence, Performance & Price Analysis",
    "\u2022\nReleased April 2026",
    "Granite 4.1 3B was released on April 29, 2026."
   ],
   "access_type_evidence": [
    "Granite 4.1 3B Intelligence, Performance & Price Analysis",
    "Open weights model",
    "Granite 4.1 3B is amongst the leading models in intelligence and well priced when comparing to other open weight non-reasoning models of similar size."
   ],
   "license_evidence": [
    "Granite 4.1 3B Intelligence, Performance & Price Analysis",
    "| License | [Apache 2.0](https://www.apache.org/licenses/LICENSE-2.0) |",
    "Granite 4.1 3B is released under the Apache 2.0 license."
   ],
   "significance_evidence": [
    "Granite 4.1 3B Intelligence, Performance & Price Analysis",
    "Granite 4.1 3B is amongst the leading models in intelligence and well priced when comparing to other open weight non-reasoning models of similar size."
   ]
  },
  {
   "url": "https://micronomicon.com/ai-tooling/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - Micronomicon",
   "published_at": "2026-07-15T10:57:01",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor",
   "organization_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "**Chinese AI labs have released four frontier-class open models in roughly eight weeks,** including DeepSeek V4, MiniMax M3, Kimi K2.7-Code, and GLM-5.2.",
    "The Chinese open model ecosystem has expanded from a single lab two years ago to four distinct families: DeepSeek, Z.ai, Moonshot, and Alibaba, each with unique strategic focuses."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "Between late April and mid-June 2026, Chinese labs launched four major open-weight models: DeepSeek V4 on April 24, MiniMax M3 on June 1, and Kimi K2.7-Code and GLM-5.2 within days of each other in mid-June.",
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor"
   ]
  },
  {
   "url": "https://micronomicon.com/ai-tooling/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - Micronomicon",
   "published_at": "2026-07-15T10:57:01",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "cheap 1M-token context, native multimodal, modified-MIT",
   "organization_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "**Chinese AI labs have released four frontier-class open models in roughly eight weeks,** including DeepSeek V4, MiniMax M3, Kimi K2.7-Code, and GLM-5.2.",
    "The Chinese open model ecosystem has expanded from a single lab two years ago to four distinct families: DeepSeek, Z.ai, Moonshot, and Alibaba, each with unique strategic focuses."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "Between late April and mid-June 2026, Chinese labs launched four major open-weight models: DeepSeek V4 on April 24, MiniMax M3 on June 1, and Kimi K2.7-Code and GLM-5.2 within days of each other in mid-June.",
    "**MiniMax M3**cheap 1M-token context, native multimodal, modified-MIT"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "**MiniMax M3**cheap 1M-token context, native multimodal, modified-MIT"
   ]
  },
  {
   "url": "https://micronomicon.com/ai-tooling/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - Micronomicon",
   "published_at": "2026-07-15T10:57:01",
   "organization": "Moonshot",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "agent-run specialist, ~30% fewer thinking tokens than K2.6",
   "organization_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "**Chinese AI labs have released four frontier-class open models in roughly eight weeks,** including DeepSeek V4, MiniMax M3, Kimi K2.7-Code, and GLM-5.2.",
    "The Chinese open model ecosystem has expanded from a single lab two years ago to four distinct families: DeepSeek, Z.ai, Moonshot, and Alibaba, each with unique strategic focuses."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "Between late April and mid-June 2026, Chinese labs launched four major open-weight models: DeepSeek V4 on April 24, MiniMax M3 on June 1, and Kimi K2.7-Code and GLM-5.2 within days of each other in mid-June.",
    "**Kimi K2.7-Code (Moonshot)**agent-run specialist, ~30% fewer thinking tokens than K2.6"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "**Kimi K2.7-Code (Moonshot)**agent-run specialist, ~30% fewer thinking tokens than K2.6"
   ]
  },
  {
   "url": "https://micronomicon.com/ai-tooling/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - Micronomicon",
   "published_at": "2026-07-15T10:57:01",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "753B MoE, MIT, top open-weight on Artificial Analysis index",
   "organization_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "**Chinese AI labs have released four frontier-class open models in roughly eight weeks,** including DeepSeek V4, MiniMax M3, Kimi K2.7-Code, and GLM-5.2.",
    "The Chinese open model ecosystem has expanded from a single lab two years ago to four distinct families: DeepSeek, Z.ai, Moonshot, and Alibaba, each with unique strategic focuses."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "Between late April and mid-June 2026, Chinese labs launched four major open-weight models: DeepSeek V4 on April 24, MiniMax M3 on June 1, and Kimi K2.7-Code and GLM-5.2 within days of each other in mid-June.",
    "**GLM-5.2 (Z.ai)**753B MoE, MIT, top open-weight on Artificial Analysis index"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "**GLM-5.2 (Z.ai)**753B MoE, MIT, top open-weight on Artificial Analysis index"
   ]
  },
  {
   "url": "https://thequantumdispatch.com/articles/inkling-open-weights-model-thinking-machines",
   "title": "Inkling Is a 975B Open-Weights Model Under Apache 2.0",
   "published_at": "2026-07-21T22:48:53",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "975B-parameter model with 41B active, 1M-token context window, and native four-modality reasoning",
   "organization_evidence": [
    "# Inkling Is a 975B Open-Weights Model Under Apache 2.0",
    "Thinking Machines released Inkling, a 975B-parameter Apache 2.0 model with 41B active, a 1M-token context window, and native four-modality reasoning.",
    "Thinking Machines Lab released Inkling on July 15, 2026 \u2014 its first open-weights model, and by parameter count one of the largest ever published under a permissive license."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Inkling Is a 975B Open-Weights Model Under Apache 2.0",
    "Thinking Machines Lab released Inkling on July 15, 2026 \u2014 its first open-weights model, and by parameter count one of the largest ever published under a permissive license."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Inkling Is a 975B Open-Weights Model Under Apache 2.0",
    "Thinking Machines released Inkling, a 975B-parameter Apache 2.0 model with 41B active, a 1M-token context window, and native four-modality reasoning."
   ]
  },
  {
   "url": "https://llm-releases.com/coding-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-07-31",
   "access_type": null,
   "license": null,
   "significance": "retrained on a substantially improved post-training pipeline targeting coding, agents, reasoning, and tool use, with no change to the base architecture",
   "organization_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### DeepSeek-V4-Flash-0731\nAvailable\nDeepSeekOpen source\nThe production release of DeepSeek's V4-Flash tier \u2014 the April V4-Flash preview retrained on a substantially improved post-training pipeline targeting coding, agents, reasoning, and tool use, with no change to the base architecture."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Software and agentic coding",
    "Last updated Jul 31, 2026",
    "Coding LLM releases",
    "[### DeepSeek-V4-Flash-0731\nAvailable\nDeepSeekOpen source\nThe production release of DeepSeek's V4-Flash tier \u2014 the April V4-Flash preview retrained on a substantially improved post-training pipeline targeting coding, agents, reasoning, and tool use, with no change to the base architecture."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### DeepSeek-V4-Flash-0731\nAvailable\nDeepSeekOpen source\nThe production release of DeepSeek's V4-Flash tier \u2014 the April V4-Flash preview retrained on a substantially improved post-training pipeline targeting coding, agents, reasoning, and tool use, with no change to the base architecture."
   ]
  },
  {
   "url": "https://llm-releases.com/coding-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Ant Group (inclusionAI)",
   "model": null,
   "release_date": "2026-07-23",
   "access_type": null,
   "license": null,
   "significance": "efficiency-focused Mixture-of-Experts model, 124B total parameters activating only ~5.1B per token",
   "organization_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Ling-3.0-flash\nAvailable\nAnt Group (inclusionAI)Open weights\nAnt Group's efficiency-focused Mixture-of-Experts model, released July 23 2026 by its inclusionAI lab: 124B total parameters activating only ~5.1B per token (1/64 expert activation)."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Ling-3.0-flash\nAvailable\nAnt Group (inclusionAI)Open weights\nAnt Group's efficiency-focused Mixture-of-Experts model, released July 23 2026 by its inclusionAI lab: 124B total parameters activating only ~5.1B per token (1/64 expert activation)."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Ling-3.0-flash\nAvailable\nAnt Group (inclusionAI)Open weights\nAnt Group's efficiency-focused Mixture-of-Experts model, released July 23 2026 by its inclusionAI lab: 124B total parameters activating only ~5.1B per token (1/64 expert activation)."
   ]
  },
  {
   "url": "https://llm-releases.com/coding-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Poolside",
   "model": null,
   "release_date": "2026-05-22",
   "access_type": null,
   "license": null,
   "significance": "agentic-coding model and a scale-up of the Laguna XS family: a 118B-total / 8B-active Mixture-of-Experts",
   "organization_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Laguna S 2.1\nAvailable\nPoolsideOpen source\nPoolside's open-weight agentic-coding model and a scale-up of the Laguna XS family (same pre-training data as XS 2.1): a 118B-total / 8B-active Mixture-of-Experts that activates only ~6.8% of its parameters per token, giving larger-model behavior while staying cheap to serve, with a 1M-token context in both thinking and no-thinking modes."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "Trained in under nine weeks on 4,096 NVIDIA H200 GPUs (pre-training began 22 May 2026); first Poolside model with RL in FP8."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Laguna S 2.1\nAvailable\nPoolsideOpen source\nPoolside's open-weight agentic-coding model and a scale-up of the Laguna XS family (same pre-training data as XS 2.1): a 118B-total / 8B-active Mixture-of-Experts that activates only ~6.8% of its parameters per token, giving larger-model behavior while staying cheap to serve, with a 1M-token context in both thinking and no-thinking modes."
   ]
  },
  {
   "url": "https://llm-releases.com/coding-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-16",
   "access_type": null,
   "license": null,
   "significance": "flagship open-weight agentic model and the largest open model released to date: a 2.8T-parameter MoE",
   "organization_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Kimi K3\nAvailable\nMoonshot AIFrontierOpen weights\nMoonshot's flagship open-weight agentic model and the largest open model released to date: a 2.8T-parameter MoE (896 experts, 16 active per token) using Kimi Delta Attention and Attention Residuals, with native multimodal input and a 1M-token context."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "Launched via API on Jul 16, 2026 at $3/$15 per Mtok (cached input $0.30); full open weights published to Hugging Face on Jul 26, 2026 \u2014 a day ahead of the announced Jul 27 target \u2014 under a Modified MIT license, making it freely downloadable and self-hostable."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Kimi K3\nAvailable\nMoonshot AIFrontierOpen weights\nMoonshot's flagship open-weight agentic model and the largest open model released to date: a 2.8T-parameter MoE (896 experts, 16 active per token) using Kimi Delta Attention and Attention Residuals, with native multimodal input and a 1M-token context."
   ]
  },
  {
   "url": "https://llm-releases.com/coding-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "first model and the leading U.S. open-weights release: a natively multimodal Mixture-of-Experts with 975B total / 41B active parameters",
   "organization_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Inkling\nAvailable\nThinking Machines LabFrontierOpen source\nThinking Machines Lab's first model and the leading U.S. open-weights release: a natively multimodal Mixture-of-Experts with 975B total / 41B active parameters that reasons across text, image, and audio inputs and emits text."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Inkling\nAvailable\nThinking Machines LabFrontierOpen source\nThinking Machines Lab's first model and the leading U.S. open-weights release: a natively multimodal Mixture-of-Experts with 975B total / 41B active parameters that reasons across text, image, and audio inputs and emits text."
   ]
  },
  {
   "url": "https://llm-releases.com/coding-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Tencent Hunyuan",
   "model": null,
   "release_date": "2026-07-06",
   "access_type": null,
   "license": null,
   "significance": "a 295B-total / 21B-active Transformer MoE with an additional 3.8B multi-token-prediction (MTP) layer and a 256K-token context",
   "organization_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Hunyuan Hy3\nAvailable\nTencent HunyuanFrontierOpen source\nThe general-availability release of Tencent's third-generation Hunyuan (Hunyuan 3.0), officially launched and open-sourced on July 6, 2026 after April's \"Hy3 preview\"."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Hunyuan Hy3\nAvailable\nTencent HunyuanFrontierOpen source\nThe general-availability release of Tencent's third-generation Hunyuan (Hunyuan 3.0), officially launched and open-sourced on July 6, 2026 after April's \"Hy3 preview\"."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "A 295B-total / 21B-active Transformer MoE with an additional 3.8B multi-token-prediction (MTP) layer and a 256K-token context, offering three selectable inference modes that blend fast and slow thinking."
   ]
  },
  {
   "url": "https://llm-releases.com/coding-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "NVIDIA",
   "model": null,
   "release_date": "2026-07-06",
   "access_type": null,
   "license": null,
   "significance": "a compressed variant of Nemotron-3-Super-120B-A12B produced with \"Iterative Puzzle\", a post-training compression framework",
   "organization_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Nemotron-Labs-3-Puzzle-75B-A9B\nAvailable\nNVIDIAOpen weights\nA deployment-optimized open-weight model from NVIDIA, released July 6, 2026 \u2014 a compressed variant of Nemotron-3-Super-120B-A12B produced with \"Iterative Puzzle\", a post-training compression framework that jointly prunes MoE experts, active-parameter budget, and Mamba state to boost inference efficiency while preserving accuracy."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Nemotron-Labs-3-Puzzle-75B-A9B\nAvailable\nNVIDIAOpen weights\nA deployment-optimized open-weight model from NVIDIA, released July 6, 2026 \u2014 a compressed variant of Nemotron-3-Super-120B-A12B produced with \"Iterative Puzzle\", a post-training compression framework that jointly prunes MoE experts, active-parameter budget, and Mamba state to boost inference efficiency while preserving accuracy."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Nemotron-Labs-3-Puzzle-75B-A9B\nAvailable\nNVIDIAOpen weights\nA deployment-optimized open-weight model from NVIDIA, released July 6, 2026 \u2014 a compressed variant of Nemotron-3-Super-120B-A12B produced with \"Iterative Puzzle\", a post-training compression framework that jointly prunes MoE experts, active-parameter budget, and Mamba state to boost inference efficiency while preserving accuracy."
   ]
  },
  {
   "url": "https://llm-releases.com/coding-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Poolside",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "open-weight small coding model: a 33B-total / 3B-active Mixture-of-Experts built for agentic coding and long-horizon work on a local machine",
   "organization_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Laguna XS 2.1\nAvailable\nPoolsideOpen source\nPoolside's open-weight small coding model: a 33B-total / 3B-active Mixture-of-Experts built for agentic coding and long-horizon work on a local machine, served at 256K context."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Laguna XS 2.1\nAvailable\nPoolsideOpen source\nPoolside's open-weight small coding model: a 33B-total / 3B-active Mixture-of-Experts built for agentic coding and long-horizon work on a local machine, served at 256K context."
   ]
  },
  {
   "url": "https://llm-releases.com/coding-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Meituan (LongCat)",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "open-weight flagship: a 1.6-trillion-parameter Mixture-of-Experts model built for agentic coding",
   "organization_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### LongCat-2.0\nAvailable\nMeituan (LongCat)FrontierOpen source\nMeituan's open-weight flagship: a 1.6-trillion-parameter Mixture-of-Experts model (~48B active per token, dynamically routed between ~33B and ~56B) with a 1M-token context, built for agentic coding."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### LongCat-2.0\nAvailable\nMeituan (LongCat)FrontierOpen source\nMeituan's open-weight flagship: a 1.6-trillion-parameter Mixture-of-Experts model (~48B active per token, dynamically routed between ~33B and ~56B) with a 1M-token context, built for agentic coding."
   ]
  },
  {
   "url": "https://llm-releases.com/coding-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Tencent Hunyuan",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "rebuilt from scratch in ~90 days and open-sourced as the \"Hy3 preview\"",
   "organization_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Hunyuan Hy3-preview\nAvailable\nTencent HunyuanFrontierOpen weights\nTencent's third-generation Hunyuan, rebuilt from scratch in ~90 days and open-sourced as the \"Hy3 preview\"."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Hunyuan Hy3-preview\nAvailable\nTencent HunyuanFrontierOpen weights\nTencent's third-generation Hunyuan, rebuilt from scratch in ~90 days and open-sourced as the \"Hy3 preview\"."
   ]
  },
  {
   "url": "https://llm-releases.com/coding-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Xiaomi (MiMo)",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "open-weight flagship: a 1.02T-parameter Mixture-of-Experts model with ~42B active parameters, a hybrid-attention architecture, and a 1M-token context window",
   "organization_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### MiMo-V2.5-Pro\nAvailable\nXiaomi (MiMo)FrontierOpen source\nXiaomi's open-weight flagship: a 1.02T-parameter Mixture-of-Experts model with ~42B active parameters, a hybrid-attention architecture, and a 1M-token context window."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### MiMo-V2.5-Pro\nAvailable\nXiaomi (MiMo)FrontierOpen source\nXiaomi's open-weight flagship: a 1.02T-parameter Mixture-of-Experts model with ~42B active parameters, a hybrid-attention architecture, and a 1M-token context window."
   ]
  },
  {
   "url": "https://llm-releases.com/coding-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Xiaomi (MiMo)",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "open-weight sparse-MoE model: ~310B total parameters with ~15B active",
   "organization_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### MiMo-V2.5\nAvailable\nXiaomi (MiMo)Open source\nXiaomi's open-weight sparse-MoE model: ~310B total parameters with ~15B active, trained on ~48T tokens, with a 1M-token context window."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### MiMo-V2.5\nAvailable\nXiaomi (MiMo)Open source\nXiaomi's open-weight sparse-MoE model: ~310B total parameters with ~15B active, trained on ~48T tokens, with a 1M-token context window."
   ]
  },
  {
   "url": "https://llm-releases.com/coding-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Google DeepMind",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "open-weight text-diffusion model built on the Gemma 4 26B-A4B MoE backbone",
   "organization_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### DiffusionGemma 26B-A4B\nAvailable\nGoogle DeepMindOpen source\nAn open-weight text-diffusion model built on the Gemma 4 26B-A4B MoE backbone (25.2B total / 3.8B active)."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### DiffusionGemma 26B-A4B\nAvailable\nGoogle DeepMindOpen source\nAn open-weight text-diffusion model built on the Gemma 4 26B-A4B MoE backbone (25.2B total / 3.8B active)."
   ]
  },
  {
   "url": "https://llm-releases.com/coding-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Cohere",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "first developer-focused model and the first in its North family of code agents: a 30B-total / 3B-active MoE",
   "organization_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### North Mini Code 1.0\nAvailable\nCohereOpen source\nCohere's first developer-focused model and the first in its North family of code agents."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### North Mini Code 1.0\nAvailable\nCohereOpen source\nCohere's first developer-focused model and the first in its North family of code agents. A 30B-total / 3B-active MoE for agentic coding with a 256K context and up to 64K output, sized to run locally for enterprise coding agents. Apache-2.0."
   ]
  },
  {
   "url": "https://llm-releases.com/coding-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Google DeepMind",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "dense 12B member of the Gemma 4 family with a unified, encoder-free multimodal architecture",
   "organization_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Gemma 4 12B\nAvailable\nGoogle DeepMindOpen source\nA dense 12B member of the Gemma 4 family with a unified, encoder-free multimodal architecture: vision and audio are projected straight into the LLM backbone."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Gemma 4 12B\nAvailable\nGoogle DeepMindOpen source\nA dense 12B member of the Gemma 4 family with a unified, encoder-free multimodal architecture: vision and audio are projected straight into the LLM backbone."
   ]
  },
  {
   "url": "https://llm-releases.com/coding-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Nex AGI",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "open-weight agentic flagship, post-trained on Qwen3.5-397B-A17B",
   "organization_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Nex-N2-Pro\nAvailable\nNex AGIOpen source\nNex AGI's open-weight agentic flagship, post-trained on Qwen3.5-397B-A17B (397B total / ~17B active MoE) by the Shanghai Innovation Institute-led Nex alliance."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Nex-N2-Pro\nAvailable\nNex AGIOpen source\nNex AGI's open-weight agentic flagship, post-trained on Qwen3.5-397B-A17B (397B total / ~17B active MoE) by the Shanghai Innovation Institute-led Nex alliance."
   ]
  },
  {
   "url": "https://llm-releases.com/coding-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "StepFun",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "high-efficiency multimodal sparse-MoE successor to Step-3.5-Flash",
   "organization_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Step-3.7-Flash\nAvailable\nStepFunOpen source\nStepFun's high-efficiency multimodal sparse-MoE successor to Step-3.5-Flash: a ~196B-total / ~11B-active vision-language model with native image and video understanding, a 256K context, and selectable reasoning tiers (high/medium/low)."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Step-3.7-Flash\nAvailable\nStepFunOpen source\nStepFun's high-efficiency multimodal sparse-MoE successor to Step-3.5-Flash: a ~196B-total / ~11B-active vision-language model with native image and video understanding, a 256K context, and selectable reasoning tiers (high/medium/low)."
   ]
  },
  {
   "url": "https://llm-releases.com/coding-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Liquid AI",
   "model": null,
   "release_date": "2026-05-28",
   "access_type": null,
   "license": null,
   "significance": "on-device Mixture-of-Experts model: 8.3B total parameters with only ~1.5B active per forward pass",
   "organization_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### LFM2.5-8B-A1B\nAvailable\nLiquid AIOpen weights\nLiquid AI's on-device Mixture-of-Experts model: 8.3B total parameters with only ~1.5B active per forward pass (32 experts, 4 active per token)."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "Released May 28 2026 under the LFM Open License; caught in a July catalog-gap sweep."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### LFM2.5-8B-A1B\nAvailable\nLiquid AIOpen weights\nLiquid AI's on-device Mixture-of-Experts model: 8.3B total parameters with only ~1.5B active per forward pass (32 experts, 4 active per token)."
   ]
  },
  {
   "url": "https://llm-releases.com/coding-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Alibaba (Qwen)",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Dense 27B that punches far above its weight on agentic coding",
   "organization_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Qwen3.6-27B\nAvailable\nAlibaba (Qwen)Open source\nDense 27B that punches far above its weight on agentic coding \u2014 easy to self-host on a single GPU node."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Qwen3.6-27B\nAvailable\nAlibaba (Qwen)Open source\nDense 27B that punches far above its weight on agentic coding \u2014 easy to self-host on a single GPU node."
   ]
  },
  {
   "url": "https://llm-releases.com/coding-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Mistral AI",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "first Mistral model to unify reasoning, multimodal understanding, and agentic coding into one Apache 2.0 model",
   "organization_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Mistral Small 4\nAvailable\nMistral AIOpen source\nMistral's March 2026 Small release: the first Mistral model to unify reasoning (Magistral), multimodal understanding (Pixtral), and agentic coding (Devstral) into one Apache 2.0 model."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Mistral Small 4\nAvailable\nMistral AIOpen source\nMistral's March 2026 Small release: the first Mistral model to unify reasoning (Magistral), multimodal understanding (Pixtral), and agentic coding (Devstral) into one Apache 2.0 model."
   ]
  },
  {
   "url": "https://llm-releases.com/coding-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "StepFun",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Apache-licensed sparse MoE model for fast agentic execution, coding, math, browsing, and tool-use workflows",
   "organization_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Step-3.5-Flash\nAvailable\nStepFunOpen source\nStepFun's Apache-licensed sparse MoE model for fast agentic execution, coding, math, browsing, and tool-use workflows."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Step-3.5-Flash\nAvailable\nStepFunOpen source\nStepFun's Apache-licensed sparse MoE model for fast agentic execution, coding, math, browsing, and tool-use workflows."
   ]
  },
  {
   "url": "https://llm-releases.com/coding-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Sarvam AI",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Apache-licensed Indian-context MoE from Sarvam AI, optimized for reasoning, coding, agentic tasks, and 22 Indian languages",
   "organization_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Sarvam-105B\nAvailable\nSarvam AIOpen source\nApache-licensed Indian-context MoE from Sarvam AI, optimized for reasoning, coding, agentic tasks, and 22 Indian languages."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Sarvam-105B\nAvailable\nSarvam AIOpen source\nApache-licensed Indian-context MoE from Sarvam AI, optimized for reasoning, coding, agentic tasks, and 22 Indian languages."
   ]
  },
  {
   "url": "https://llm-releases.com/coding-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Alibaba (Qwen)",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "flagship of Alibaba's small dense Qwen3.5 models",
   "organization_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Qwen3.5-9B\nAvailable\nAlibaba (Qwen)Open source\nThe flagship of Alibaba's small dense Qwen3.5 models."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Qwen3.5-9B\nAvailable\nAlibaba (Qwen)Open source\nThe flagship of Alibaba's small dense Qwen3.5 models."
   ]
  },
  {
   "url": "https://llm-releases.com/coding-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Alibaba (Qwen)",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "dense 4B in Alibaba's small Qwen3.5 family, rated by Artificial Analysis as the most intelligent model under 5B parameters at launch",
   "organization_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Qwen3.5-4B\nAvailable\nAlibaba (Qwen)Open source\nA dense 4B in Alibaba's small Qwen3.5 family, rated by Artificial Analysis as the most intelligent model under 5B parameters at launch \u2014 outscoring several 7B\u20139B peers despite roughly half the parameters."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Qwen3.5-4B\nAvailable\nAlibaba (Qwen)Open source\nA dense 4B in Alibaba's small Qwen3.5 family, rated by Artificial Analysis as the most intelligent model under 5B parameters at launch \u2014 outscoring several 7B\u20139B peers despite roughly half the parameters."
   ]
  },
  {
   "url": "https://llm-releases.com/coding-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Alibaba (Qwen)",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "dense 2B Qwen3.5 model built for high-throughput, low-latency edge and on-device use",
   "organization_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Qwen3.5-2B\nAvailable\nAlibaba (Qwen)Open source\nA dense 2B Qwen3.5 model built for high-throughput, low-latency edge and on-device use."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Qwen3.5-2B\nAvailable\nAlibaba (Qwen)Open source\nA dense 2B Qwen3.5 model built for high-throughput, low-latency edge and on-device use."
   ]
  },
  {
   "url": "https://llm-releases.com/coding-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Alibaba (Qwen)",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "smallest Qwen3.5 model \u2014 a dense 0.8B designed for the most constrained on-device deployments",
   "organization_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Qwen3.5-0.8B\nAvailable\nAlibaba (Qwen)Open source\nThe smallest Qwen3.5 model \u2014 a dense 0.8B designed for the most constrained on-device deployments, operating in non-thinking (instruct) mode by default."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Qwen3.5-0.8B\nAvailable\nAlibaba (Qwen)Open source\nThe smallest Qwen3.5 model \u2014 a dense 0.8B designed for the most constrained on-device deployments, operating in non-thinking (instruct) mode by default."
   ]
  },
  {
   "url": "https://llm-releases.com/coding-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Allen Institute for AI (Ai2)",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "fully open thinking model with public weights, code, data, checkpoints, and training details",
   "organization_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### OLMo 3 Think 32B\nAvailable\nAllen Institute for AI (Ai2)Open source\nAi2's fully open thinking model with public weights, code, data, checkpoints, and training details across the OLMo 3 pipeline."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### OLMo 3 Think 32B\nAvailable\nAllen Institute for AI (Ai2)Open source\nAi2's fully open thinking model with public weights, code, data, checkpoints, and training details across the OLMo 3 pipeline."
   ]
  },
  {
   "url": "https://llm-releases.com/coding-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Alibaba (Qwen)",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Apache-licensed Qwen3-Next coding-agent model with 80B total / 3B active parameters",
   "organization_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Qwen3-Coder-Next\nAvailable\nAlibaba (Qwen)Open source\nApache-licensed Qwen3-Next coding-agent model with 80B total / 3B active parameters, 256K context, and long-horizon tool-use training."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Qwen3-Coder-Next\nAvailable\nAlibaba (Qwen)Open source\nApache-licensed Qwen3-Next coding-agent model with 80B total / 3B active parameters, 256K context, and long-horizon tool-use training."
   ]
  },
  {
   "url": "https://llm-releases.com/coding-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Moonshot AI",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "MIT-licensed hybrid linear-attention model using Kimi Delta Attention, built for million-token contexts",
   "organization_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Kimi-Linear-48B-A3B-Instruct\nAvailable\nMoonshot AIOpen source\nMIT-licensed hybrid linear-attention model using Kimi Delta Attention, built for million-token contexts with much lower KV-cache usage."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Kimi-Linear-48B-A3B-Instruct\nAvailable\nMoonshot AIOpen source\nMIT-licensed hybrid linear-attention model using Kimi Delta Attention, built for million-token contexts with much lower KV-cache usage."
   ]
  },
  {
   "url": "https://llm-releases.com/coding-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "ByteDance Seed",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Apache-licensed long-context reasoning and agent model, with controllable thinking budgets and a native 512K context",
   "organization_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Seed-OSS-36B-Instruct\nAvailable\nByteDance SeedOpen source\nByteDance Seed's Apache-licensed long-context reasoning and agent model, with controllable thinking budgets and a native 512K context."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Seed-OSS-36B-Instruct\nAvailable\nByteDance SeedOpen source\nByteDance Seed's Apache-licensed long-context reasoning and agent model, with controllable thinking budgets and a native 512K context."
   ]
  },
  {
   "url": "https://llm-releases.com/coding-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "OpenAI",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Smaller gpt-oss reasoning model optimized for local inference on systems with about 16GB of memory",
   "organization_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### gpt-oss-20b\nAvailable\nOpenAIOpen source\nSmaller gpt-oss reasoning model optimized for local inference on systems with about 16GB of memory."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### gpt-oss-20b\nAvailable\nOpenAIOpen source\nSmaller gpt-oss reasoning model optimized for local inference on systems with about 16GB of memory."
   ]
  },
  {
   "url": "https://llm-releases.com/coding-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "OpenAI",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "larger open-weight reasoning model, a 117B-total / 5.1B-active MoE with 128K context for local and self-hosted deployment",
   "organization_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### gpt-oss-120b\nAvailable\nOpenAIOpen source\nOpenAI's larger open-weight reasoning model, a 117B-total / 5.1B-active MoE with 128K context for local and self-hosted deployment."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### gpt-oss-120b\nAvailable\nOpenAIOpen source\nOpenAI's larger open-weight reasoning model, a 117B-total / 5.1B-active MoE with 128K context for local and self-hosted deployment."
   ]
  },
  {
   "url": "https://llm-releases.com/coding-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Alibaba (Qwen)",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "large open coding-agent model: a 480B-total / 35B-active MoE released under Apache-2.0",
   "organization_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Qwen3-Coder-480B-A35B-Instruct\nAvailable\nAlibaba (Qwen)Open source\nAlibaba Qwen's large open coding-agent model: a 480B-total / 35B-active MoE released under Apache-2.0, tuned for code generation, repository-level software engineering, tool calling, and long-horizon agent workflows with a 256K-token native context."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Qwen3-Coder-480B-A35B-Instruct\nAvailable\nAlibaba (Qwen)Open source\nAlibaba Qwen's large open coding-agent model: a 480B-total / 35B-active MoE released under Apache-2.0, tuned for code generation, repository-level software engineering, tool calling, and long-horizon agent workflows with a 256K-token native context."
   ]
  },
  {
   "url": "https://llm-releases.com/coding-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Moonshot AI",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "MIT-licensed coding LLM trained with repository-level reinforcement learning for software issue resolution",
   "organization_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Kimi-Dev-72B\nAvailable\nMoonshot AIOpen source\nMIT-licensed coding LLM trained with repository-level reinforcement learning for software issue resolution."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Kimi-Dev-72B\nAvailable\nMoonshot AIOpen source\nMIT-licensed coding LLM trained with repository-level reinforcement learning for software issue resolution."
   ]
  },
  {
   "url": "https://llm-releases.com/coding-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Hugging Face",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "fully open 3B multilingual long-context model with optional reasoning mode and 128K context",
   "organization_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### SmolLM3 3B\nAvailable\nHugging FaceOpen source\nHugging Face's fully open 3B multilingual long-context model with optional reasoning mode and 128K context."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### SmolLM3 3B\nAvailable\nHugging FaceOpen source\nHugging Face's fully open 3B multilingual long-context model with optional reasoning mode and 128K context."
   ]
  },
  {
   "url": "https://llm-releases.com/coding-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Baidu",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "open ERNIE 4.5 language MoE, part of a 10-variant Apache-licensed model family",
   "organization_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### ERNIE-4.5-300B-A47B\nAvailable\nBaiduOpen source\nBaidu's open ERNIE 4.5 language MoE, part of a 10-variant Apache-licensed model family built with heterogeneous multimodal MoE training."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### ERNIE-4.5-300B-A47B\nAvailable\nBaiduOpen source\nBaidu's open ERNIE 4.5 language MoE, part of a 10-variant Apache-licensed model family built with heterogeneous multimodal MoE training."
   ]
  },
  {
   "url": "https://llm-releases.com/coding-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Mistral AI",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "open coding-agent model, released as a 24B Apache-2.0 research preview for software engineering tasks",
   "organization_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Devstral Small 2505\nPreview\nMistral AIOpen source\nMistral and All Hands AI's open coding-agent model, released as a 24B Apache-2.0 research preview for software engineering tasks."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Devstral Small 2505\nPreview\nMistral AIOpen source\nMistral and All Hands AI's open coding-agent model, released as a 24B Apache-2.0 research preview for software engineering tasks."
   ]
  },
  {
   "url": "https://llm-releases.com/coding-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "IBM",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Granite 3.3 text update for enterprise chat, RAG, and instruction-following workflows",
   "organization_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Granite 3.3 8B\nAvailable\nIBMOpen source\nGranite 3.3 text update for enterprise chat, RAG, and instruction-following workflows."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Granite 3.3 8B\nAvailable\nIBMOpen source\nGranite 3.3 text update for enterprise chat, RAG, and instruction-following workflows."
   ]
  },
  {
   "url": "https://llm-releases.com/coding-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Alibaba (Qwen)",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Largest open Qwen3 MoE, introducing hybrid thinking/non-thinking modes and 119-language coverage",
   "organization_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Qwen3-235B-A22B\nAvailable\nAlibaba (Qwen)Open source\nLargest open Qwen3 MoE, introducing hybrid thinking/non-thinking modes and 119-language coverage."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Qwen3-235B-A22B\nAvailable\nAlibaba (Qwen)Open source\nLargest open Qwen3 MoE, introducing hybrid thinking/non-thinking modes and 119-language coverage."
   ]
  },
  {
   "url": "https://llm-releases.com/coding-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Allen Institute for AI (Ai2)",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "fully open model \u2014 weights, data, and training code all public \u2014 and the first such to beat GPT-3.5 / GPT-4o mini",
   "organization_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### OLMo 2 32B\nAvailable\nAllen Institute for AI (Ai2)Open source\nA fully open model \u2014 weights, data, and training code all public \u2014 and the first such to beat GPT-3.5 / GPT-4o mini."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### OLMo 2 32B\nAvailable\nAllen Institute for AI (Ai2)Open source\nA fully open model \u2014 weights, data, and training code all public \u2014 and the first such to beat GPT-3.5 / GPT-4o mini."
   ]
  },
  {
   "url": "https://llm-releases.com/coding-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "IBM",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Granite 3.2 update with reasoning controls and multimodal/document-oriented Granite variants",
   "organization_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Granite 3.2 8B\nAvailable\nIBMOpen source\nGranite 3.2 update with reasoning controls and multimodal/document-oriented Granite variants."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Granite 3.2 8B\nAvailable\nIBMOpen source\nGranite 3.2 update with reasoning controls and multimodal/document-oriented Granite variants."
   ]
  },
  {
   "url": "https://llm-releases.com/coding-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Moonshot AI",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "MIT-licensed 16B/3B-active MoE trained with Moonshot's scalable Muon optimizer experiments",
   "organization_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Moonlight-16B-A3B-Instruct\nAvailable\nMoonshot AIOpen source\nMIT-licensed 16B/3B-active MoE trained with Moonshot's scalable Muon optimizer experiments."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Moonlight-16B-A3B-Instruct\nAvailable\nMoonshot AIOpen source\nMIT-licensed 16B/3B-active MoE trained with Moonshot's scalable Muon optimizer experiments."
   ]
  },
  {
   "url": "https://llm-releases.com/coding-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Mistral AI",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "latency-optimized 24B dense model under Apache-2.0",
   "organization_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Mistral Small 3\nAvailable\nMistral AIOpen source\nA latency-optimized 24B dense model under Apache-2.0 \u2014 a popular local-deployment workhorse."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Mistral Small 3\nAvailable\nMistral AIOpen source\nA latency-optimized 24B dense model under Apache-2.0 \u2014 a popular local-deployment workhorse."
   ]
  },
  {
   "url": "https://llm-releases.com/coding-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "IBM",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "enterprise-focused open model with a 128k context, Apache-2.0 licensed",
   "organization_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Granite 3.1 8B\nAvailable\nIBMOpen source\nIBM's enterprise-focused open model with a 128k context, Apache-2.0 licensed."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Granite 3.1 8B\nAvailable\nIBMOpen source\nIBM's enterprise-focused open model with a 128k context, Apache-2.0 licensed."
   ]
  },
  {
   "url": "https://llm-releases.com/coding-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Microsoft",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "14B dense model that rivals far larger ones on math and reasoning, under a permissive MIT license",
   "organization_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Phi-4\nAvailable\nMicrosoftOpen source\nA 14B dense model that rivals far larger ones on math and reasoning, under a permissive MIT license."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Phi-4\nAvailable\nMicrosoftOpen source\nA 14B dense model that rivals far larger ones on math and reasoning, under a permissive MIT license."
   ]
  },
  {
   "url": "https://llm-releases.com/coding-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Alibaba (Qwen)",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Qwen's first public reasoning-preview model, aimed at math, coding, and deliberate problem solving",
   "organization_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### QwQ-32B-Preview\nAvailable\nAlibaba (Qwen)Open source\nQwen's first public reasoning-preview model, aimed at math, coding, and deliberate problem solving."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### QwQ-32B-Preview\nAvailable\nAlibaba (Qwen)Open source\nQwen's first public reasoning-preview model, aimed at math, coding, and deliberate problem solving."
   ]
  },
  {
   "url": "https://llm-releases.com/coding-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Allen Institute for AI (Ai2)",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "post-trained open instruction model line, scaling the Tulu recipe to Llama 3.1 405B",
   "organization_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Tulu 3 405B\nAvailable\nAllen Institute for AI (Ai2)Open weights\nAi2's post-trained open instruction model line, scaling the Tulu recipe to Llama 3.1 405B."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Tulu 3 405B\nAvailable\nAllen Institute for AI (Ai2)Open weights\nAi2's post-trained open instruction model line, scaling the Tulu recipe to Llama 3.1 405B."
   ]
  },
  {
   "url": "https://llm-releases.com/coding-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Alibaba (Qwen)",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Code-specialized Qwen2.5 model family, with the 32B checkpoint as the flagship open coding model",
   "organization_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Qwen2.5-Coder-32B\nAvailable\nAlibaba (Qwen)Open source\nCode-specialized Qwen2.5 model family, with the 32B checkpoint as the flagship open coding model."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Qwen2.5-Coder-32B\nAvailable\nAlibaba (Qwen)Open source\nCode-specialized Qwen2.5 model family, with the 32B checkpoint as the flagship open coding model."
   ]
  },
  {
   "url": "https://llm-releases.com/coding-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Hugging Face",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Compact on-device model family trained on 11T tokens, popular for lightweight local chat and experimentation",
   "organization_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### SmolLM2 1.7B\nAvailable\nHugging FaceOpen source\nCompact on-device model family trained on 11T tokens, popular for lightweight local chat and experimentation."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### SmolLM2 1.7B\nAvailable\nHugging FaceOpen source\nCompact on-device model family trained on 11T tokens, popular for lightweight local chat and experimentation."
   ]
  },
  {
   "url": "https://llm-releases.com/coding-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "IBM",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Apache-licensed Granite 3.0 text model, part of IBM's push toward enterprise-friendly open models",
   "organization_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Granite 3.0 8B\nAvailable\nIBMOpen source\nApache-licensed Granite 3.0 text model, part of IBM's push toward enterprise-friendly open models."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Granite 3.0 8B\nAvailable\nIBMOpen source\nApache-licensed Granite 3.0 text model, part of IBM's push toward enterprise-friendly open models."
   ]
  },
  {
   "url": "https://llm-releases.com/coding-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Allen Institute for AI (Ai2)",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Fully open sparse MoE model with 7B total and about 1B active parameters",
   "organization_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### OLMoE 1B-7B\nAvailable\nAllen Institute for AI (Ai2)Open source\nFully open sparse MoE model with 7B total and about 1B active parameters."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### OLMoE 1B-7B\nAvailable\nAllen Institute for AI (Ai2)Open source\nFully open sparse MoE model with 7B total and about 1B active parameters."
   ]
  },
  {
   "url": "https://llm-releases.com/coding-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Mistral AI",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Apache-licensed 12B model co-developed with NVIDIA, including a 128K context window and strong multilingual tokenization",
   "organization_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Mistral NeMo\nAvailable\nMistral AIOpen source\nApache-licensed 12B model co-developed with NVIDIA, including a 128K context window and strong multilingual tokenization."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Mistral NeMo\nAvailable\nMistral AIOpen source\nApache-licensed 12B model co-developed with NVIDIA, including a 128K context window and strong multilingual tokenization."
   ]
  },
  {
   "url": "https://llm-releases.com/coding-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "DeepSeek",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Open code-focused MoE built from DeepSeek-V2, expanding programming-language coverage and coding benchmark performance",
   "organization_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### DeepSeek-Coder-V2\nAvailable\nDeepSeekOpen source\nOpen code-focused MoE built from DeepSeek-V2, expanding programming-language coverage and coding benchmark performance."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### DeepSeek-Coder-V2\nAvailable\nDeepSeekOpen source\nOpen code-focused MoE built from DeepSeek-V2, expanding programming-language coverage and coding benchmark performance."
   ]
  },
  {
   "url": "https://llm-releases.com/coding-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Alibaba (Qwen)",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Qwen2's largest dense model, introducing stronger multilingual support, coding/math gains, and long-context variants",
   "organization_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Qwen2-72B\nAvailable\nAlibaba (Qwen)Open weights\nQwen2's largest dense model, introducing stronger multilingual support, coding/math gains, and long-context variants."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Qwen2-72B\nAvailable\nAlibaba (Qwen)Open weights\nQwen2's largest dense model, introducing stronger multilingual support, coding/math gains, and long-context variants."
   ]
  },
  {
   "url": "https://llm-releases.com/coding-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Mistral AI",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Mistral's first code-specialized model, trained for code generation, fill-in-the-middle, and multi-language programming tasks",
   "organization_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Codestral 22B\nAvailable\nMistral AIOpen weights\nMistral's first code-specialized model, trained for code generation, fill-in-the-middle, and multi-language programming tasks."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Codestral 22B\nAvailable\nMistral AIOpen weights\nMistral's first code-specialized model, trained for code generation, fill-in-the-middle, and multi-language programming tasks."
   ]
  },
  {
   "url": "https://llm-releases.com/coding-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "DeepSeek",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "DeepSeek's first major MoE general model with Multi-head Latent Attention and low-cost API positioning",
   "organization_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### DeepSeek-V2\nAvailable\nDeepSeekOpen source\nDeepSeek's first major MoE general model with Multi-head Latent Attention and low-cost API positioning."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### DeepSeek-V2\nAvailable\nDeepSeekOpen source\nDeepSeek's first major MoE general model with Multi-head Latent Attention and low-cost API positioning."
   ]
  },
  {
   "url": "https://llm-releases.com/coding-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "IBM",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Apache-2.0 code model from IBM's Granite Code family, used for local code generation and enterprise coding assistants",
   "organization_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Granite Code 34B\nAvailable\nIBMOpen source\nApache-2.0 code model from IBM's Granite Code family, used for local code generation and enterprise coding assistants."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Granite Code 34B\nAvailable\nIBMOpen source\nApache-2.0 code model from IBM's Granite Code family, used for local code generation and enterprise coding assistants."
   ]
  },
  {
   "url": "https://llm-releases.com/coding-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Snowflake AI Research",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Apache-2.0 enterprise LLM with 480B total / 17B active parameters, optimized for SQL, code, and instr",
   "organization_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Snowflake Arctic\nAvailable\nSnowflake AI ResearchOpen source\nApache-2.0 enterprise LLM with 480B total / 17B active parameters, optimized for SQL, code, and instr"
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Software and agentic coding",
    "Coding LLM releases",
    "[### Snowflake Arctic\nAvailable\nSnowflake AI ResearchOpen source\nApache-2.0 enterprise LLM with 480B total / 17B active parameters, optimized for SQL, code, and instr"
   ]
  },
  {
   "url": "https://dev.to/doremonai/openais-gpt-oss-the-moment-they-finally-went-open-weight-with-120b-and-20b-models-5g4m",
   "title": "OpenAI's gpt-oss: The Moment They Finally Went Open-Weight With 120B and 20B Models",
   "published_at": "2026-06-21T18:16:46",
   "organization": "OpenAI",
   "model": null,
   "release_date": "2026-06-21",
   "access_type": null,
   "license": null,
   "significance": "state-of-the-art large language models available under permissive open-weight licenses",
   "organization_evidence": [
    "# OpenAI's gpt-oss: The Moment They Finally Went Open-Weight With 120B and 20B Models",
    "Yesterday, Sam Altman announced **gpt-oss-120b** and **gpt-oss-20b**, two state-of-the-art large language models available under permissive open-weight licenses."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "*All benchmark data from OpenAI's official announcement. Posted on Jun 21, 2026.*",
    "# OpenAI's gpt-oss: The Moment They Finally Went Open-Weight With 120B and 20B Models"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Yesterday, Sam Altman announced **gpt-oss-120b** and **gpt-oss-20b**, two state-of-the-art large language models available under permissive open-weight licenses.",
    "# OpenAI's gpt-oss: The Moment They Finally Went Open-Weight With 120B and 20B Models"
   ]
  },
  {
   "url": "https://pondero.ai/news/2026-07-24-poolside-laguna-s-21",
   "title": "Poolside releases Laguna S 2.1, a 118B open-weight model that tops SWE-Bench Multilingual and outperforms rivals several times its effective size",
   "published_at": "2026-07-24T00:00:00",
   "organization": "Poolside",
   "model": null,
   "release_date": "2026-07-21",
   "access_type": null,
   "license": null,
   "significance": "118B mixture-of-experts coding model that scored 78.5% on SWE-Bench Multilingual and 70.2% on Terminal-Bench 2.1, runs on a single Nvidia DGX Spark",
   "organization_evidence": [
    "## Poolside releases Laguna S 2.1, a 118B open-weight model that tops SWE-Bench Multilingual and outperforms rivals several times its effective size",
    "Poolside released Laguna S 2.1 on July 21, 2026, a mixture-of-experts coding model that climbed to the top of the SWE-Bench Multilingual public leaderboard at 78.5% and beat larger open-weight rivals while needing only a single Nvidia DGX Spark to run."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## Poolside releases Laguna S 2.1, a 118B open-weight model that tops SWE-Bench Multilingual and outperforms rivals several times its effective size",
    "Poolside released Laguna S 2.1 on July 21, 2026, a mixture-of-experts coding model that climbed to the top of the SWE-Bench Multilingual public leaderboard at 78.5% and beat larger open-weight rivals while needing only a single Nvidia DGX Spark to run."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## Poolside releases Laguna S 2.1, a 118B open-weight model that tops SWE-Bench Multilingual and outperforms rivals several times its effective size",
    "Poolside's Laguna S 2.1 is a 118B mixture-of-experts coding model that scored 78.5% on SWE-Bench Multilingual and 70.2% on Terminal-Bench 2.1, runs on a single Nvidia DGX Spark, and ships free on Hugging Face under a commercial-use license."
   ]
  },
  {
   "url": "https://en.tibet3.com/culture/2026-03/18/content_500420011.html",
   "title": "Tibetan Large Language Model DeepZang Launched in Lhasa",
   "published_at": null,
   "organization": "Xizang CHOKNOR Digital Industry Co., Ltd.",
   "model": null,
   "release_date": "2026-03-15",
   "access_type": null,
   "license": null,
   "significance": "first Tibetan large language model in China to have completed national registration for generative artificial intelligence (AI)",
   "organization_evidence": [
    "## Tibetan Large Language Model DeepZang Launched in Lhasa",
    "DeepZang was independently developed by Xizang CHOKNOR Digital Industry Co., Ltd."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## Tibetan Large Language Model DeepZang Launched in Lhasa",
    "On March 15, DeepZang, a Tibetan large language model with its intelligent applications, was launched in Lhasa, Xizang."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## Tibetan Large Language Model DeepZang Launched in Lhasa",
    "On March 15, DeepZang, a Tibetan large language model with its intelligent applications, was launched in Lhasa, Xizang. This is the first Tibetan large language model in China to have completed national registration for generative artificial intelligence (AI)."
   ]
  },
  {
   "url": "https://the-ledger.net/events/thinking-machines-inkling-open-weights",
   "title": "Thinking Machines Lab Releases Inkling, Its First Open-Weights Model \u2014 The Ledger",
   "published_at": "2026-06-01T00:00:00",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "A multimodal mixture-of-experts transformer with 975 billion total parameters, 41 billion active parameters per token, and a context window of up to one million tokens.",
   "organization_evidence": [
    "# Thinking Machines Lab Releases Inkling, Its First Open-Weights Model",
    "On July 15, 2026, Thinking Machines Lab released Inkling, its first model trained from scratch and its first publicly downloadable model weights."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Thinking Machines Lab Releases Inkling, Its First Open-Weights Model",
    "A ledger entry in the open source archive, dated 2026-07-15.",
    "On July 15, 2026, Thinking Machines Lab released Inkling, its first model trained from scratch and its first publicly downloadable model weights."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Thinking Machines Lab Releases Inkling, Its First Open-Weights Model",
    "Inkling is a multimodal mixture-of-experts transformer with 975 billion total parameters, 41 billion active parameters per token, and a context window of up to one million tokens."
   ]
  },
  {
   "url": "https://huggingface.co/blog/mlabonne/qwen35",
   "title": "Qwen3.5: Nobody Agrees on Attention Anymore",
   "published_at": "2026-02-17T13:00:51",
   "organization": "Alibaba",
   "model": "Qwen3.5-397B-A17B",
   "release_date": "2026-02-16",
   "access_type": "open model",
   "license": null,
   "significance": "a 397 billion parameter Mixture-of-Experts model with only 17 billion active parameters per token, featuring a hybrid attention architecture, native multimodality, and 201 supported languages",
   "organization_evidence": [
    "On February 16th, 2026, Alibaba's Qwen team released **Qwen3.5-397B-A17B** , their next-generation foundation model."
   ],
   "model_evidence": [
    "On February 16th, 2026, Alibaba's Qwen team released **Qwen3.5-397B-A17B** , their next-generation foundation model."
   ],
   "release_date_evidence": [
    "On February 16th, 2026, Alibaba's Qwen team released **Qwen3.5-397B-A17B** , their next-generation foundation model."
   ],
   "access_type_evidence": [
    "201 Languages.",
    "This is the broadest language coverage of any open model I\u2019m aware of."
   ],
   "license_evidence": null,
   "significance_evidence": [
    "Qwen3.5-397B-A17B is a 397 billion parameter Mixture-of-Experts model with only 17 billion active parameters per token.",
    "Qwen3.5 builds on the Qwen3-Next lineage, combining Gated Delta Networks (a linear attention variant) with sparse Mixture-of-Experts.",
    "Unlike Qwen3, which had separate text and vision model lines (Qwen3 and Qwen3-VL), Qwen3.5 is natively multimodal from the ground up."
   ]
  },
  {
   "url": "https://huggingface.co/blog/mlabonne/qwen35",
   "title": "Qwen3.5: Nobody Agrees on Attention Anymore",
   "published_at": "2026-02-17T13:00:51",
   "organization": "Zhipu",
   "model": "GLM-5",
   "release_date": "2026-02-11",
   "access_type": null,
   "license": null,
   "significance": "uses Multi-head Latent Attention and DeepSeek Sparse Attention",
   "organization_evidence": [
    "This follows the trend we've seen from MiniMax's Forge and Zhipu's Slime: asynchronous RL infrastructure designed to handle the long-horizon, multi-step nature of agentic tasks."
   ],
   "model_evidence": [
    "GLM-5 shipped on February 11th."
   ],
   "release_date_evidence": [
    "GLM-5 shipped on February 11th."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Kimi K2.5 and GLM-5 both use **Multi-head Latent Attention** (MLA), but GLM-5 also integrates **DeepSeek Sparse Attention** (DSA) to induce token-level sparsity on top of it."
   ]
  },
  {
   "url": "https://huggingface.co/blog/mlabonne/qwen35",
   "title": "Qwen3.5: Nobody Agrees on Attention Anymore",
   "published_at": "2026-02-17T13:00:51",
   "organization": "MiniMax",
   "model": "MiniMax M2.5",
   "release_date": "2026-02-11",
   "access_type": null,
   "license": null,
   "significance": "full attention model with Multi-Head Attention and a small active parameter count",
   "organization_evidence": [
    "MiniMax M2.5 landed the same day."
   ],
   "model_evidence": [
    "MiniMax M2.5 landed the same day."
   ],
   "release_date_evidence": [
    "MiniMax M2.5 landed the same day."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Finally, MiniMax-M2.5 is the only full attention model (see this article) with **Multi-Head Attention** (MHA) for reliability purposes.",
    "Notably, MiniMax M2.5 has the smallest active parameter count at 10B and the same activation ratio as Qwen3.5."
   ]
  },
  {
   "url": "https://huggingface.co/blog/mlabonne/qwen35",
   "title": "Qwen3.5: Nobody Agrees on Attention Anymore",
   "published_at": "2026-02-17T13:00:51",
   "organization": null,
   "model": "Kimi K2.5",
   "release_date": "2026-01-27",
   "access_type": null,
   "license": null,
   "significance": "uses Multi-head Latent Attention and is significantly bigger with 1T total parameters",
   "organization_evidence": null,
   "model_evidence": [
    "Kimi K2.5 arrived on January 27th."
   ],
   "release_date_evidence": [
    "Kimi K2.5 arrived on January 27th."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Kimi K2.5 and GLM-5 both use **Multi-head Latent Attention** (MLA), but GLM-5 also integrates **DeepSeek Sparse Attention** (DSA) to induce token-level sparsity on top of it.",
    "Kimi K2.5 is even sparser but also significantly bigger with 1T total parameters."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/gpt-5-6-sol-xhigh-vs-glm-5-2",
   "title": "GPT-5.6 Sol (xhigh) vs GLM-5.2 (max): Model Comparison | Artificial Analysis",
   "published_at": null,
   "organization": "Z AI",
   "model": null,
   "release_date": "2026-06",
   "access_type": null,
   "license": null,
   "significance": "A fast and cheap open-weights model scoring 53 on the Artificial Analysis Intelligence Index with 1.0M context window",
   "organization_evidence": [
    "# GPT-5.6 Sol (xhigh) vs. GLM-5.2 (max)",
    "Comparison between GPT-5.6 Sol (xhigh) and GLM-5.2 (max) across intelligence, price, speed, context window and more.",
    "| | ![OpenAI logo](/img/logos/openai_small.svg)OpenAI GPT-5.6 Sol (xhigh) | ![Z AI logo](/img/logos/zai_small.svg)Z AI GLM-5.2 (max) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# GPT-5.6 Sol (xhigh) vs. GLM-5.2 (max)",
    "Comparison between GPT-5.6 Sol (xhigh) and GLM-5.2 (max) across intelligence, price, speed, context window and more.",
    "| Release Date | July 2026 | June 2026 | GPT-5.6 Sol (xhigh) has a more recent release date than GLM-5.2 (max) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# GPT-5.6 Sol (xhigh) vs. GLM-5.2 (max)",
    "Comparison between GPT-5.6 Sol (xhigh) and GLM-5.2 (max) across intelligence, price, speed, context window and more.",
    "| Intelligence Index | 59 | 53 | GPT-5.6 Sol (xhigh) is more intelligent than GLM-5.2 (max) |",
    "| Price per 1M Tokens | $4.35 | $0.90 | GLM-5.2 (max) is cheaper than GPT-5.6 Sol (xhigh) |",
    "| Output Speed | 72 tokens/s | 98 tokens/s | GLM-5.2 (max) is faster than GPT-5.6 Sol (xhigh) |",
    "| Context Window | 1000k tokens~1,500 A4 pages of size 12 Arial font | 1000k tokens~1,500 A4 pages of size 12 Arial font | Both GPT-5.6 Sol (xhigh) and GLM-5.2 (max) have the same sized context window |"
   ]
  },
  {
   "url": "http://eng.tibet.cn/en/economic/202604/t20260430_7967667.html",
   "title": "Tibetan Large Language Model DeepZang Launched in Lhasa_Economic&Society_2026Home_China Tibet Online",
   "published_at": "2026-03-16T10:14:44",
   "organization": "Xizang CHOKNOR Digital Industry Co., Ltd.",
   "model": null,
   "release_date": "2026-03-15",
   "access_type": null,
   "license": null,
   "significance": "the first Tibetan large language model in China to have completed national registration for generative artificial intelligence (AI)",
   "organization_evidence": [
    "Tibetan Large Language Model DeepZang Launched in Lhasa",
    "DeepZang was independently developed by Xizang CHOKNOR Digital Industry Co., Ltd."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Tibetan Large Language Model DeepZang Launched in Lhasa",
    "On March 15, DeepZang, a Tibetan large language model with its intelligent applications, was launched in Lhasa, Xizang."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Tibetan Large Language Model DeepZang Launched in Lhasa",
    "On March 15, DeepZang, a Tibetan large language model with its intelligent applications, was launched in Lhasa, Xizang.",
    "This is the first Tibetan large language model in China to have completed national registration for generative artificial intelligence (AI)."
   ]
  },
  {
   "url": "https://technology.org/2026/07/16/thinking-machines-inkling-open-weights-model",
   "title": "Thinking Machines Releases Inkling Open-Weights Model - Technology Org",
   "published_at": "2026-07-16T07:53:40",
   "organization": "Thinking Machines",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "A 975B parameter MoE model with 41B active, a 1M token context window, pretrained on 45 trillion tokens of text, image, audio and video",
   "organization_evidence": [
    "## Thinking Machines Ships Inkling",
    "Thinking Machines Lab released its first in-house model on Wednesday, and unlike the flagships from OpenAI, Anthropic or Google, anyone can download it."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## Thinking Machines Ships Inkling",
    "Thinking Machines Lab released its first in-house model on Wednesday, and unlike the flagships from OpenAI, Anthropic or Google, anyone can download it.",
    "That is a different bet from the metered API model OpenAI and Anthropic sell, and [reporting on the release frames it as a test](https://techcrunch.com/2026/07/15/thinking-machines-amps-up-its-bet-against-one-size-fits-all-ai-with-its-first-open-model-inkling/) of whether customisation beats raw capability in the enterprise market."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## Thinking Machines Ships Inkling",
    "- Inkling is a 975B parameter MoE model with 41B active, a 1M token context window, pretrained on 45 trillion tokens of text, image, audio and video, with full weights on Hugging Face."
   ]
  },
  {
   "url": "https://huggingface.co/blog/FINAL-Bench/darwin-tts",
   "title": "Darwin-TTS: We Gave a TTS Model 3% of an LLM's Brain \u2014 It Started Showing Emotion",
   "published_at": "2026-04-15T04:57:21",
   "organization": "VIDRAFT",
   "model": null,
   "release_date": "2026-04-15",
   "access_type": null,
   "license": null,
   "significance": "world's first cross-modal LLM\u2192TTS FFN transfer model showing emotion without training",
   "organization_evidence": [
    "Darwin-TTS: We Gave a TTS Model 3% of an LLM's Brain \u2014 It Started Showing Emotion",
    "Built by [VIDRAFT](https://vidraft.net) (\ube44\ub4dc\ub798\ud504\ud2b8) Apache 2.0. Use it, break it, improve it."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Darwin-TTS: We Gave a TTS Model 3% of an LLM's Brain \u2014 It Started Showing Emotion",
    "Here's our full timeline from April 15, 2026:",
    "*This research was conducted on April 15, 2026, using 1\u00d7 H100 GPU for inference only."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Darwin-TTS: We Gave a TTS Model 3% of an LLM's Brain \u2014 It Started Showing Emotion",
    "Today we release [Darwin-TTS-1.7B-Cross](https://huggingface.co/FINAL-Bench/Darwin-TTS-1.7B-Cross), the world's first cross-modal LLM\u2192TTS FFN transfer model."
   ]
  },
  {
   "url": "https://strongmocha.com/ai-infrastructure-data-centers/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - StrongMocha",
   "published_at": "2026-07-15T12:06:52",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor",
   "organization_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "China\u2019s Release Cadence Is the Story",
    "Between late April and mid-June 2026, Chinese labs launched four major open-weight models: **DeepSeek V4** on April 24, **MiniMax M3** on June 1, and **Kimi K2.7-Code** and **GLM-5.2** within days of each other in mid-June."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "China\u2019s Release Cadence Is the Story",
    "Between late April and mid-June 2026, Chinese labs launched four major open-weight models: **DeepSeek V4** on April 24, **MiniMax M3** on June 1, and **Kimi K2.7-Code** and **GLM-5.2** within days of each other in mid-June."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "China\u2019s Release Cadence Is the Story",
    "### The production line \u2014 spring 2026",
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor"
   ]
  },
  {
   "url": "https://strongmocha.com/ai-infrastructure-data-centers/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - StrongMocha",
   "published_at": "2026-07-15T12:06:52",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "cheap 1M-token context, native multimodal, modified-MIT",
   "organization_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "China\u2019s Release Cadence Is the Story",
    "Between late April and mid-June 2026, Chinese labs launched four major open-weight models: **DeepSeek V4** on April 24, **MiniMax M3** on June 1, and **Kimi K2.7-Code** and **GLM-5.2** within days of each other in mid-June."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "China\u2019s Release Cadence Is the Story",
    "Between late April and mid-June 2026, Chinese labs launched four major open-weight models: **DeepSeek V4** on April 24, **MiniMax M3** on June 1, and **Kimi K2.7-Code** and **GLM-5.2** within days of each other in mid-June."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "China\u2019s Release Cadence Is the Story",
    "### The production line \u2014 spring 2026",
    "**MiniMax M3**cheap 1M-token context, native multimodal, modified-MIT"
   ]
  },
  {
   "url": "https://strongmocha.com/ai-infrastructure-data-centers/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - StrongMocha",
   "published_at": "2026-07-15T12:06:52",
   "organization": "Moonshot",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "agent-run specialist, ~30% fewer thinking tokens than K2.6",
   "organization_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "China\u2019s Release Cadence Is the Story",
    "Between late April and mid-June 2026, Chinese labs launched four major open-weight models: **DeepSeek V4** on April 24, **MiniMax M3** on June 1, and **Kimi K2.7-Code** and **GLM-5.2** within days of each other in mid-June."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "China\u2019s Release Cadence Is the Story",
    "Between late April and mid-June 2026, Chinese labs launched four major open-weight models: **DeepSeek V4** on April 24, **MiniMax M3** on June 1, and **Kimi K2.7-Code** and **GLM-5.2** within days of each other in mid-June."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "China\u2019s Release Cadence Is the Story",
    "### The production line \u2014 spring 2026",
    "**Kimi K2.7-Code (Moonshot)**agent-run specialist, ~30% fewer thinking tokens than K2.6"
   ]
  },
  {
   "url": "https://strongmocha.com/ai-infrastructure-data-centers/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - StrongMocha",
   "published_at": "2026-07-15T12:06:52",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-06-16",
   "access_type": null,
   "license": null,
   "significance": "753B MoE, MIT, top open-weight on Artificial Analysis index",
   "organization_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "China\u2019s Release Cadence Is the Story",
    "Between late April and mid-June 2026, Chinese labs launched four major open-weight models: **DeepSeek V4** on April 24, **MiniMax M3** on June 1, and **Kimi K2.7-Code** and **GLM-5.2** within days of each other in mid-June."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "China\u2019s Release Cadence Is the Story",
    "Between late April and mid-June 2026, Chinese labs launched four major open-weight models: **DeepSeek V4** on April 24, **MiniMax M3** on June 1, and **Kimi K2.7-Code** and **GLM-5.2** within days of each other in mid-June."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "China\u2019s Release Cadence Is the Story",
    "### The production line \u2014 spring 2026",
    "**GLM-5.2 (Z.ai)**753B MoE, MIT, top open-weight on Artificial Analysis index"
   ]
  },
  {
   "url": "https://fazm.ai/t/open-source-llm-releases-may-2026",
   "title": "Fazm - AI Computer Agent for macOS | Voice-Controlled Desktop Automation",
   "published_at": "2026-05-13T07:12:17",
   "organization": "OpenBMB",
   "model": null,
   "release_date": "2026-05-11",
   "access_type": null,
   "license": null,
   "significance": "1.3B parameter multimodal SLM optimized for on-device inference",
   "organization_evidence": [
    "# Open source LLM releases in May 2026: one new model so far, four April tail releases, and the Swift block that decides which of them drive your Mac",
    "OpenBMB shipped MiniCPM-V 4.6 1.3B Instruct on May 11, 2026 under Apache 2.0."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Open source LLM releases in May 2026: one new model so far, four April tail releases, and the Swift block that decides which of them drive your Mac",
    "OpenBMB shipped MiniCPM-V 4.6 1.3B Instruct on May 11, 2026 under Apache 2.0."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Open source LLM releases in May 2026: one new model so far, four April tail releases, and the Swift block that decides which of them drive your Mac",
    "The label on the box says \"1.3B parameter multimodal SLM optimized for on-device inference\", and that is what it does."
   ]
  },
  {
   "url": "https://fazm.ai/t/open-source-llm-releases-may-2026",
   "title": "Fazm - AI Computer Agent for macOS | Voice-Controlled Desktop Automation",
   "published_at": "2026-05-13T07:12:17",
   "organization": "Xiaomi",
   "model": null,
   "release_date": "2026-04-28",
   "access_type": null,
   "license": null,
   "significance": "1.02T total parameters with 42B active per token",
   "organization_evidence": [
    "# Open source LLM releases in May 2026: one new model so far, four April tail releases, and the Swift block that decides which of them drive your Mac",
    "April 28, 2026 - Xiaomi",
    "### MiMo-V2.5-Pro (1.02T MoE / 42B active)"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Open source LLM releases in May 2026: one new model so far, four April tail releases, and the Swift block that decides which of them drive your Mac",
    "April 28, 2026 - Xiaomi",
    "### MiMo-V2.5-Pro (1.02T MoE / 42B active)"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Open source LLM releases in May 2026: one new model so far, four April tail releases, and the Swift block that decides which of them drive your Mac",
    "1.02T total parameters with 42B active per token."
   ]
  },
  {
   "url": "https://fazm.ai/t/open-source-llm-releases-may-2026",
   "title": "Fazm - AI Computer Agent for macOS | Voice-Controlled Desktop Automation",
   "published_at": "2026-05-13T07:12:17",
   "organization": "NVIDIA",
   "model": null,
   "release_date": "2026-04-28",
   "access_type": null,
   "license": null,
   "significance": "Single open multimodal model that unifies vision, speech, and language",
   "organization_evidence": [
    "# Open source LLM releases in May 2026: one new model so far, four April tail releases, and the Swift block that decides which of them drive your Mac",
    "April 28, 2026 - NVIDIA",
    "### Nemotron 3 Nano Omni (30B / 3B active multimodal)"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Open source LLM releases in May 2026: one new model so far, four April tail releases, and the Swift block that decides which of them drive your Mac",
    "April 28, 2026 - NVIDIA",
    "### Nemotron 3 Nano Omni (30B / 3B active multimodal)"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Open source LLM releases in May 2026: one new model so far, four April tail releases, and the Swift block that decides which of them drive your Mac",
    "Single open multimodal model that unifies vision, speech, and language."
   ]
  },
  {
   "url": "https://fazm.ai/t/open-source-llm-releases-may-2026",
   "title": "Fazm - AI Computer Agent for macOS | Voice-Controlled Desktop Automation",
   "published_at": "2026-05-13T07:12:17",
   "organization": "IBM",
   "model": null,
   "release_date": "2026-04-29",
   "access_type": null,
   "license": null,
   "significance": "Three dense decoder-only models under Apache 2.0",
   "organization_evidence": [
    "# Open source LLM releases in May 2026: one new model so far, four April tail releases, and the Swift block that decides which of them drive your Mac",
    "April 29, 2026 - IBM",
    "### Granite 4.1 (3B / 8B / 30B dense)"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Open source LLM releases in May 2026: one new model so far, four April tail releases, and the Swift block that decides which of them drive your Mac",
    "April 29, 2026 - IBM",
    "### Granite 4.1 (3B / 8B / 30B dense)"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Open source LLM releases in May 2026: one new model so far, four April tail releases, and the Swift block that decides which of them drive your Mac",
    "Three dense decoder-only models under Apache 2.0."
   ]
  },
  {
   "url": "https://fazm.ai/t/open-source-llm-releases-may-2026",
   "title": "Fazm - AI Computer Agent for macOS | Voice-Controlled Desktop Automation",
   "published_at": "2026-05-13T07:12:17",
   "organization": "Mistral",
   "model": null,
   "release_date": "2026-04-29",
   "access_type": null,
   "license": null,
   "significance": "128B dense multimodal model under a Modified MIT license",
   "organization_evidence": [
    "# Open source LLM releases in May 2026: one new model so far, four April tail releases, and the Swift block that decides which of them drive your Mac",
    "April 29, 2026 - Mistral",
    "### Medium 3.5 (128B dense)"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Open source LLM releases in May 2026: one new model so far, four April tail releases, and the Swift block that decides which of them drive your Mac",
    "April 29, 2026 - Mistral",
    "### Medium 3.5 (128B dense)"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Open source LLM releases in May 2026: one new model so far, four April tail releases, and the Swift block that decides which of them drive your Mac",
    "128B dense multimodal model under a Modified MIT license, 256K token context, 77.6% on SWE-Bench Verified."
   ]
  },
  {
   "url": "https://huggingface.co/blog/FINAL-Bench/darwin-gpqa",
   "title": "\"Darwin-27B-Opus: Surpassing the Foundation Model Without Training\"",
   "published_at": "2026-04-13T01:11:13",
   "organization": "VIDRAFT",
   "model": null,
   "release_date": "2026-04-12",
   "access_type": null,
   "license": null,
   "significance": "GPQA Diamond 86.9%, World #5",
   "organization_evidence": [
    "Darwin-27B-Opus: Surpassing the Foundation Model Without Training",
    "Darwin-27B-Opus",
    "*Darwin is developed by [VIDRAFT](https://huggingface.co/FINAL-Bench). All models are released under Apache 2.0.*"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Darwin-27B-Opus: Surpassing the Foundation Model Without Training",
    "On April 12, 2026, a 27-billion-parameter model that had never undergone a single gradient update surpassed its own foundation model on one of the most demanding scientific reasoning benchmarks in existence."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Available Models",
    "| Model | Highlight | Link |",
    "| Darwin-27B-Opus | GPQA 86.9%, World #5 | [Model](https://huggingface.co/FINAL-Bench/Darwin-27B-Opus) |"
   ]
  },
  {
   "url": "https://huggingface.co/blog/FINAL-Bench/darwin-gpqa",
   "title": "\"Darwin-27B-Opus: Surpassing the Foundation Model Without Training\"",
   "published_at": "2026-04-13T01:11:13",
   "organization": "VIDRAFT",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Korean hybrid vigor, CLIcK 75.59%",
   "organization_evidence": [
    "Darwin-27B-Opus: Surpassing the Foundation Model Without Training",
    "Darwin-27B-KR",
    "*Darwin is developed by [VIDRAFT](https://huggingface.co/FINAL-Bench). All models are released under Apache 2.0.*"
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Available Models",
    "| Model | Highlight | Link |",
    "| Darwin-27B-KR | Korean hybrid vigor, CLIcK 75.59% | [Model](https://huggingface.co/FINAL-Bench/Darwin-27B-KR) |"
   ]
  },
  {
   "url": "https://huggingface.co/blog/FINAL-Bench/darwin-gpqa",
   "title": "\"Darwin-27B-Opus: Surpassing the Foundation Model Without Training\"",
   "published_at": "2026-04-13T01:11:13",
   "organization": "VIDRAFT",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "First cross-architecture FFN breeding",
   "organization_evidence": [
    "Darwin-27B-Opus: Surpassing the Foundation Model Without Training",
    "Darwin-4B-Genesis",
    "*Darwin is developed by [VIDRAFT](https://huggingface.co/FINAL-Bench). All models are released under Apache 2.0.*"
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Available Models",
    "| Model | Highlight | Link |",
    "| Darwin-4B-Genesis | First cross-architecture FFN breeding | [Model](https://huggingface.co/FINAL-Bench/Darwin-4B-Genesis) |"
   ]
  },
  {
   "url": "https://lisapedrosa.com/china-ai-flood-2026",
   "title": "The Flood: China's Four Open-Source AI Models in 12 Days",
   "published_at": "2026-05-13T10:45:23",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-04-07",
   "access_type": null,
   "license": null,
   "significance": "BenchLM score 83, MIT-licensed open-weights model within striking range of GPT-5.5",
   "organization_evidence": [
    "Between April 7 and April 24, 2026, four Chinese AI laboratories released frontier-class open-weights models.",
    "Z.ai released GLM-5.1.",
    "Section 01",
    "## Twelve Days"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The sequence began on April 7th, the same day Anthropic formally announced Mythos.",
    "Z.ai released GLM-5.1.",
    "Section 01",
    "## Twelve Days"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Kimi K2.6 scores 84. GLM-5.1 scores 83.",
    "Section 02",
    "## What They Build",
    "On BenchLM, the most widely used aggregated capability benchmark in 2026, DeepSeek V4 Pro scores 87. Kimi K2.6 scores 84. GLM-5.1 scores 83."
   ]
  },
  {
   "url": "https://lisapedrosa.com/china-ai-flood-2026",
   "title": "The Flood: China's Four Open-Source AI Models in 12 Days",
   "published_at": "2026-05-13T10:45:23",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-04-18",
   "access_type": null,
   "license": null,
   "significance": "BenchLM score 84, MIT-licensed open-weights model within striking range of GPT-5.5",
   "organization_evidence": [
    "Between April 7 and April 24, 2026, four Chinese AI laboratories released frontier-class open-weights models.",
    "On April 18th, Moonshot dropped Kimi K2.6.",
    "Section 01",
    "## Twelve Days"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "On April 18th, Moonshot dropped Kimi K2.6.",
    "Section 01",
    "## Twelve Days"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Kimi K2.6 scores 84.",
    "Section 02",
    "## What They Build",
    "On BenchLM, the most widely used aggregated capability benchmark in 2026, DeepSeek V4 Pro scores 87. Kimi K2.6 scores 84. GLM-5.1 scores 83."
   ]
  },
  {
   "url": "https://lisapedrosa.com/china-ai-flood-2026",
   "title": "The Flood: China's Four Open-Source AI Models in 12 Days",
   "published_at": "2026-05-13T10:45:23",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-21",
   "access_type": null,
   "license": null,
   "significance": "BenchLM score 87, MIT-licensed open-weights model topping Chinese leaderboards",
   "organization_evidence": [
    "Between April 7 and April 24, 2026, four Chinese AI laboratories released frontier-class open-weights models.",
    "DeepSeek, which had effectively restarted the Chinese open-source AI conversation in early 2025, completed the sequence with V4 Pro on April 21st and V4 Flash on April 24th.",
    "Section 01",
    "## Twelve Days"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "DeepSeek, which had effectively restarted the Chinese open-source AI conversation in early 2025, completed the sequence with V4 Pro on April 21st and V4 Flash on April 24th.",
    "Section 01",
    "## Twelve Days"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "87 BenchLM score \u2014 DeepSeek V4 Pro (Chinese leaderboard #1)",
    "Section 02",
    "## What They Build",
    "On BenchLM, the most widely used aggregated capability benchmark in 2026, DeepSeek V4 Pro scores 87."
   ]
  },
  {
   "url": "https://lisapedrosa.com/china-ai-flood-2026",
   "title": "The Flood: China's Four Open-Source AI Models in 12 Days",
   "published_at": "2026-05-13T10:45:23",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "BenchLM score 81, MIT-licensed open-weights model with extremely low inference pricing",
   "organization_evidence": [
    "Between April 7 and April 24, 2026, four Chinese AI laboratories released frontier-class open-weights models.",
    "DeepSeek, which had effectively restarted the Chinese open-source AI conversation in early 2025, completed the sequence with V4 Pro on April 21st and V4 Flash on April 24th.",
    "Section 01",
    "## Twelve Days"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "DeepSeek, which had effectively restarted the Chinese open-source AI conversation in early 2025, completed the sequence with V4 Pro on April 21st and V4 Flash on April 24th.",
    "Section 01",
    "## Twelve Days"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "DeepSeek V4-Flash costs $0.28 per million output tokens.",
    "Section 03",
    "## The Price Weapon"
   ]
  },
  {
   "url": "https://developersdigest.tech/blog/inkling-open-weights-thinking-machines",
   "title": "Inkling: Thinking Machines Lab Drops a 975B Open-Weights Model",
   "published_at": "2026-07-15T00:00:00",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "A 975 billion parameter open-weights model with multimodal capabilities, 1M token context, and competitive benchmarks.",
   "organization_evidence": [
    "Inkling: Thinking Machines Lab Drops a 975B Open-Weights Model",
    "Thinking Machines Lab just released Inkling, a 975 billion parameter open-[weights](/glossary#weights) model that represents the most capable American-made open model to date."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Inkling: Thinking Machines Lab Drops a 975B Open-Weights Model",
    "**Last updated:** July 15, 2026",
    "Thinking Machines Lab just released Inkling, a 975 billion parameter open-[weights](/glossary#weights) model that represents the most capable American-made open model to date."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Inkling: Thinking Machines Lab Drops a 975B Open-Weights Model",
    "A new American open-weights frontier model with multimodal capabilities, 1M token context, and competitive benchmarks. Here's what the HN community thinks.",
    "Thinking Machines Lab just released Inkling, a 975 billion parameter open-[weights](/glossary#weights) model that represents the most capable American-made open model to date."
   ]
  },
  {
   "url": "https://llm-releases.com/multimodal-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-26",
   "access_type": null,
   "license": null,
   "significance": "largest open model released to date: a 2.8T-parameter MoE",
   "organization_evidence": [
    "# Multimodal LLM releases",
    "Large language model releases with multimodal capabilities, including vision-language, audio, video, image-generation, and document-understanding models.",
    "Moonshot's flagship open-weight agentic model and the largest open model released to date: a 2.8T-parameter MoE (896 experts, 16 active per token) using Kimi Delta Attention and Attention Residuals, with native multimodal input and a 1M-token context."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Multimodal LLM releases",
    "Large language model releases with multimodal capabilities, including vision-language, audio, video, image-generation, and document-understanding models.",
    "Launched via API on Jul 16, 2026 at $3/$15 per Mtok (cached input $0.30); full open weights published to Hugging Face on Jul 26, 2026 \u2014 a day ahead of the announced Jul 27 target \u2014 under a Modified MIT license, making it freely downloadable and self-hostable."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Multimodal LLM releases",
    "Large language model releases with multimodal capabilities, including vision-language, audio, video, image-generation, and document-understanding models.",
    "Moonshot's flagship open-weight agentic model and the largest open model released to date: a 2.8T-parameter MoE (896 experts, 16 active per token) using Kimi Delta Attention and Attention Residuals, with native multimodal input and a 1M-token context."
   ]
  },
  {
   "url": "https://llm-releases.com/multimodal-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "leading U.S. open-weights release: a natively multimodal Mixture-of-Experts with 975B total / 41B active parameters",
   "organization_evidence": [
    "# Multimodal LLM releases",
    "Large language model releases with multimodal capabilities, including vision-language, audio, video, image-generation, and document-understanding models.",
    "Thinking Machines Lab's first model and the leading U.S. open-weights release: a natively multimodal Mixture-of-Experts with 975B total / 41B active parameters that reasons across text, image, and audio inputs and emits text."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Multimodal LLM releases",
    "Large language model releases with multimodal capabilities, including vision-language, audio, video, image-generation, and document-understanding models.",
    "Thinking Machines Lab's first model and the leading U.S. open-weights release: a natively multimodal Mixture-of-Experts with 975B total / 41B active parameters that reasons across text, image, and audio inputs and emits text."
   ]
  },
  {
   "url": "https://llm-releases.com/multimodal-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Google DeepMind",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "text-diffusion model built on the Gemma 4 26B-A4B MoE backbone",
   "organization_evidence": [
    "# Multimodal LLM releases",
    "Large language model releases with multimodal capabilities, including vision-language, audio, video, image-generation, and document-understanding models.",
    "An open-weight text-diffusion model built on the Gemma 4 26B-A4B MoE backbone (25.2B total / 3.8B active)."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Multimodal LLM releases",
    "Large language model releases with multimodal capabilities, including vision-language, audio, video, image-generation, and document-understanding models.",
    "An open-weight text-diffusion model built on the Gemma 4 26B-A4B MoE backbone (25.2B total / 3.8B active)."
   ]
  },
  {
   "url": "https://llm-releases.com/multimodal-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Google DeepMind",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "dense 12B member of the Gemma 4 family with a unified, encoder-free multimodal architecture",
   "organization_evidence": [
    "# Multimodal LLM releases",
    "Large language model releases with multimodal capabilities, including vision-language, audio, video, image-generation, and document-understanding models.",
    "A dense 12B member of the Gemma 4 family with a unified, encoder-free multimodal architecture: vision and audio are projected straight into the LLM backbone."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Multimodal LLM releases",
    "Large language model releases with multimodal capabilities, including vision-language, audio, video, image-generation, and document-understanding models.",
    "A dense 12B member of the Gemma 4 family with a unified, encoder-free multimodal architecture: vision and audio are projected straight into the LLM backbone."
   ]
  },
  {
   "url": "https://llm-releases.com/multimodal-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Nex AGI",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "open-weight agentic flagship, post-trained on Qwen3.5-397B-A17B",
   "organization_evidence": [
    "# Multimodal LLM releases",
    "Large language model releases with multimodal capabilities, including vision-language, audio, video, image-generation, and document-understanding models.",
    "Nex AGI's open-weight agentic flagship, post-trained on Qwen3.5-397B-A17B (397B total / ~17B active MoE) by the Shanghai Innovation Institute-led Nex alliance."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Multimodal LLM releases",
    "Large language model releases with multimodal capabilities, including vision-language, audio, video, image-generation, and document-understanding models.",
    "Nex AGI's open-weight agentic flagship, post-trained on Qwen3.5-397B-A17B (397B total / ~17B active MoE) by the Shanghai Innovation Institute-led Nex alliance."
   ]
  },
  {
   "url": "https://llm-releases.com/multimodal-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "StepFun",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "high-efficiency multimodal sparse-MoE successor to Step-3.5-Flash",
   "organization_evidence": [
    "# Multimodal LLM releases",
    "Large language model releases with multimodal capabilities, including vision-language, audio, video, image-generation, and document-understanding models.",
    "StepFun's high-efficiency multimodal sparse-MoE successor to Step-3.5-Flash: a ~196B-total / ~11B-active vision-language model with native image and video understanding, a 256K context, and selectable reasoning tiers (high/medium/low)."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Multimodal LLM releases",
    "Large language model releases with multimodal capabilities, including vision-language, audio, video, image-generation, and document-understanding models.",
    "StepFun's high-efficiency multimodal sparse-MoE successor to Step-3.5-Flash: a ~196B-total / ~11B-active vision-language model with native image and video understanding, a 256K context, and selectable reasoning tiers (high/medium/low)."
   ]
  },
  {
   "url": "https://llm-releases.com/multimodal-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Google DeepMind",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "advanced-reasoning open model for personal computers",
   "organization_evidence": [
    "# Multimodal LLM releases",
    "Large language model releases with multimodal capabilities, including vision-language, audio, video, image-generation, and document-understanding models.",
    "Google DeepMind's Gemma 4 advanced-reasoning open model for personal computers, part of the April 2026 Gemma 4 family."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Multimodal LLM releases",
    "Large language model releases with multimodal capabilities, including vision-language, audio, video, image-generation, and document-understanding models.",
    "Google DeepMind's Gemma 4 advanced-reasoning open model for personal computers, part of the April 2026 Gemma 4 family."
   ]
  },
  {
   "url": "https://llm-releases.com/multimodal-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Alibaba (Qwen)",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "most intelligent model under 10B parameters at launch",
   "organization_evidence": [
    "# Multimodal LLM releases",
    "Large language model releases with multimodal capabilities, including vision-language, audio, video, image-generation, and document-understanding models.",
    "The flagship of Alibaba's small dense Qwen3.5 models."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Multimodal LLM releases",
    "Large language model releases with multimodal capabilities, including vision-language, audio, video, image-generation, and document-understanding models.",
    "The flagship of Alibaba's small dense Qwen3.5 models."
   ]
  },
  {
   "url": "https://llm-releases.com/multimodal-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Alibaba (Qwen)",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "most intelligent model under 5B parameters at launch",
   "organization_evidence": [
    "# Multimodal LLM releases",
    "Large language model releases with multimodal capabilities, including vision-language, audio, video, image-generation, and document-understanding models.",
    "A dense 4B in Alibaba's small Qwen3.5 family, rated by Artificial Analysis as the most intelligent model under 5B parameters at launch \u2014 outscoring several 7B\u20139B peers despite roughly half the parameters."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Multimodal LLM releases",
    "Large language model releases with multimodal capabilities, including vision-language, audio, video, image-generation, and document-understanding models.",
    "A dense 4B in Alibaba's small Qwen3.5 family, rated by Artificial Analysis as the most intelligent model under 5B parameters at launch \u2014 outscoring several 7B\u20139B peers despite roughly half the parameters."
   ]
  },
  {
   "url": "https://llm-releases.com/multimodal-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Alibaba (Qwen)",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "built for high-throughput, low-latency edge and on-device use",
   "organization_evidence": [
    "# Multimodal LLM releases",
    "Large language model releases with multimodal capabilities, including vision-language, audio, video, image-generation, and document-understanding models.",
    "A dense 2B Qwen3.5 model built for high-throughput, low-latency edge and on-device use."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Multimodal LLM releases",
    "Large language model releases with multimodal capabilities, including vision-language, audio, video, image-generation, and document-understanding models.",
    "A dense 2B Qwen3.5 model built for high-throughput, low-latency edge and on-device use."
   ]
  },
  {
   "url": "https://llm-releases.com/multimodal-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Alibaba (Qwen)",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "designed for the most constrained on-device deployments",
   "organization_evidence": [
    "# Multimodal LLM releases",
    "Large language model releases with multimodal capabilities, including vision-language, audio, video, image-generation, and document-understanding models.",
    "The smallest Qwen3.5 model \u2014 a dense 0.8B designed for the most constrained on-device deployments, operating in non-thinking (instruct) mode by default."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Multimodal LLM releases",
    "Large language model releases with multimodal capabilities, including vision-language, audio, video, image-generation, and document-understanding models.",
    "The smallest Qwen3.5 model \u2014 a dense 0.8B designed for the most constrained on-device deployments, operating in non-thinking (instruct) mode by default."
   ]
  },
  {
   "url": "https://llm-releases.com/multimodal-models",
   "title": "LLM Releases",
   "published_at": null,
   "organization": "Mistral AI",
   "model": null,
   "release_date": "2026-03",
   "access_type": null,
   "license": null,
   "significance": "first Mistral model to unify reasoning, multimodal understanding, and agentic coding",
   "organization_evidence": [
    "# Multimodal LLM releases",
    "Large language model releases with multimodal capabilities, including vision-language, audio, video, image-generation, and document-understanding models.",
    "Mistral's March 2026 Small release: the first Mistral model to unify reasoning (Magistral), multimodal understanding (Pixtral), and agentic coding (Devstral) into one Apache 2.0 model."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Multimodal LLM releases",
    "Large language model releases with multimodal capabilities, including vision-language, audio, video, image-generation, and document-understanding models.",
    "Mistral's March 2026 Small release: the first Mistral model to unify reasoning (Magistral), multimodal understanding (Pixtral), and agentic coding (Devstral) into one Apache 2.0 model."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Multimodal LLM releases",
    "Large language model releases with multimodal capabilities, including vision-language, audio, video, image-generation, and document-understanding models.",
    "Mistral's March 2026 Small release: the first Mistral model to unify reasoning (Magistral), multimodal understanding (Pixtral), and agentic coding (Devstral) into one Apache 2.0 model."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/qwen3-5-122b-a10b",
   "title": "Qwen3.5 122B A10B - Intelligence, Performance & Price Analysis",
   "published_at": null,
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-02-24",
   "access_type": null,
   "license": null,
   "significance": "Supports text and image input, outputs text, and has a 262k tokens context window.",
   "organization_evidence": [
    "Qwen3.5 122B A10B (Reasoning) Intelligence, Performance & Price Analysis",
    "Qwen3.5 122B A10B (Reasoning) was created by Alibaba."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Qwen3.5 122B A10B (Reasoning) Intelligence, Performance & Price Analysis",
    "Qwen3.5 122B A10B (Reasoning) was released on February 24, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Qwen3.5 122B A10B (Reasoning) Intelligence, Performance & Price Analysis",
    "It's also faster than average, however very verbose. The model supports text and image input, outputs text, and has a 262k tokens context window."
   ]
  },
  {
   "url": "https://nanomachines.net/business-markets/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - NanoMachines",
   "published_at": "2026-07-15T12:08:03",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor",
   "organization_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "China\u2019s Release Cadence Is the Story",
    "Between April 24 and June 16, 2026, Chinese laboratories launched four major open-weight models: DeepSeek V4, MiniMax M3, Kimi K2.7-Code, and GLM-5.2."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "China\u2019s Release Cadence Is the Story",
    "Between April 24 and June 16, 2026, Chinese laboratories launched four major open-weight models: DeepSeek V4, MiniMax M3, Kimi K2.7-Code, and GLM-5.2.",
    "APR 24",
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "China\u2019s Release Cadence Is the Story",
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor"
   ]
  },
  {
   "url": "https://nanomachines.net/business-markets/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - NanoMachines",
   "published_at": "2026-07-15T12:08:03",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "cheap 1M-token context, native multimodal, modified-MIT",
   "organization_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "China\u2019s Release Cadence Is the Story",
    "Between April 24 and June 16, 2026, Chinese laboratories launched four major open-weight models: DeepSeek V4, MiniMax M3, Kimi K2.7-Code, and GLM-5.2."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "China\u2019s Release Cadence Is the Story",
    "JUN 01"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "China\u2019s Release Cadence Is the Story"
   ]
  },
  {
   "url": "https://nanomachines.net/business-markets/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - NanoMachines",
   "published_at": "2026-07-15T12:08:03",
   "organization": "Moonshot",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "agent-run specialist, ~30% fewer thinking tokens than K2.6",
   "organization_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "China\u2019s Release Cadence Is the Story",
    "Between April 24 and June 16, 2026, Chinese laboratories launched four major open-weight models: DeepSeek V4, MiniMax M3, Kimi K2.7-Code, and GLM-5.2."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "China\u2019s Release Cadence Is the Story",
    "JUN 13"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "China\u2019s Release Cadence Is the Story"
   ]
  },
  {
   "url": "https://nanomachines.net/business-markets/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - NanoMachines",
   "published_at": "2026-07-15T12:08:03",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "753B MoE, MIT, top open-weight on Artificial Analysis index",
   "organization_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "China\u2019s Release Cadence Is the Story",
    "Between April 24 and June 16, 2026, Chinese laboratories launched four major open-weight models: DeepSeek V4, MiniMax M3, Kimi K2.7-Code, and GLM-5.2."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "China\u2019s Release Cadence Is the Story",
    "JUN 13\u201316"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "China\u2019s Release Cadence Is the Story"
   ]
  },
  {
   "url": "https://fireworks.ai/blog/best-open-source-llms?from_theconsensus=1",
   "title": "Best Open Source LLMs in 2026",
   "published_at": "2026-07-08T00:00:00",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "Strongest benchmark profile among currently open models in this set",
   "organization_evidence": [
    "The best open source LLMs at a glance",
    "## GLM 5.2"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The best open source LLMs at a glance",
    "## GLM 5.2",
    "### What is GLM 5.2?",
    "- \u2022**Release date:** June 2026",
    "| [GLM 5.2](/models/fireworks/glm-5p2) | June 2026 | 743B total | 1,040k tokens | Strongest benchmark profile among currently open models in this set. First eval for broad reasoning, coding, and long-context agents. | [Try in playground](/models/fireworks/glm-5p2) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The best open source LLMs at a glance",
    "## GLM 5.2",
    "| [GLM 5.2](/models/fireworks/glm-5p2) | June 2026 | 743B total | 1,040k tokens | Strongest benchmark profile among currently open models in this set. First eval for broad reasoning, coding, and long-context agents. | [Try in playground](/models/fireworks/glm-5p2) |"
   ]
  },
  {
   "url": "https://fireworks.ai/blog/best-open-source-llms?from_theconsensus=1",
   "title": "Best Open Source LLMs in 2026",
   "published_at": "2026-07-08T00:00:00",
   "organization": "Moonshot",
   "model": null,
   "release_date": "2026-07-16",
   "access_type": null,
   "license": null,
   "significance": "Benchmark leader awaiting full weights and a Fireworks listing",
   "organization_evidence": [
    "The best open source LLMs at a glance",
    "## Kimi K3",
    "[Kimi K3](https://www.kimi.com/blog/kimi-k3) is Moonshot's 2.8-trillion-parameter mixture-of-experts model."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The best open source LLMs at a glance",
    "## Kimi K3",
    "### What is Kimi K3?",
    "- \u2022**Release date:** July 16, 2026"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The best open source LLMs at a glance",
    "## Kimi K3",
    "| [Kimi K3](https://www.kimi.com/blog/kimi-k3) | July 2026 | 2.8T total, 16 of 896 experts active | 1M tokens | Benchmark leader awaiting full weights and a Fireworks listing. License details remain unpublished. | Not yet listed |"
   ]
  },
  {
   "url": "https://fireworks.ai/blog/best-open-source-llms?from_theconsensus=1",
   "title": "Best Open Source LLMs in 2026",
   "published_at": "2026-07-08T00:00:00",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "Coding agents, repository work, patch planning, and multimodal developer tools",
   "organization_evidence": [
    "The best open source LLMs at a glance",
    "## Kimi K2.7 Code",
    "[Kimi K2.7 Code](/models/fireworks/kimi-k2p7-code) is a coding-specialized 1.02T-parameter MoE model from [Moonshot AI](https://www.moonshot.ai/) with a 262k-token context window."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The best open source LLMs at a glance",
    "## Kimi K2.7 Code",
    "| [Kimi K2.7 Code](/models/fireworks/kimi-k2p7-code) | June 2026 | 1.02T total | 262k tokens | Coding agents, repository work, patch planning, and multimodal developer tools. Thinking mode is mandatory. | [Try in playground](/models/fireworks/kimi-k2p7-code) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The best open source LLMs at a glance",
    "## Kimi K2.7 Code",
    "| [Kimi K2.7 Code](/models/fireworks/kimi-k2p7-code) | June 2026 | 1.02T total | 262k tokens | Coding agents, repository work, patch planning, and multimodal developer tools. Thinking mode is mandatory. | [Try in playground](/models/fireworks/kimi-k2p7-code) |"
   ]
  },
  {
   "url": "https://fireworks.ai/blog/best-open-source-llms?from_theconsensus=1",
   "title": "Best Open Source LLMs in 2026",
   "published_at": "2026-07-08T00:00:00",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-01",
   "access_type": null,
   "license": null,
   "significance": "Long-context reasoning and coding from a separate open-source family",
   "organization_evidence": [
    "The best open source LLMs at a glance",
    "## DeepSeek-V4-Pro",
    "[DeepSeek-V4-Pro](/models/fireworks/deepseek-v4-pro) is a large MoE model for reasoning, coding, and long-context work from [DeepSeek](https://www.deepseek.com/)."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The best open source LLMs at a glance",
    "## DeepSeek-V4-Pro",
    "| [DeepSeek-V4-Pro](/models/fireworks/deepseek-v4-pro) | April 2026 | 1.6T total | 1,040k tokens | Long-context reasoning and coding from a separate open-source family. Second eval when GLM 5.2 misses on your repo. | [Try in playground](/models/fireworks/deepseek-v4-pro) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The best open source LLMs at a glance",
    "## DeepSeek-V4-Pro",
    "| [DeepSeek-V4-Pro](/models/fireworks/deepseek-v4-pro) | April 2026 | 1.6T total | 1,040k tokens | Long-context reasoning and coding from a separate open-source family. Second eval when GLM 5.2 misses on your repo. | [Try in playground](/models/fireworks/deepseek-v4-pro) |"
   ]
  },
  {
   "url": "https://fireworks.ai/blog/best-open-source-llms?from_theconsensus=1",
   "title": "Best Open Source LLMs in 2026",
   "published_at": "2026-07-08T00:00:00",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-01",
   "access_type": null,
   "license": null,
   "significance": "Same 1,040k context class as Pro at higher throughput and lower cost",
   "organization_evidence": [
    "The best open source LLMs at a glance",
    "## DeepSeek-V4-Flash",
    "[DeepSeek-V4-Flash](/models/fireworks/deepseek-v4-flash) is the faster V4 route for long-context reasoning and agent work."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The best open source LLMs at a glance",
    "## DeepSeek-V4-Flash",
    "| [DeepSeek-V4-Flash](/models/fireworks/deepseek-v4-flash) | April 2026 | 284B total | 1,040k tokens | Same 1,040k context class as Pro at higher throughput and lower cost. Default DeepSeek route for high-volume workloads. | [Try in playground](/models/fireworks/deepseek-v4-flash) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The best open source LLMs at a glance",
    "## DeepSeek-V4-Flash",
    "| [DeepSeek-V4-Flash](/models/fireworks/deepseek-v4-flash) | April 2026 | 284B total | 1,040k tokens | Same 1,040k context class as Pro at higher throughput and lower cost. Default DeepSeek route for high-volume workloads. | [Try in playground](/models/fireworks/deepseek-v4-flash) |"
   ]
  },
  {
   "url": "https://fireworks.ai/blog/best-open-source-llms?from_theconsensus=1",
   "title": "Best Open Source LLMs in 2026",
   "published_at": "2026-07-08T00:00:00",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "Native image and video input, second-highest GPQA score in the table",
   "organization_evidence": [
    "The best open source LLMs at a glance",
    "## MiniMax M3",
    "[MiniMax M3](/models/fireworks/minimax-m3) is a native multimodal MoE model built for long-context, agentic, and multimodal workflows."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The best open source LLMs at a glance",
    "## MiniMax M3",
    "| [MiniMax M3](/models/fireworks/minimax-m3) | June 202622, 2025 | 428B total, about 23B activated | 512k tokens | Native image and video input, second-highest GPQA score in the table. First eval when multimodality sets the constraint. | [Try in playground](/models/fireworks/minimax-m3) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The best open source LLMs at a glance",
    "## MiniMax M3",
    "| [MiniMax M3](/models/fireworks/minimax-m3) | June 202622, 2025 | 428B total, about 23B activated | 512k tokens | Native image and video input, second-highest GPQA score in the table. First eval when multimodality sets the constraint. | [Try in playground](/models/fireworks/minimax-m3) |"
   ]
  },
  {
   "url": "https://fireworks.ai/blog/best-open-source-llms?from_theconsensus=1",
   "title": "Best Open Source LLMs in 2026",
   "published_at": "2026-07-08T00:00:00",
   "organization": "Google",
   "model": null,
   "release_date": "2026-04-01",
   "access_type": null,
   "license": null,
   "significance": "Smaller dense multimodal model with Apache-2.0 licensing and an On-Demand deployment path",
   "organization_evidence": [
    "The best open source LLMs at a glance",
    "## Gemma 4 31B IT",
    "[Gemma 4 31B IT](/models/fireworks/gemma-4-31b-it) is a smaller multimodal instruction-tuned model from Google."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The best open source LLMs at a glance",
    "## Gemma 4 31B IT",
    "| [Gemma 4 31B IT](/models/fireworks/gemma-4-31b-it) | April 2026 | 32.2B dense | 262k tokens | Smaller dense multimodal model with Apache-2.0 licensing and an On-Demand deployment path. First eval when adaptation and deployment control matter more than frontier benchmark rank. | [View On-Demand model](/models/fireworks/gemma-4-31b-it) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The best open source LLMs at a glance",
    "## Gemma 4 31B IT",
    "| [Gemma 4 31B IT](/models/fireworks/gemma-4-31b-it) | April 2026 | 32.2B dense | 262k tokens | Smaller dense multimodal model with Apache-2.0 licensing and an On-Demand deployment path. First eval when adaptation and deployment control matter more than frontier benchmark rank. | [View On-Demand model](/models/fireworks/gemma-4-31b-it) |"
   ]
  },
  {
   "url": "https://nerdleveltech.com/minimax-m3-open-weight-coding-sparse-attention",
   "title": "MiniMax M3: Open-Weight Coding at 1/10 the Cost (2026)",
   "published_at": "2026-06-09T00:00:00",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "pairs frontier-level coding with a 1-million-token context window and native multimodality",
   "organization_evidence": [
    "ai-ml",
    "# MiniMax M3: Open-Weight Coding at 1/10 the Cost (2026)",
    "MiniMax M3 is a large language model from **MiniMax**, the Chinese AI company founded in 2022 and backed by Alibaba and Tencent that listed in Hong Kong in January 2026.10"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "ai-ml",
    "# MiniMax M3: Open-Weight Coding at 1/10 the Cost (2026)",
    "June 9, 2026",
    "MiniMax M3 is a Chinese open-weight language model, released June 1, 2026, that pairs frontier-level coding with a 1-million-token context window and native multimodality \u2014 at roughly a tenth of the per-token price of GPT-5.5 or Claude Opus.12"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "ai-ml",
    "# MiniMax M3: Open-Weight Coding at 1/10 the Cost (2026)",
    "MiniMax M3 is a Chinese open-weight language model, released June 1, 2026, that pairs frontier-level coding with a 1-million-token context window and native multimodality \u2014 at roughly a tenth of the per-token price of GPT-5.5 or Claude Opus.12"
   ]
  },
  {
   "url": "https://huggingface.co/blog/mlabonne/kimik25",
   "title": "Kimi K2.5: Still Worth It After Two Weeks?",
   "published_at": "2026-02-23T11:54:28",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-01-27",
   "access_type": null,
   "license": null,
   "significance": "1.04 trillion parameters with 32B activated parameters per token, featuring Agent Swarm and a 256K context window.",
   "organization_evidence": [
    "Beijing-based Moonshot AI released **Kimi K2.5** on January 27, 2026."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Beijing-based Moonshot AI released **Kimi K2.5** on January 27, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Beijing-based Moonshot AI released **Kimi K2.5** on January 27, 2026.",
    "Kimi K2.5 is one of the largest open-weight models with 1.04 trillion parameters and 32B activated parameters per token."
   ]
  },
  {
   "url": "https://techcrunch.com/2026/04/24/deepseek-previews-new-ai-model-that-closes-the-gap-with-frontier-models",
   "title": "DeepSeek previews new AI model that \u2018closes the gap\u2019 with frontier models",
   "published_at": "2026-04-24T13:30:59",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "Mixture-of-experts models with 1 million token context windows, featuring 1.6 trillion parameters for the Pro model and 284 billion for the Flash model, outperforming previous versions and approaching frontier model performance.",
   "organization_evidence": [
    "Chinese AI lab DeepSeek has launched two preview versions of its newest large language model, [DeepSeek V4](https://huggingface.co/collections/deepseek-ai/deepseek-v4), a much-awaited update to last year\u2019s V3.2 model and the accompanying [R1 reasoning model](https://techcrunch.com/2025/01/27/deepseek-claims-its-reasoning-model-beats-openais-o1-on-certain-benchmarks/) that took the [AI world by storm](https://techcrunch.com/2025/09/29/deepseek-everything-you-need-to-know-about-the-ai-chatbot-app/)."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "![](https://techcrunch.com/wp-content/uploads/2026/04/260424_18h44m20s_screenshot.png?w=680)",
    "Chinese AI lab DeepSeek has launched two preview versions of its newest large language model, [DeepSeek V4](https://huggingface.co/collections/deepseek-ai/deepseek-v4), a much-awaited update to last year\u2019s V3.2 model and the accompanying [R1 reasoning model](https://techcrunch.com/2025/01/27/deepseek-claims-its-reasoning-model-beats-openais-o1-on-certain-benchmarks/) that took the [AI world by storm](https://techcrunch.com/2025/09/29/deepseek-everything-you-need-to-know-about-the-ai-chatbot-app/)."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The company says both DeepSeek V4 Flash and V4 Pro are mixture-of-experts models with context windows of 1 million tokens each \u2014 enough to allow large codebases or documents to be used in prompts.",
    "Chinese AI lab DeepSeek has launched two preview versions of its newest large language model, [DeepSeek V4](https://huggingface.co/collections/deepseek-ai/deepseek-v4), a much-awaited update to last year\u2019s V3.2 model and the accompanying [R1 reasoning model](https://techcrunch.com/2025/01/27/deepseek-claims-its-reasoning-model-beats-openais-o1-on-certain-benchmarks/) that took the [AI world by storm](https://techcrunch.com/2025/09/29/deepseek-everything-you-need-to-know-about-the-ai-chatbot-app/)."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/articles?page=5",
   "title": "Articles | Artificial Analysis",
   "published_at": null,
   "organization": "Cohere",
   "model": null,
   "release_date": "2026-05-20",
   "access_type": null,
   "license": null,
   "significance": "more than a year since the Command A release",
   "organization_evidence": [
    "# Articles",
    "Cohere launches open weights model Command A+, more than a year since the Command A release"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Articles",
    "Cohere launches open weights model Command A+, more than a year since the Command A release"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Articles",
    "Cohere launches open weights model Command A+, more than a year since the Command A release"
   ]
  },
  {
   "url": "https://artificialanalysis.ai/articles?page=5",
   "title": "Articles | Artificial Analysis",
   "published_at": null,
   "organization": "Cohere",
   "model": "Command A+",
   "release_date": "2026-05-20",
   "access_type": "open weights",
   "license": null,
   "significance": "more than a year since the Command A release",
   "organization_evidence": [
    "# Articles",
    "Cohere launches open weights model Command A+, more than a year since the Command A release"
   ],
   "model_evidence": [
    "# Articles",
    "Cohere launches open weights model Command A+, more than a year since the Command A release"
   ],
   "release_date_evidence": [
    "# Articles",
    "Cohere launches open weights model Command A+, more than a year since the Command A release"
   ],
   "access_type_evidence": [
    "# Articles",
    "Cohere launches open weights model Command A+, more than a year since the Command A release"
   ],
   "license_evidence": null,
   "significance_evidence": [
    "# Articles",
    "Cohere launches open weights model Command A+, more than a year since the Command A release"
   ]
  },
  {
   "url": "https://artificialanalysis.ai/articles?page=5",
   "title": "Articles | Artificial Analysis",
   "published_at": null,
   "organization": "OpenBMB",
   "model": "MiniCPM-V 4.6 1.3B Instruct",
   "release_date": "2026-05-11",
   "access_type": null,
   "license": null,
   "significance": null,
   "organization_evidence": [
    "# Articles",
    "OpenBMB launches MiniCPM-V 4.6 1.3B Instruct"
   ],
   "model_evidence": [
    "# Articles",
    "OpenBMB launches MiniCPM-V 4.6 1.3B Instruct"
   ],
   "release_date_evidence": [
    "# Articles",
    "OpenBMB launches MiniCPM-V 4.6 1.3B Instruct"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": null
  },
  {
   "url": "https://artificialanalysis.ai/articles?page=5",
   "title": "Articles | Artificial Analysis",
   "published_at": null,
   "organization": "DeepSeek",
   "model": "V4 Pro and V4 Flash",
   "release_date": "2026-04-24",
   "access_type": "open weights",
   "license": null,
   "significance": "back among the leading open weights models",
   "organization_evidence": [
    "# Articles",
    "DeepSeek is back among the leading open weights models with V4 Pro and V4 Flash"
   ],
   "model_evidence": [
    "# Articles",
    "DeepSeek is back among the leading open weights models with V4 Pro and V4 Flash"
   ],
   "release_date_evidence": [
    "# Articles",
    "DeepSeek is back among the leading open weights models with V4 Pro and V4 Flash"
   ],
   "access_type_evidence": [
    "# Articles",
    "DeepSeek is back among the leading open weights models with V4 Pro and V4 Flash"
   ],
   "license_evidence": null,
   "significance_evidence": [
    "# Articles",
    "DeepSeek is back among the leading open weights models with V4 Pro and V4 Flash"
   ]
  },
  {
   "url": "https://artificialanalysis.ai/articles?page=5",
   "title": "Articles | Artificial Analysis",
   "published_at": null,
   "organization": null,
   "model": "Kimi K2.6",
   "release_date": "2026-04-20",
   "access_type": "open weights",
   "license": null,
   "significance": "The new leading open weights model",
   "organization_evidence": null,
   "model_evidence": [
    "# Articles",
    "Kimi K2.6: The new leading open weights model"
   ],
   "release_date_evidence": [
    "# Articles",
    "Kimi K2.6: The new leading open weights model"
   ],
   "access_type_evidence": [
    "# Articles",
    "Kimi K2.6: The new leading open weights model"
   ],
   "license_evidence": null,
   "significance_evidence": [
    "# Articles",
    "Kimi K2.6: The new leading open weights model"
   ]
  },
  {
   "url": "https://leftbrainmarketing.net/ai-tooling/inside-china-s-fast-paced-ai-model-deployment-signal-s-four-open-versions",
   "title": "Inside China\u2019s Fast-Paced AI Model Deployment: Signal\u2019s Four Open Versions - leftbrainmarketing.net",
   "published_at": "2026-07-15T06:14:13",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor",
   "organization_evidence": [
    "Inside China\u2019s Fast-Paced AI Model Deployment: Signal\u2019s Four Open Versions",
    "The production line \u2014 spring 2026",
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Inside China\u2019s Fast-Paced AI Model Deployment: Signal\u2019s Four Open Versions",
    "The production line \u2014 spring 2026",
    "APR 24",
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Inside China\u2019s Fast-Paced AI Model Deployment: Signal\u2019s Four Open Versions",
    "The production line \u2014 spring 2026",
    "APR 24",
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor"
   ]
  },
  {
   "url": "https://leftbrainmarketing.net/ai-tooling/inside-china-s-fast-paced-ai-model-deployment-signal-s-four-open-versions",
   "title": "Inside China\u2019s Fast-Paced AI Model Deployment: Signal\u2019s Four Open Versions - leftbrainmarketing.net",
   "published_at": "2026-07-15T06:14:13",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "cheap 1M-token context, native multimodal, modified-MIT",
   "organization_evidence": [
    "Inside China\u2019s Fast-Paced AI Model Deployment: Signal\u2019s Four Open Versions",
    "The production line \u2014 spring 2026",
    "JUN 01"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Inside China\u2019s Fast-Paced AI Model Deployment: Signal\u2019s Four Open Versions",
    "The production line \u2014 spring 2026",
    "JUN 01"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Inside China\u2019s Fast-Paced AI Model Deployment: Signal\u2019s Four Open Versions",
    "The production line \u2014 spring 2026",
    "JUN 01"
   ]
  },
  {
   "url": "https://leftbrainmarketing.net/ai-tooling/inside-china-s-fast-paced-ai-model-deployment-signal-s-four-open-versions",
   "title": "Inside China\u2019s Fast-Paced AI Model Deployment: Signal\u2019s Four Open Versions - leftbrainmarketing.net",
   "published_at": "2026-07-15T06:14:13",
   "organization": "Moonshot",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "agent-run specialist, ~30% fewer thinking tokens than K2.6",
   "organization_evidence": [
    "Inside China\u2019s Fast-Paced AI Model Deployment: Signal\u2019s Four Open Versions",
    "The production line \u2014 spring 2026",
    "JUN 13"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Inside China\u2019s Fast-Paced AI Model Deployment: Signal\u2019s Four Open Versions",
    "The production line \u2014 spring 2026",
    "JUN 13"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Inside China\u2019s Fast-Paced AI Model Deployment: Signal\u2019s Four Open Versions",
    "The production line \u2014 spring 2026",
    "JUN 13"
   ]
  },
  {
   "url": "https://leftbrainmarketing.net/ai-tooling/inside-china-s-fast-paced-ai-model-deployment-signal-s-four-open-versions",
   "title": "Inside China\u2019s Fast-Paced AI Model Deployment: Signal\u2019s Four Open Versions - leftbrainmarketing.net",
   "published_at": "2026-07-15T06:14:13",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "753B MoE, MIT, top open-weight on Artificial Analysis index",
   "organization_evidence": [
    "Inside China\u2019s Fast-Paced AI Model Deployment: Signal\u2019s Four Open Versions",
    "The production line \u2014 spring 2026",
    "JUN 13\u201316"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Inside China\u2019s Fast-Paced AI Model Deployment: Signal\u2019s Four Open Versions",
    "The production line \u2014 spring 2026",
    "JUN 13\u201316"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Inside China\u2019s Fast-Paced AI Model Deployment: Signal\u2019s Four Open Versions",
    "The production line \u2014 spring 2026",
    "JUN 13\u201316"
   ]
  },
  {
   "url": "https://promptzone.com/ai-model-releases",
   "title": "Latest AI Model Releases: July 2026 Timeline \u2014 PromptZone - Leading AI Community for Prompt Engineering and AI Enthusiasts",
   "published_at": null,
   "organization": "Zhipu AI",
   "model": null,
   "release_date": "Jun 16, 2026",
   "access_type": null,
   "license": null,
   "significance": "Open-weights flagship that topped Hugging Face trending and runs on surprisingly modest hardware.",
   "organization_evidence": [
    "A chronological timeline of major AI model releases across 2025 and 2026 \u2014 LLMs, reasoning models, open-weights families, and image & video generators, each with its date, why it mattered, and the official announcement.",
    "Hand-curated \u2014 last verified: July 29, 2026",
    "## June 2026",
    "3. Jun 16, 2026 \u00b7 Zhipu AI"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "A chronological timeline of major AI model releases across 2025 and 2026 \u2014 LLMs, reasoning models, open-weights families, and image & video generators, each with its date, why it mattered, and the official announcement.",
    "Hand-curated \u2014 last verified: July 29, 2026",
    "## June 2026",
    "3. Jun 16, 2026 \u00b7 Zhipu AI"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "A chronological timeline of major AI model releases across 2025 and 2026 \u2014 LLMs, reasoning models, open-weights families, and image & video generators, each with its date, why it mattered, and the official announcement.",
    "Hand-curated \u2014 last verified: July 29, 2026",
    "## June 2026",
    "Open-weights flagship that topped Hugging Face trending and runs on surprisingly modest hardware."
   ]
  },
  {
   "url": "https://medium.com/@augustozz/thinking-machines-launches-inkling-a-new-open-weight-model-and-a-game-changer-in-ai-7b501a86daaf",
   "title": "Thinking Machines Launches Inkling: A New Open Weight Model and A Game Changer in AI",
   "published_at": "2026-07-16T14:10:40",
   "organization": "Thinking Machines",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "Scores 41 on Artificial Analysis\u2019 Intelligence Index, the highest of any US built open weight model",
   "organization_evidence": [
    "Thinking Machines Launches Inkling: A New Open Weight Model and A Game Changer in AI",
    "Thinking Machines just released Inkling, a free open weight AI model. Here\u2019s what open weight AI means, how Inkling compares to GPT and Claude, and why it matters."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Thinking Machines Launches Inkling: A New Open Weight Model and A Game Changer in AI",
    "- Released July 15, 2026, under a fully open Apache 2.0 license"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Thinking Machines Launches Inkling: A New Open Weight Model and A Game Changer in AI",
    "- Scores 41 on Artificial Analysis\u2019 Intelligence Index, the highest of any US built open weight model, but still behind Claude Opus 4.8 (56) and GPT-5.6 (59)"
   ]
  },
  {
   "url": "https://venturebeat.com/technology/cerebras-says-its-chips-run-a-trillion-parameter-ai-model-nearly-7-times-faster-than-gpu-clouds",
   "title": "Cerebras says its chips run a trillion-parameter AI model nearly 7 times faster than GPU clouds",
   "published_at": "2026-05-20T19:59:22",
   "organization": "Moonshot AI",
   "model": "Kimi K2.6",
   "release_date": "2026-04-20",
   "access_type": "open-weight",
   "license": null,
   "significance": "trillion-parameter Mixture-of-Experts model that tops SWE-Bench Pro at 58.6, matching GPT-5.4",
   "organization_evidence": [
    "On Monday, the Sunnyvale-based chipmaker announced that it is now running [Kimi K2.6](https://www.kimi.com/ai-models/kimi-k2-6) \u2014 a trillion-parameter open-weight model developed by Beijing-based [Moonshot AI](https://www.moonshot.ai/) \u2014 for enterprise customers at nearly 1,000 tokens per second, a speed no GPU-based provider has come close to matching.",
    "Released on April 20 by Moonshot AI \u2014 a Beijing-based company founded in 2023 by Tsinghua University alumni and dubbed one of China's \"AI Tiger\" companies \u2014 K2.6 is a trillion-parameter Mixture-of-Experts model that has rapidly established itself as the most capable open-weight model available for coding and agentic tasks."
   ],
   "model_evidence": [
    "On Monday, the Sunnyvale-based chipmaker announced that it is now running [Kimi K2.6](https://www.kimi.com/ai-models/kimi-k2-6) \u2014 a trillion-parameter open-weight model developed by Beijing-based [Moonshot AI](https://www.moonshot.ai/) \u2014 for enterprise customers at nearly 1,000 tokens per second, a speed no GPU-based provider has come close to matching.",
    "Released on April 20 by Moonshot AI \u2014 a Beijing-based company founded in 2023 by Tsinghua University alumni and dubbed one of China's \"AI Tiger\" companies \u2014 K2.6 is a trillion-parameter Mixture-of-Experts model that has rapidly established itself as the most capable open-weight model available for coding and agentic tasks."
   ],
   "release_date_evidence": [
    "Released on April 20 by Moonshot AI \u2014 a Beijing-based company founded in 2023 by Tsinghua University alumni and dubbed one of China's \"AI Tiger\" companies \u2014 K2.6 is a trillion-parameter Mixture-of-Experts model that has rapidly established itself as the most capable open-weight model available for coding and agentic tasks."
   ],
   "access_type_evidence": [
    "On Monday, the Sunnyvale-based chipmaker announced that it is now running [Kimi K2.6](https://www.kimi.com/ai-models/kimi-k2-6) \u2014 a trillion-parameter open-weight model developed by Beijing-based [Moonshot AI](https://www.moonshot.ai/) \u2014 for enterprise customers at nearly 1,000 tokens per second, a speed no GPU-based provider has come close to matching.",
    "Released on April 20 by Moonshot AI \u2014 a Beijing-based company founded in 2023 by Tsinghua University alumni and dubbed one of China's \"AI Tiger\" companies \u2014 K2.6 is a trillion-parameter Mixture-of-Experts model that has rapidly established itself as the most capable open-weight model available for coding and agentic tasks."
   ],
   "license_evidence": null,
   "significance_evidence": [
    "Released on April 20 by Moonshot AI \u2014 a Beijing-based company founded in 2023 by Tsinghua University alumni and dubbed one of China's \"AI Tiger\" companies \u2014 K2.6 is a trillion-parameter Mixture-of-Experts model that has rapidly established itself as the most capable open-weight model available for coding and agentic tasks.",
    "The model tops [SWE-Bench Pro](https://labs.scale.com/leaderboard/swe_bench_pro_public) at 58.6, outperforming Claude Opus 4.6 and matching GPT-5.4, while posting leading scores on agentic benchmarks like [Humanity's Last Exam](https://agi.safe.ai/) and [DeepSearchQA](https://huggingface.co/datasets/google/deepsearchqa)."
   ]
  },
  {
   "url": "https://micronomicon.com/ai-tooling/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is-2",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - Micronomicon",
   "published_at": "2026-07-15T12:06:53",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor",
   "organization_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "From April 24 to mid-June 2026, Chinese labs introduced four major open-weight models: **DeepSeek V4** on April 24, **MiniMax M3** on June 1, and **Kimi K2.7-Code** along with **GLM-5.2** in mid-June.",
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "From April 24 to mid-June 2026, Chinese labs introduced four major open-weight models: **DeepSeek V4** on April 24, **MiniMax M3** on June 1, and **Kimi K2.7-Code** along with **GLM-5.2** in mid-June.",
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor"
   ]
  },
  {
   "url": "https://micronomicon.com/ai-tooling/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is-2",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - Micronomicon",
   "published_at": "2026-07-15T12:06:53",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "cheap 1M-token context, native multimodal, modified-MIT",
   "organization_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "From April 24 to mid-June 2026, Chinese labs introduced four major open-weight models: **DeepSeek V4** on April 24, **MiniMax M3** on June 1, and **Kimi K2.7-Code** along with **GLM-5.2** in mid-June.",
    "**MiniMax M3**cheap 1M-token context, native multimodal, modified-MIT"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "From April 24 to mid-June 2026, Chinese labs introduced four major open-weight models: **DeepSeek V4** on April 24, **MiniMax M3** on June 1, and **Kimi K2.7-Code** along with **GLM-5.2** in mid-June.",
    "**MiniMax M3**cheap 1M-token context, native multimodal, modified-MIT"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "**MiniMax M3**cheap 1M-token context, native multimodal, modified-MIT"
   ]
  },
  {
   "url": "https://micronomicon.com/ai-tooling/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is-2",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - Micronomicon",
   "published_at": "2026-07-15T12:06:53",
   "organization": "Moonshot",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "agent-run specialist, ~30% fewer thinking tokens than K2.6",
   "organization_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "From April 24 to mid-June 2026, Chinese labs introduced four major open-weight models: **DeepSeek V4** on April 24, **MiniMax M3** on June 1, and **Kimi K2.7-Code** along with **GLM-5.2** in mid-June.",
    "**Kimi K2.7-Code (Moonshot)**agent-run specialist, ~30% fewer thinking tokens than K2.6"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "From April 24 to mid-June 2026, Chinese labs introduced four major open-weight models: **DeepSeek V4** on April 24, **MiniMax M3** on June 1, and **Kimi K2.7-Code** along with **GLM-5.2** in mid-June.",
    "**Kimi K2.7-Code (Moonshot)**agent-run specialist, ~30% fewer thinking tokens than K2.6"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "**Kimi K2.7-Code (Moonshot)**agent-run specialist, ~30% fewer thinking tokens than K2.6"
   ]
  },
  {
   "url": "https://micronomicon.com/ai-tooling/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is-2",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - Micronomicon",
   "published_at": "2026-07-15T12:06:53",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "753B MoE, MIT, top open-weight on Artificial Analysis index",
   "organization_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "From April 24 to mid-June 2026, Chinese labs introduced four major open-weight models: **DeepSeek V4** on April 24, **MiniMax M3** on June 1, and **Kimi K2.7-Code** along with **GLM-5.2** in mid-June.",
    "**GLM-5.2 (Z.ai)**753B MoE, MIT, top open-weight on Artificial Analysis index"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "From April 24 to mid-June 2026, Chinese labs introduced four major open-weight models: **DeepSeek V4** on April 24, **MiniMax M3** on June 1, and **Kimi K2.7-Code** along with **GLM-5.2** in mid-June.",
    "**GLM-5.2 (Z.ai)**753B MoE, MIT, top open-weight on Artificial Analysis index"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "**GLM-5.2 (Z.ai)**753B MoE, MIT, top open-weight on Artificial Analysis index"
   ]
  },
  {
   "url": "https://aifloxium.online/blog/best-open-source-ai-models",
   "title": "Best Open Source AI Models to Try in June 2026: Complete Guide",
   "published_at": "2026-06-03T22:58:10",
   "organization": "Xiaomi",
   "model": null,
   "release_date": "2026-04-27",
   "access_type": null,
   "license": null,
   "significance": "Trillion-parameter frontier model ranking #3 globally on agent benchmarks under an MIT license.",
   "organization_evidence": [
    "# Best Open Source AI Models to Try in June 2026: Complete Guide",
    "### MiMo-V2.5-Pro (Xiaomi) \u2014 The Undercover Agent Champion",
    "Xiaomi shocked the AI world by open-sourcing a trillion-parameter frontier model under the permissive **MIT license**."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Best Open Source AI Models to Try in June 2026: Complete Guide",
    "### MiMo-V2.5-Pro (Xiaomi) \u2014 The Undercover Agent Champion",
    "When Xiaomi revealed the identity on April 27, it immediately became the most discussed **open source LLM** on AI Twitter."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Best Open Source AI Models to Try in June 2026: Complete Guide",
    "### MiMo-V2.5-Pro (Xiaomi) \u2014 The Undercover Agent Champion",
    "Xiaomi shocked the AI world by open-sourcing a trillion-parameter frontier model under the permissive **MIT license**."
   ]
  },
  {
   "url": "https://dinimiciuillabs.com/blog/the-month-open-models-caught-up",
   "title": "The month open models caught up: Kimi K3, Inkling and GLM 5.2",
   "published_at": "2026-07-17T00:00:00",
   "organization": "Moonshot",
   "model": null,
   "release_date": "2026-07-01",
   "access_type": null,
   "license": null,
   "significance": "2.8 trillion parameter model, the largest open weight release ever, and it promptly topped the Frontend Code Arena ahead of the leading closed models",
   "organization_evidence": [
    "In July 2026, Moonshot released Kimi K3, a 2.8 trillion parameter model, the largest open weight release ever, and it promptly topped the Frontend Code Arena ahead of the leading closed models, per [Tom's Hardware's report](https://www.tomshardware.com/tech-industry/artificial-intelligence/moonshot-releases-2-8-trillion-parameter-kimi-k3).",
    "In short"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "In July 2026, Moonshot released Kimi K3, a 2.8 trillion parameter model, the largest open weight release ever, and it promptly topped the Frontend Code Arena ahead of the leading closed models, per [Tom's Hardware's report](https://www.tomshardware.com/tech-industry/artificial-intelligence/moonshot-releases-2-8-trillion-parameter-kimi-k3).",
    "In short"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "In July 2026, Moonshot released Kimi K3, a 2.8 trillion parameter model, the largest open weight release ever, and it promptly topped the Frontend Code Arena ahead of the leading closed models, per [Tom's Hardware's report](https://www.tomshardware.com/tech-industry/artificial-intelligence/moonshot-releases-2-8-trillion-parameter-kimi-k3).",
    "In short"
   ]
  },
  {
   "url": "https://dinimiciuillabs.com/blog/the-month-open-models-caught-up",
   "title": "The month open models caught up: Kimi K3, Inkling and GLM 5.2",
   "published_at": "2026-07-17T00:00:00",
   "organization": "Thinking Machines",
   "model": null,
   "release_date": "2026-07-01",
   "access_type": null,
   "license": null,
   "significance": "a mixture of experts design with 975 billion total and 41 billion active parameters, multimodal, trained on 45 trillion tokens",
   "organization_evidence": [
    "In short"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "In short"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "In short"
   ]
  },
  {
   "url": "https://ainchina.com/blog/china-ai-model-wars-summer-2026",
   "title": "China's AI Model Wars: The Summer That Changed Global AI",
   "published_at": "2026-08-09T00:00:00",
   "organization": "DeepSeek",
   "model": "V4",
   "release_date": "2026-04-24",
   "access_type": "open-weight MoE architecture",
   "license": "MIT",
   "significance": "1.6 trillion parameters of open-weight MoE architecture that undercut GPT-5 on cost by a factor of ten",
   "organization_evidence": [
    "China's AI Model Wars: The Summer That Changed Global AI",
    "It died on **April 24**, when DeepSeek released the V4 family\u20141.6 trillion parameters of open-weight MoE architecture that undercut GPT-5 on cost by a factor of ten."
   ],
   "model_evidence": [
    "China's AI Model Wars: The Summer That Changed Global AI",
    "It died on **April 24**, when DeepSeek released the V4 family\u20141.6 trillion parameters of open-weight MoE architecture that undercut GPT-5 on cost by a factor of ten."
   ],
   "release_date_evidence": [
    "China's AI Model Wars: The Summer That Changed Global AI",
    "It died on **April 24**, when DeepSeek released the V4 family\u20141.6 trillion parameters of open-weight MoE architecture that undercut GPT-5 on cost by a factor of ten."
   ],
   "access_type_evidence": [
    "China's AI Model Wars: The Summer That Changed Global AI",
    "It died on **April 24**, when DeepSeek released the V4 family\u20141.6 trillion parameters of open-weight MoE architecture that undercut GPT-5 on cost by a factor of ten."
   ],
   "license_evidence": [
    "China's AI Model Wars: The Summer That Changed Global AI",
    "| DeepSeek | V4-Pro | 1.6T (MoE) | April 24, 2026 | Yes (MIT) |"
   ],
   "significance_evidence": [
    "China's AI Model Wars: The Summer That Changed Global AI",
    "It died on **April 24**, when DeepSeek released the V4 family\u20141.6 trillion parameters of open-weight MoE architecture that undercut GPT-5 on cost by a factor of ten."
   ]
  },
  {
   "url": "https://ainchina.com/blog/china-ai-model-wars-summer-2026",
   "title": "China's AI Model Wars: The Summer That Changed Global AI",
   "published_at": "2026-08-09T00:00:00",
   "organization": "Moonshot AI",
   "model": "Kimi K3",
   "release_date": "2026-07-16",
   "access_type": "open-source",
   "license": "Modified MIT",
   "significance": "2.8 trillion parameters, becoming the largest open-source model ever built and ranking #4 globally on independent benchmarks",
   "organization_evidence": [
    "China's AI Model Wars: The Summer That Changed Global AI",
    "It died again on **July 16**, when Moonshot AI's Kimi K3 launched at 2.8 trillion parameters, becoming the largest open-source model ever built and ranking #4 globally on independent benchmarks."
   ],
   "model_evidence": [
    "China's AI Model Wars: The Summer That Changed Global AI",
    "It died again on **July 16**, when Moonshot AI's Kimi K3 launched at 2.8 trillion parameters, becoming the largest open-source model ever built and ranking #4 globally on independent benchmarks."
   ],
   "release_date_evidence": [
    "China's AI Model Wars: The Summer That Changed Global AI",
    "It died again on **July 16**, when Moonshot AI's Kimi K3 launched at 2.8 trillion parameters, becoming the largest open-source model ever built and ranking #4 globally on independent benchmarks."
   ],
   "access_type_evidence": [
    "China's AI Model Wars: The Summer That Changed Global AI",
    "It died again on **July 16**, when Moonshot AI's Kimi K3 launched at 2.8 trillion parameters, becoming the largest open-source model ever built and ranking #4 globally on independent benchmarks."
   ],
   "license_evidence": [
    "China's AI Model Wars: The Summer That Changed Global AI",
    "| Moonshot AI | Kimi K3 | 2.8T (MoE) | July 16, 2026 | Yes (Modified MIT) |"
   ],
   "significance_evidence": [
    "China's AI Model Wars: The Summer That Changed Global AI",
    "It died again on **July 16**, when Moonshot AI's Kimi K3 launched at 2.8 trillion parameters, becoming the largest open-source model ever built and ranking #4 globally on independent benchmarks."
   ]
  },
  {
   "url": "https://xenospectrum.com/en/thinking-machines-inkling-small-open-weights",
   "title": "Thinking Machines Officially Releases \"Inkling-Small,\" Maintaining Reasoning and Coding Performance at About 30% the Size of the Large Version",
   "published_at": "2026-08-01T21:45:05",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-30",
   "access_type": null,
   "license": null,
   "significance": "Shrunk to under 30% of the large version's size while maintaining reasoning and coding performance, with 276 billion total parameters.",
   "organization_evidence": [
    "# Thinking Machines Officially Releases \"Inkling-Small,\" Maintaining Reasoning and Coding Performance at About 30% the Size of the Large Version",
    "Thinking Machines Lab officially released its multimodal AI \"Inkling-Small\" as an open-weight model on July 30, 2026."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Thinking Machines Officially Releases \"Inkling-Small,\" Maintaining Reasoning and Coding Performance at About 30% the Size of the Large Version",
    "Thinking Machines Lab officially released its multimodal AI \"Inkling-Small\" as an open-weight model on July 30, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Thinking Machines Officially Releases \"Inkling-Small,\" Maintaining Reasoning and Coding Performance at About 30% the Size of the Large Version",
    "Thinking Machines Lab officially released its multimodal AI \"Inkling-Small\" as an open-weight model on July 30, 2026.",
    "Notably, the model\u2014shrunk to under 30% of the large version's size\u2014actually outperformed the large model on some reasoning and coding evaluations.",
    "Inkling-Small has a total of 276 billion parameters, with 12 billion active parameters used per token processed."
   ]
  },
  {
   "url": "https://nivaalabs.com/april-2026-the-biggest-month-ever-for-open-source-ai",
   "title": "April 2026: The Biggest Month Ever for Open-Source AI",
   "published_at": "2026-04-22T05:45:23",
   "organization": "Google",
   "model": null,
   "release_date": "2026-04-02",
   "access_type": null,
   "license": null,
   "significance": "31B beats Llama 4 on math & coding; offline phone/edge support",
   "organization_evidence": [
    "April 2026: The Biggest Month Ever for Open-Source AI",
    "### \ud83d\udd37 Gemma 4 \u2014 April 2 (Google, Apache 2.0)",
    "Google\u2019s most important open model release since the original Gemma."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "April 2026: The Biggest Month Ever for Open-Source AI",
    "### \ud83d\udd37 Gemma 4 \u2014 April 2 (Google, Apache 2.0)"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "April 2026: The Biggest Month Ever for Open-Source AI",
    "| Apr 2 | **Gemma 4** (E2B/E4B/26B/31B) | Google | Apache 2.0 \u2705 | 31B beats Llama 4 on math & coding; offline phone/edge support |",
    "Full Release Table: April 1\u201322, 2026"
   ]
  },
  {
   "url": "https://nivaalabs.com/april-2026-the-biggest-month-ever-for-open-source-ai",
   "title": "April 2026: The Biggest Month Ever for Open-Source AI",
   "published_at": "2026-04-22T05:45:23",
   "organization": "Ai2",
   "model": null,
   "release_date": "2026-04-03",
   "access_type": null,
   "license": null,
   "significance": "Only fully open training data, code & weights at this scale",
   "organization_evidence": [
    "April 2026: The Biggest Month Ever for Open-Source AI",
    "### \ud83d\udce6 OLMo 2 32B \u2014 April 3 (Ai2, Apache 2.0)",
    "The Allen Institute for AI shipped OLMo 2 32B \u2014 the only frontier-scale model in this entire wave where the training data, training code, and evaluation pipeline are completely open alongside the weights."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "April 2026: The Biggest Month Ever for Open-Source AI",
    "### \ud83d\udce6 OLMo 2 32B \u2014 April 3 (Ai2, Apache 2.0)"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "April 2026: The Biggest Month Ever for Open-Source AI",
    "| Apr 3 | **OLMo 2 32B** | Ai2 | Apache 2.0 \u2705 | Only fully open training data, code & weights at this scale |",
    "Full Release Table: April 1\u201322, 2026"
   ]
  },
  {
   "url": "https://nivaalabs.com/april-2026-the-biggest-month-ever-for-open-source-ai",
   "title": "April 2026: The Biggest Month Ever for Open-Source AI",
   "published_at": "2026-04-22T05:45:23",
   "organization": "Meta",
   "model": null,
   "release_date": "2026-04-05",
   "access_type": null,
   "license": null,
   "significance": "First native multimodal MoE; 10M context (Scout); consumer GPU viable",
   "organization_evidence": [
    "April 2026: The Biggest Month Ever for Open-Source AI",
    "### \ud83e\udd99 Llama 4 Scout + Maverick \u2014 April 5 (Meta, Custom License)",
    "Meta\u2019s first MoE model family and first natively multimodal release."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "April 2026: The Biggest Month Ever for Open-Source AI",
    "### \ud83e\udd99 Llama 4 Scout + Maverick \u2014 April 5 (Meta, Custom License)"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "April 2026: The Biggest Month Ever for Open-Source AI",
    "| Apr 5 | **Llama 4 Scout + Maverick** | Meta | Custom \u26a0\ufe0f | First native multimodal MoE; 10M context (Scout); consumer GPU viable |",
    "Full Release Table: April 1\u201322, 2026"
   ]
  },
  {
   "url": "https://nivaalabs.com/april-2026-the-biggest-month-ever-for-open-source-ai",
   "title": "April 2026: The Biggest Month Ever for Open-Source AI",
   "published_at": "2026-04-22T05:45:23",
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-04-05",
   "access_type": null,
   "license": null,
   "significance": "Top dense model on reasoning; thinking + fast modes",
   "organization_evidence": [
    "April 2026: The Biggest Month Ever for Open-Source AI",
    "### \ud83e\udd16 Qwen 3 72B \u2014 April 5 (Alibaba, Apache 2.0)",
    "Alibaba\u2019s top dense model in its Q2 open-source push \u2014 72B parameters, Apache 2.0, with reasoning and coding performance that exceeded GPT-4o on MMLU-Pro."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "April 2026: The Biggest Month Ever for Open-Source AI",
    "### \ud83e\udd16 Qwen 3 72B \u2014 April 5 (Alibaba, Apache 2.0)"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "April 2026: The Biggest Month Ever for Open-Source AI",
    "| Apr 5 | **Qwen 3 72B** | Alibaba | Apache 2.0 \u2705 | Top dense model on reasoning; thinking + fast modes |",
    "Full Release Table: April 1\u201322, 2026"
   ]
  },
  {
   "url": "https://nivaalabs.com/april-2026-the-biggest-month-ever-for-open-source-ai",
   "title": "April 2026: The Biggest Month Ever for Open-Source AI",
   "published_at": "2026-04-22T05:45:23",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-04-07",
   "access_type": null,
   "license": null,
   "significance": "#1 SWE-Bench Pro (58.4); no Nvidia chips; $1/$3.20 per 1M tokens",
   "organization_evidence": [
    "April 2026: The Biggest Month Ever for Open-Source AI",
    "### \ud83d\udc09 GLM-5.1 (Z.ai, MIT) \u2014 April 7",
    "Z.ai dropped open weights for GLM-5.1 \u2014 754B MoE (40B active), MIT license, trained entirely on Huawei Ascend 910B chips \u2014 with a 58.4 SWE-Bench Pro score that edged past GPT-5.4 (57.7) and Claude Opus 4.6 (57.3)."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "April 2026: The Biggest Month Ever for Open-Source AI",
    "### \ud83d\udc09 GLM-5.1 (Z.ai, MIT) \u2014 April 7"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "April 2026: The Biggest Month Ever for Open-Source AI",
    "| Apr 7 | **GLM-5.1** | Z.ai | MIT \u2705 | #1 SWE-Bench Pro (58.4); no Nvidia chips; $1/$3.20 per 1M tokens |",
    "Full Release Table: April 1\u201322, 2026"
   ]
  },
  {
   "url": "https://nivaalabs.com/april-2026-the-biggest-month-ever-for-open-source-ai",
   "title": "April 2026: The Biggest Month Ever for Open-Source AI",
   "published_at": "2026-04-22T05:45:23",
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-04-08",
   "access_type": null,
   "license": null,
   "significance": "235B/22A MoE; near-frontier at low active param cost",
   "organization_evidence": [
    "April 2026: The Biggest Month Ever for Open-Source AI",
    "### \ud83e\udd16 Qwen 3 MoE 235B \u2014 April 8 (Alibaba, Apache 2.0)",
    "Alibaba followed up the 72B Dense with the heavier MoE variant \u2014 235B total, 22B active, Apache 2.0."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "April 2026: The Biggest Month Ever for Open-Source AI",
    "### \ud83e\udd16 Qwen 3 MoE 235B \u2014 April 8 (Alibaba, Apache 2.0)"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "April 2026: The Biggest Month Ever for Open-Source AI",
    "| Apr 8 | **Qwen 3 MoE 235B** | Alibaba | Apache 2.0 \u2705 | 235B/22A MoE; near-frontier at low active param cost |",
    "Full Release Table: April 1\u201322, 2026"
   ]
  },
  {
   "url": "https://nivaalabs.com/april-2026-the-biggest-month-ever-for-open-source-ai",
   "title": "April 2026: The Biggest Month Ever for Open-Source AI",
   "published_at": "2026-04-22T05:45:23",
   "organization": "Google",
   "model": null,
   "release_date": "2026-04-09",
   "access_type": null,
   "license": null,
   "significance": "4B effective / 2B footprint; fully offline on phones; 128K context",
   "organization_evidence": [
    "April 2026: The Biggest Month Ever for Open-Source AI",
    "### \ud83d\udcf1 Gemma 3n \u2014 April 9 (Google, Gemma License)",
    "Google\u2019s next-generation on-device model \u2014 4B effective parameters, 2B memory footprint \u2014 built specifically for phones and tablets."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "April 2026: The Biggest Month Ever for Open-Source AI",
    "### \ud83d\udcf1 Gemma 3n \u2014 April 9 (Google, Gemma License)"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "April 2026: The Biggest Month Ever for Open-Source AI",
    "| Apr 9 | **Gemma 3n** | Google | Gemma \u26a0\ufe0f | 4B effective / 2B footprint; fully offline on phones; 128K context |",
    "Full Release Table: April 1\u201322, 2026"
   ]
  },
  {
   "url": "https://nivaalabs.com/april-2026-the-biggest-month-ever-for-open-source-ai",
   "title": "April 2026: The Biggest Month Ever for Open-Source AI",
   "published_at": "2026-04-22T05:45:23",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-04-11",
   "access_type": null,
   "license": null,
   "significance": "Self-evolving; 230B/10B active; #1 GDPval-AA open-weight; $0.30/$1.20",
   "organization_evidence": [
    "April 2026: The Biggest Month Ever for Open-Source AI",
    "### \ud83e\udd16 MiniMax M2.7 \u2014 April 11 (Research License \u26a0\ufe0f)",
    "MiniMax\u2019s self-evolving agentic model \u2014 230B total, 10B active, the smallest active-parameter footprint in the Tier-1 coding class."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "April 2026: The Biggest Month Ever for Open-Source AI",
    "### \ud83e\udd16 MiniMax M2.7 \u2014 April 11 (Research License \u26a0\ufe0f)"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "April 2026: The Biggest Month Ever for Open-Source AI",
    "| Apr 11 | **MiniMax M2.7** | MiniMax | Research \u26a0\ufe0f | Self-evolving; 230B/10B active; #1 GDPval-AA open-weight; $0.30/$1.20 |",
    "Full Release Table: April 1\u201322, 2026"
   ]
  },
  {
   "url": "https://nivaalabs.com/april-2026-the-biggest-month-ever-for-open-source-ai",
   "title": "April 2026: The Biggest Month Ever for Open-Source AI",
   "published_at": "2026-04-22T05:45:23",
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-04-16",
   "access_type": null,
   "license": null,
   "significance": "73.4% SWE-Bench Verified; 3B active / 35B total; runs on RTX 4090",
   "organization_evidence": [
    "April 2026: The Biggest Month Ever for Open-Source AI",
    "### \ud83e\udde0 Qwen 3.6-35B-A3B \u2014 April 16 (Alibaba, Apache 2.0)",
    "Alibaba\u2019s Qwen team shipped a 35B parameter MoE model that activates only 3B parameters per query \u2014 a 90% compute reduction versus models it outperforms."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "April 2026: The Biggest Month Ever for Open-Source AI",
    "### \ud83e\udde0 Qwen 3.6-35B-A3B \u2014 April 16 (Alibaba, Apache 2.0)"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "April 2026: The Biggest Month Ever for Open-Source AI",
    "| Apr 16 | **Qwen 3.6-35B-A3B** | Alibaba | Apache 2.0 \u2705 | 73.4% SWE-Bench Verified; 3B active / 35B total; runs on RTX 4090 |",
    "Full Release Table: April 1\u201322, 2026"
   ]
  },
  {
   "url": "https://nivaalabs.com/april-2026-the-biggest-month-ever-for-open-source-ai",
   "title": "April 2026: The Biggest Month Ever for Open-Source AI",
   "published_at": "2026-04-22T05:45:23",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-04-20",
   "access_type": null,
   "license": null,
   "significance": "New #1 SWE-Bench Pro (58.6); 1T/32B MoE; 300-agent swarms; 12-hr runs",
   "organization_evidence": [
    "April 2026: The Biggest Month Ever for Open-Source AI",
    "### \ud83c\udf19 Kimi K2.6 GA \u2014 April 20 (Moonshot AI, Modified MIT)",
    "Eight days after the preview, Moonshot AI shipped Kimi K2.6 as generally available \u2014 and immediately claimed the new SWE-Bench Pro #1 with 58.6 (beating GLM-5.1\u2019s 58.4 that had held the top spot since April 7)."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "April 2026: The Biggest Month Ever for Open-Source AI",
    "### \ud83c\udf19 Kimi K2.6 GA \u2014 April 20 (Moonshot AI, Modified MIT)"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "April 2026: The Biggest Month Ever for Open-Source AI",
    "| Apr 20 | **Kimi K2.6 GA** | Moonshot AI | Modified MIT \u26a0\ufe0f | New #1 SWE-Bench Pro (58.6); 1T/32B MoE; 300-agent swarms; 12-hr runs |",
    "Full Release Table: April 1\u201322, 2026"
   ]
  },
  {
   "url": "https://vinoth12940.github.io/blog/articles/genai-20260718-kimi-k3-launch",
   "title": "Kimi K3: The 2.8 Trillion Parameter Open Model That Just Reset the Frontier",
   "published_at": "2026-07-18T00:00:00",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-16",
   "access_type": null,
   "license": null,
   "significance": "a 2.8 trillion parameter open-weight model with a 1 million token context window",
   "organization_evidence": [
    "July 18, 2026 \u00b7 AI Models Open Source Moonshot AI",
    "Let's cut through the marketing. Moonshot AI (the Beijing-based company behind the Kimi chatbot) published K3 on July 14, 2026, with general availability following on July 16."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Let's cut through the marketing. Moonshot AI (the Beijing-based company behind the Kimi chatbot) published K3 on July 14, 2026, with general availability following on July 16."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "July 18, 2026 \u00b7 AI Models Open Source Moonshot AI",
    "Two days ago, Moonshot AI launched Kimi K3 \u2014 a 2.8 trillion parameter open-weight model with a 1 million token context window."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/glm-5-1-non-reasoning-vs-qwen3-6-plus",
   "title": "GLM-5.1 (Non-reasoning) vs Qwen3.6 Plus: Model Comparison",
   "published_at": null,
   "organization": "Z AI",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "200k tokens context window with open source weights",
   "organization_evidence": [
    "# GLM-5.1 (Non-reasoning) vs. Qwen3.6 Plus",
    "## Model Comparison",
    "| Metric | ![Z AI logo](/img/logos/zai_small.svg)GLM-5.1 (Non-reasoning) | ![Alibaba logo](/img/logos/alibaba_small.svg)Qwen3.6 Plus | Analysis |",
    "| Creator | ![Z AI](/img/logos/zai_small.svg)[](https://z.ai/) | ![Alibaba](/img/logos/alibaba_small.svg)[](https://www.alibabacloud.com/) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# GLM-5.1 (Non-reasoning) vs. Qwen3.6 Plus",
    "## Model Comparison",
    "| Metric | ![Z AI logo](/img/logos/zai_small.svg)GLM-5.1 (Non-reasoning) | ![Alibaba logo](/img/logos/alibaba_small.svg)Qwen3.6 Plus | Analysis |",
    "| Release Date | April, 2026 | April, 2026 | GLM-5.1 (Non-reasoning) has a more recent release date than Qwen3.6 Plus |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# GLM-5.1 (Non-reasoning) vs. Qwen3.6 Plus",
    "## Model Comparison",
    "| Metric | ![Z AI logo](/img/logos/zai_small.svg)GLM-5.1 (Non-reasoning) | ![Alibaba logo](/img/logos/alibaba_small.svg)Qwen3.6 Plus | Analysis |",
    "| Context Window | 200k tokens (~300 A4 pages of size 12 Arial font) | 1000k tokens (~1500 A4 pages of size 12 Arial font) | GLM-5.1 (Non-reasoning) is smaller than Qwen3.6 Plus |",
    "| Open Source (Weights) | Yes[](https://huggingface.co/zai-org/GLM-5.1) | No | GLM-5.1 (Non-reasoning) is open source while Qwen3.6 Plus is proprietary |"
   ]
  },
  {
   "url": "https://agenticbrew.ai/news/05c05277-f513-4465-945e-10131501086d/thinking-machines-lab-launches-inkling-open-weights-model",
   "title": "Thinking Machines Lab launches Inkling open-weights model",
   "published_at": "2026-07-16T00:27:26",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "a 975B-parameter Mixture-of-Experts transformer with 41B active parameters, pretrained from scratch on 45 trillion tokens of text, image, audio, and video data with a 1M-token context window",
   "organization_evidence": [
    "Thinking Machines Lab released Inkling on July 15, 2026, a 975B-parameter Mixture-of-Experts transformer with 41B active parameters, pretrained from scratch on 45 trillion tokens of text, image, audio, and video data with a 1M-token context window."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Thinking Machines Lab released Inkling on July 15, 2026, a 975B-parameter Mixture-of-Experts transformer with 41B active parameters, pretrained from scratch on 45 trillion tokens of text, image, audio, and video data with a 1M-token context window."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Thinking Machines Lab released Inkling on July 15, 2026, a 975B-parameter Mixture-of-Experts transformer with 41B active parameters, pretrained from scratch on 45 trillion tokens of text, image, audio, and video data with a 1M-token context window."
   ]
  },
  {
   "url": "https://medium.com/@pioneer-fastino/closing-the-gap-open-weight-vs-proprietary-frontier-language-models-cb50ad1662fc",
   "title": "Closing the Gap: Open-Weight vs. Proprietary Frontier Language Models",
   "published_at": "2026-07-20T00:00:12",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-06-16",
   "access_type": null,
   "license": null,
   "significance": "A 744-billion-parameter mixture-of-experts model with about 40 billion active parameters per forward pass, a 1-million-token context window, and up to 128K output tokens.",
   "organization_evidence": [
    "[Z.ai](http://z.ai/)\u2019s GLM-5.2 follows with a score of 51.",
    "[Z.ai](http://z.ai/) shipped GLM-5.2 in mid-June under an MIT license, API live on June 16."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "[Z.ai](http://z.ai/) shipped GLM-5.2 in mid-June under an MIT license, API live on June 16."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "**GLM-5.2** is a 744-billion-parameter mixture-of-experts (MoE) model with about 40 billion active parameters per forward pass, a 1-million-token context window, and up to 128K output tokens.",
    "What they shipped"
   ]
  },
  {
   "url": "https://medium.com/@pioneer-fastino/closing-the-gap-open-weight-vs-proprietary-frontier-language-models-cb50ad1662fc",
   "title": "Closing the Gap: Open-Weight vs. Proprietary Frontier Language Models",
   "published_at": "2026-07-20T00:00:12",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-27",
   "access_type": null,
   "license": null,
   "significance": "A 2.8-trillion-parameter MoE model, the first open model to reach the 3-trillion class.",
   "organization_evidence": [
    "Moonshot AI\u2019s Kimi K3 scores 57 on the Artificial Analysis Intelligence Index, a composite of nine evaluations.",
    "A month later, on July 16, Moonshot AI announced Kimi K3, with API access at launch and full weights arriving on Hugging Face by July 27."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "A month later, on July 16, Moonshot AI announced Kimi K3, with API access at launch and full weights arriving on Hugging Face by July 27."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "**Kimi K3** is a 2.8-trillion-parameter MoE model, the first open model to reach the 3-trillion class.",
    "What they shipped"
   ]
  },
  {
   "url": "http://artificialanalysis.ai/models/muse-glimmer",
   "title": "Muse Glimmer (high) - Intelligence, Performance & Price Analysis | Artificial Analysis",
   "published_at": null,
   "organization": "Meta",
   "model": null,
   "release_date": "2026-08-10",
   "access_type": null,
   "license": null,
   "significance": "A 30B parameter reasoning and multimodal model with a 131k context window and leading intelligence among open weight models of similar size.",
   "organization_evidence": [
    "Muse Glimmer (high) Intelligence, Performance & Price Analysis",
    "Muse Glimmer (high) was created by Meta."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Muse Glimmer (high) Intelligence, Performance & Price Analysis",
    "Muse Glimmer (high) was released on August 10, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Muse Glimmer (high) Intelligence, Performance & Price Analysis",
    "Muse Glimmer (high) is amongst the leading models in intelligence, but somewhat expensive when comparing to other open weight models of similar size.",
    "Total parameters | 30B |",
    "Yes, Muse Glimmer (high) is a reasoning model."
   ]
  },
  {
   "url": "https://arxiv.org/html/2604.07190v2",
   "title": "The ATOM Report: Measuring the Open Language Model Ecosystem",
   "published_at": null,
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-02",
   "access_type": null,
   "license": null,
   "significance": "strong initial adoption and high launch momentum across multiple size classes",
   "organization_evidence": [
    "The ATOM Report:\nMeasuring the Open Language Model Ecosystem",
    "We present a comprehensive adoption snapshot of the leading open language models and who is building them, focusing on the 1.5K mainline open models from the likes of Alibaba\u2019s Qwen, DeepSeek, Meta\u2019s Llama, that are the foundation of an ecosystem crucial to researchers, entrepreneurs, and policy advisors.",
    "The early adoption numbers of Qwen3.5 (Qwen Team, 2026) since its release in February 2026 are an indication that the dominance of Qwen relative to its peers will continue (see Section 6 and specifically Fig. 15 for early data on Qwen 3.5\u2019s adoption)."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The ATOM Report:\nMeasuring the Open Language Model Ecosystem",
    "The early adoption numbers of Qwen3.5 (Qwen Team, 2026) since its release in February 2026 are an indication that the dominance of Qwen relative to its peers will continue (see Section 6 and specifically Fig. 15 for early data on Qwen 3.5\u2019s adoption)."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The ATOM Report:\nMeasuring the Open Language Model Ecosystem",
    "For example, using the 2026-Q2 RAM baseline, the RAM framework makes the February 2026 Qwen3.5 rollout easier to compare across scales."
   ]
  },
  {
   "url": "https://codersera.com/blog/kimi-k3-complete-guide-2026/amp",
   "title": "Kimi K3: Specs, Pricing & Release (2026)",
   "published_at": "2026-07-17T18:05:59",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-16",
   "access_type": null,
   "license": null,
   "significance": "A 2.8-trillion-parameter open-weight, multimodal reasoning model with a 1-million-token context window",
   "organization_evidence": [
    "# Kimi K3: Moonshot AI\u2019s 2.8T Open-Weight Model \u2014 Release, Specs & Pricing (2026)",
    "Moonshot AI shipped **Kimi K3** on July 16, 2026, and it immediately reset expectations for what an open-weight model can do."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Kimi K3: Moonshot AI\u2019s 2.8T Open-Weight Model \u2014 Release, Specs & Pricing (2026)",
    "Moonshot AI shipped **Kimi K3** on July 16, 2026, and it immediately reset expectations for what an open-weight model can do."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Kimi K3: Moonshot AI\u2019s 2.8T Open-Weight Model \u2014 Release, Specs & Pricing (2026)",
    "It's a 2.8-trillion-parameter open-weight, multimodal reasoning model \u2014 the largest open-weight model shipped to date \u2014 with a 1-million-token context window, an always-on \"thinking mode,\" and pricing of $3 per million input tokens and $15 per million output tokens."
   ]
  },
  {
   "url": "https://deploybase.ai/articles/open-source-llm-release-news",
   "title": "Open-Source LLM Release News: March 2026 Updates",
   "published_at": "2026-03-10T06:28:31",
   "organization": "Meta",
   "model": null,
   "release_date": "2026-03-15",
   "access_type": null,
   "license": null,
   "significance": "improved reasoning and mixture of experts architecture with multimodal understanding",
   "organization_evidence": [
    "Open Source LLM Release News: Highlights",
    "March 2026 marks a release surge in open-source LLMs.",
    "Meta released Llama 4 with improved reasoning."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Llama 4 Announcement",
    "Meta released Llama 4 on March 15, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Llama 4 Announcement",
    "Meta released Llama 4 on March 15, 2026.",
    "Named it Llama 4 to signal major architectural changes.",
    "Both Scout and Maverick include native multimodal understanding (vision + text)."
   ]
  },
  {
   "url": "https://deploybase.ai/articles/open-source-llm-release-news",
   "title": "Open-Source LLM Release News: March 2026 Updates",
   "published_at": "2026-03-10T06:28:31",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-03-18",
   "access_type": null,
   "license": null,
   "significance": "long-context tasks and efficiency improvements with 256K token context window",
   "organization_evidence": [
    "Open Source LLM Release News: Highlights",
    "March 2026 marks a release surge in open-source LLMs.",
    "DeepSeek published V3.1 with long-context support."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "DeepSeek V3.1 Release",
    "DeepSeek published V3.1 on March 18, 2026, nine weeks after V3.0 launch."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "DeepSeek V3.1 Release",
    "DeepSeek published V3.1 on March 18, 2026, nine weeks after V3.0 launch.",
    "Focused on long-context tasks and efficiency improvements."
   ]
  },
  {
   "url": "https://deploybase.ai/articles/open-source-llm-release-news",
   "title": "Open-Source LLM Release News: March 2026 Updates",
   "published_at": "2026-03-10T06:28:31",
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-03-12",
   "access_type": null,
   "license": null,
   "significance": "multilingual support and efficiency across 30 languages",
   "organization_evidence": [
    "Open Source LLM Release News: Highlights",
    "March 2026 marks a release surge in open-source LLMs.",
    "Alibaba shipped Qwen 2.5 with multilingual improvements."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Qwen 2.5 Updates",
    "Alibaba released Qwen 2.5 on March 12, 2026, iterating on Qwen 2 (March 2024) and Qwen 2.5-turbo (August 2025)."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Qwen 2.5 Updates",
    "Alibaba released Qwen 2.5 on March 12, 2026, iterating on Qwen 2 (March 2024) and Qwen 2.5-turbo (August 2025).",
    "Qwen 2.5 focuses on multilingual support and efficiency."
   ]
  },
  {
   "url": "https://deploybase.ai/articles/open-source-llm-release-news",
   "title": "Open-Source LLM Release News: March 2026 Updates",
   "published_at": "2026-03-10T06:28:31",
   "organization": "Google",
   "model": null,
   "release_date": "2026-03-10",
   "access_type": null,
   "license": null,
   "significance": "performance improvements on math and code",
   "organization_evidence": [
    "Open Source LLM Release News: Highlights",
    "March 2026 marks a release surge in open-source LLMs.",
    "Google incremented Gemma with performance gains."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Gemma Model Updates",
    "Google incremented Gemma with two releases in March 2026.",
    "**Gemma 2.5:** Released March 10."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Gemma Model Updates",
    "Google incremented Gemma with two releases in March 2026.",
    "**Gemma 2.5:** Released March 10.",
    "Gemma 2.5 is a minor update with performance improvements on math and code."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/muse-glimmer?intelligence-index-token-use=intelligence-vs-output-tokens-per-task&model-filters=open-source&models=muse-glimmer,qwen3-6-27b,gemma-4-31b",
   "title": "Muse Glimmer (high) - Intelligence, Performance & Price Analysis",
   "published_at": null,
   "organization": "Meta",
   "model": null,
   "release_date": "2026-08-10",
   "access_type": null,
   "license": null,
   "significance": "a 30B parameter reasoning model supporting text and image input with a 131k context window",
   "organization_evidence": [
    "Muse Glimmer (high) was created by Meta.",
    "# Muse Glimmer (high) Intelligence, Performance & Price Analysis"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Muse Glimmer (high) was released on August 10, 2026.",
    "# Muse Glimmer (high) Intelligence, Performance & Price Analysis"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Muse Glimmer (high) is amongst the leading models in intelligence, but particularly expensive when comparing to other open weight models of similar size.",
    "# Muse Glimmer (high) Intelligence, Performance & Price Analysis"
   ]
  },
  {
   "url": "https://techtrendtrove.com/science-technology/four-open-ai-models-in-record-time-china-s-rapid-signal-deployment",
   "title": "Four Open AI Models In Record Time: China\u2019s Rapid Signal Deployment - Tech Trend Trove",
   "published_at": "2026-07-15T06:14:44",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "1.6 trillion total parameters with 49 billion active, 1 million token context window",
   "organization_evidence": [
    "AI DISPATCH \u00b7 SIGNAL",
    "## Four Frontier-Class Open Models in Eight Weeks",
    "**DeepSeek** emphasizes affordability and high parameter counts with efficient activation, while **Z.ai** leads in open-weight intelligence."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "AI DISPATCH \u00b7 SIGNAL",
    "## Four Frontier-Class Open Models in Eight Weeks",
    "APR 24",
    "Between April 24 and June 15, 2026, Chinese labs launched four major open-weight models: **DeepSeek V4**, **MiniMax M3**, **Kimi K2.7-Code**, and **GLM-5.2**."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "AI DISPATCH \u00b7 SIGNAL",
    "## Four Frontier-Class Open Models in Eight Weeks",
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor",
    "**DeepSeek V4** leads in capability, with 1.6 trillion parameters but activating only 49 billion per pass, and a 1 million token context window, making it competitive with proprietary models."
   ]
  },
  {
   "url": "https://techtrendtrove.com/science-technology/four-open-ai-models-in-record-time-china-s-rapid-signal-deployment",
   "title": "Four Open AI Models In Record Time: China\u2019s Rapid Signal Deployment - Tech Trend Trove",
   "published_at": "2026-07-15T06:14:44",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "cheap 1 million token context, native multimodal",
   "organization_evidence": [
    "AI DISPATCH \u00b7 SIGNAL",
    "## Four Frontier-Class Open Models in Eight Weeks",
    "Between April 24 and June 15, 2026, Chinese labs launched four major open-weight models: **DeepSeek V4**, **MiniMax M3**, **Kimi K2.7-Code**, and **GLM-5.2**."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "AI DISPATCH \u00b7 SIGNAL",
    "## Four Frontier-Class Open Models in Eight Weeks",
    "JUN 01",
    "**MiniMax M3**cheap 1M-token context, native multimodal, modified-MIT"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "AI DISPATCH \u00b7 SIGNAL",
    "## Four Frontier-Class Open Models in Eight Weeks",
    "**MiniMax M3**cheap 1M-token context, native multimodal, modified-MIT"
   ]
  },
  {
   "url": "https://techtrendtrove.com/science-technology/four-open-ai-models-in-record-time-china-s-rapid-signal-deployment",
   "title": "Four Open AI Models In Record Time: China\u2019s Rapid Signal Deployment - Tech Trend Trove",
   "published_at": "2026-07-15T06:14:44",
   "organization": "Moonshot",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "agent-run specialist with about 30% fewer thinking tokens than K2.6",
   "organization_evidence": [
    "AI DISPATCH \u00b7 SIGNAL",
    "## Four Frontier-Class Open Models in Eight Weeks",
    "**Moonshot\u2019s Kimi** focuses on long-term agent stability, and **Alibaba\u2019s Qwen** offers broad, self-hostable variants."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "AI DISPATCH \u00b7 SIGNAL",
    "## Four Frontier-Class Open Models in Eight Weeks",
    "JUN 13",
    "**Kimi K2.7-Code (Moonshot)**agent-run specialist, ~30% fewer thinking tokens than K2.6"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "AI DISPATCH \u00b7 SIGNAL",
    "## Four Frontier-Class Open Models in Eight Weeks",
    "**Kimi K2.7-Code (Moonshot)**agent-run specialist, ~30% fewer thinking tokens than K2.6"
   ]
  },
  {
   "url": "https://techtrendtrove.com/science-technology/four-open-ai-models-in-record-time-china-s-rapid-signal-deployment",
   "title": "Four Open AI Models In Record Time: China\u2019s Rapid Signal Deployment - Tech Trend Trove",
   "published_at": "2026-07-15T06:14:44",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "753B Mixture of Experts model, top open-weight on Artificial Analysis index",
   "organization_evidence": [
    "AI DISPATCH \u00b7 SIGNAL",
    "## Four Frontier-Class Open Models in Eight Weeks",
    "**GLM-5.2 (Z.ai)**753B MoE, MIT, top open-weight on Artificial Analysis index"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "AI DISPATCH \u00b7 SIGNAL",
    "## Four Frontier-Class Open Models in Eight Weeks",
    "JUN 13\u201316",
    "**GLM-5.2 (Z.ai)**753B MoE, MIT, top open-weight on Artificial Analysis index"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "AI DISPATCH \u00b7 SIGNAL",
    "## Four Frontier-Class Open Models in Eight Weeks",
    "**GLM-5.2 (Z.ai)**753B MoE, MIT, top open-weight on Artificial Analysis index"
   ]
  },
  {
   "url": "https://groundtruth.day/news/inkling-thinking-machines-open-weights-975b.html",
   "title": "Thinking Machines releases Inkling, now the top-ranked US open-weights model",
   "published_at": "2026-07-15T22:51:59",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "A 975-billion-parameter mixture-of-experts open-weights model that is the top-ranked US open-weights model on the Artificial Analysis Intelligence Index.",
   "organization_evidence": [
    "Thinking Machines releases Inkling, now the top-ranked US open-weights model",
    "Thinking Machines Lab has released Inkling, a 975-billion-parameter open-weights model published under the permissive Apache 2.0 license."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Thinking Machines releases Inkling, now the top-ranked US open-weights model",
    "News \u00b7 [2026-07-15](2026-07-15.html)",
    "- **When:** Released July 15, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Thinking Machines releases Inkling, now the top-ranked US open-weights model",
    "Thinking Machines Lab has released Inkling, a 975-billion-parameter open-weights model published under the permissive Apache 2.0 license."
   ]
  },
  {
   "url": "https://klymentiev.com/blog/best-open-source-llm-2026",
   "title": "Best Open Source LLM 2026: Llama 4 vs Qwen 3.5 vs DeepSeek V4 vs Kimi K2.5 vs GLM-5 vs Mistral",
   "published_at": "2026-05-10T21:09:59",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "overall open-weight leaderboard leader scoring 87 on BenchLM and 89.8 on coding with a 1M token context",
   "organization_evidence": [
    "# Best Open Source LLM 2026: Llama 4 vs Qwen 3.5 vs DeepSeek V4",
    "DeepSeek V4 Pro \u2014 overall leader",
    "DeepSeek V4 (released **April 24, 2026**) ships in Pro (Max + High) and Flash variants."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Best Open Source LLM 2026: Llama 4 vs Qwen 3.5 vs DeepSeek V4",
    "DeepSeek V4 Pro \u2014 overall leader",
    "DeepSeek V4 (released **April 24, 2026**) ships in Pro (Max + High) and Flash variants."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Best Open Source LLM 2026: Llama 4 vs Qwen 3.5 vs DeepSeek V4",
    "DeepSeek V4 Pro \u2014 overall leader",
    "The best open-source LLM in May 2026 depends on workload: **DeepSeek V4 Pro (Max)** leads the overall open-weight leaderboard at BenchLM (score 87), **Qwen 3.5 397B** is the strongest Apache 2.0 all-rounder, **Kimi K2.5** leads HumanEval (99%), **GLM-5 Reasoning** leads knowledge benchmarks (MMLU 96), and **Llama 4 Scout** owns the long-context segment with a 10M token window."
   ]
  },
  {
   "url": "https://venturebeat.com/ai/western-qwen-ibm-wows-with-granite-4-llm-launch-and-hybrid-mamba-transformer",
   "title": "IBM wows with Granite 4 LLM launch and hybrid Mamba/Transformer architecture",
   "published_at": "2025-12-22T12:14:21",
   "organization": "IBM",
   "model": "Granite 4.0",
   "release_date": "2026-08-31",
   "access_type": "open source",
   "license": "Apache 2.0",
   "significance": "Newest generation hybrid Mamba-transformer model family offering high performance with lower memory and cost requirements.",
   "organization_evidence": [
    "IBM today [announced the release of Granite 4.0](https://www.ibm.com/new/announcements/ibm-granite-4-0-hyper-efficient-high-performance-hybrid-models), the newest generation of its homemade family of open source large language models (LLMs) designed to balance high performance with lower memory and cost requirements."
   ],
   "model_evidence": [
    "IBM today [announced the release of Granite 4.0](https://www.ibm.com/new/announcements/ibm-granite-4-0-hyper-efficient-high-performance-hybrid-models), the newest generation of its homemade family of open source large language models (LLMs) designed to balance high performance with lower memory and cost requirements."
   ],
   "release_date_evidence": [
    "IBM today [announced the release of Granite 4.0](https://www.ibm.com/new/announcements/ibm-granite-4-0-hyper-efficient-high-performance-hybrid-models), the newest generation of its homemade family of open source large language models (LLMs) designed to balance high performance with lower memory and cost requirements."
   ],
   "access_type_evidence": [
    "IBM today [announced the release of Granite 4.0](https://www.ibm.com/new/announcements/ibm-granite-4-0-hyper-efficient-high-performance-hybrid-models), the newest generation of its homemade family of open source large language models (LLMs) designed to balance high performance with lower memory and cost requirements."
   ],
   "license_evidence": [
    "The models are open sourced under the Apache 2.0 license, cryptographically signed for authenticity, and stand out as the first open language model family certified under ISO 42001, an international standard for AI governance and transparency."
   ],
   "significance_evidence": [
    "IBM today [announced the release of Granite 4.0](https://www.ibm.com/new/announcements/ibm-granite-4-0-hyper-efficient-high-performance-hybrid-models), the newest generation of its homemade family of open source large language models (LLMs) designed to balance high performance with lower memory and cost requirements."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/longcat-2-0",
   "title": "LongCat 2.0 - Intelligence, Performance & Price Analysis | Artificial Analysis",
   "published_at": null,
   "organization": "LongCat",
   "model": null,
   "release_date": "2026-06-29",
   "access_type": null,
   "license": null,
   "significance": "1.6 trillion parameter Mixture of Experts model with 1 million token context window",
   "organization_evidence": [
    "LongCat 2.0 Intelligence, Performance & Price Analysis",
    "LongCat 2.0 was created by LongCat."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Released June 2026",
    "LongCat 2.0 Intelligence, Performance & Price Analysis",
    "LongCat 2.0 was released on June 29, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "LongCat 2.0 Intelligence, Performance & Price Analysis",
    "LongCat 2.0 has a context window of 1.0M tokens. This determines how much text and conversation history the model can process in a single request.",
    "LongCat 2.0 is a Mixture of Experts (MoE) model with 1.6 trillion total parameters, but only 48 billion active parameters are used during inference."
   ]
  },
  {
   "url": "https://sunsetbrowser.app/blog/china-open-source-llm-eruption-2026-which-to-use-en",
   "title": "China's Open-Source LLM Eruption (2026): DeepSeek V4, Kimi K2.6, GLM-5.1, Qwen 3.6, MiniMax M3 \u2014 Which One to Pick",
   "published_at": "2026-06-08T00:00:00",
   "organization": null,
   "model": null,
   "release_date": "2026-04-07",
   "access_type": null,
   "license": null,
   "significance": "1.6T-total / 49B-active MoE with 1M context",
   "organization_evidence": null,
   "model_evidence": null,
   "release_date_evidence": [
    "Here's the timeline.",
    "In the three weeks from April 7 to April 24, 2026, GLM-5.1, Qwen 3.6, Kimi K2.6, and DeepSeek V4 all went live."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## DeepSeek V4 \u2014 1M context, MIT license, priced almost unreasonably low",
    "Per multiple April 2026 benchmark reports, V4-Pro is a 1.6T-total / 49B-active MoE, while V4-Flash is 284B-total / 13B-active."
   ]
  },
  {
   "url": "https://sunsetbrowser.app/blog/china-open-source-llm-eruption-2026-which-to-use-en",
   "title": "China's Open-Source LLM Eruption (2026): DeepSeek V4, Kimi K2.6, GLM-5.1, Qwen 3.6, MiniMax M3 \u2014 Which One to Pick",
   "published_at": "2026-06-08T00:00:00",
   "organization": null,
   "model": null,
   "release_date": "2026-04-07",
   "access_type": null,
   "license": null,
   "significance": "Qwen 3.6-35B-A3B and Qwen3.6-27B release, global usage champion",
   "organization_evidence": null,
   "model_evidence": null,
   "release_date_evidence": [
    "Here's the timeline.",
    "In the three weeks from April 7 to April 24, 2026, GLM-5.1, Qwen 3.6, Kimi K2.6, and DeepSeek V4 all went live."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## Qwen 3.6 \u2014 global usage champion, deepest ecosystem",
    "The reports also note April open-weight releases like Qwen3.6-35B-A3B and Qwen3.6-27B."
   ]
  },
  {
   "url": "https://sunsetbrowser.app/blog/china-open-source-llm-eruption-2026-which-to-use-en",
   "title": "China's Open-Source LLM Eruption (2026): DeepSeek V4, Kimi K2.6, GLM-5.1, Qwen 3.6, MiniMax M3 \u2014 Which One to Pick",
   "published_at": "2026-06-08T00:00:00",
   "organization": "Moonshot",
   "model": null,
   "release_date": "2026-04-07",
   "access_type": null,
   "license": null,
   "significance": "256K context and 300-agent swarm design",
   "organization_evidence": [
    "## Kimi K2.6 \u2014 stacking \"swarm exploration,\" not benchmarks",
    "Moonshot's Kimi K2.6 takes a different road."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Here's the timeline.",
    "In the three weeks from April 7 to April 24, 2026, GLM-5.1, Qwen 3.6, Kimi K2.6, and DeepSeek V4 all went live."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## Kimi K2.6 \u2014 stacking \"swarm exploration,\" not benchmarks",
    "Its context window is 256K and it lands around 58.6% on SWE-Bench Pro \u2014 not the flashiest numbers in this batch."
   ]
  },
  {
   "url": "https://sunsetbrowser.app/blog/china-open-source-llm-eruption-2026-which-to-use-en",
   "title": "China's Open-Source LLM Eruption (2026): DeepSeek V4, Kimi K2.6, GLM-5.1, Qwen 3.6, MiniMax M3 \u2014 Which One to Pick",
   "published_at": "2026-06-08T00:00:00",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-04-07",
   "access_type": null,
   "license": null,
   "significance": "744B MoE / 40B active under MIT license",
   "organization_evidence": [
    "## GLM-5.1 \u2014 a 744B MoE under MIT, friendly to self-hosting",
    "Z.ai's GLM-5.1 is the one in this wave I'd flag hardest for the self-hosting crowd."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Here's the timeline.",
    "In the three weeks from April 7 to April 24, 2026, GLM-5.1, Qwen 3.6, Kimi K2.6, and DeepSeek V4 all went live."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## GLM-5.1 \u2014 a 744B MoE under MIT, friendly to self-hosting",
    "Per the reports, it's a 744B MoE / 40B active, MIT-licensed, around 58.4% on SWE-Bench Pro \u2014 and the team explicitly notes it runs on 8\u00d7 H100 with vLLM."
   ]
  },
  {
   "url": "https://sunsetbrowser.app/blog/china-open-source-llm-eruption-2026-which-to-use-en",
   "title": "China's Open-Source LLM Eruption (2026): DeepSeek V4, Kimi K2.6, GLM-5.1, Qwen 3.6, MiniMax M3 \u2014 Which One to Pick",
   "published_at": "2026-06-08T00:00:00",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "First open-weight model with frontier-grade coding, 1M context, and native multimodal input",
   "organization_evidence": [
    "## MiniMax M3 \u2014 June's new king, but put an asterisk on the score",
    "Per reporting from The Decoder and TechTimes, MiniMax launched M3 on June 1, 2026, billing it as the \"first open-weight model with frontier-grade coding, 1M context, and native multimodal input\" all at once."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Here's the timeline.",
    "Then on June 1, MiniMax came back with M3."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## MiniMax M3 \u2014 June's new king, but put an asterisk on the score",
    "Per reporting from The Decoder and TechTimes, MiniMax launched M3 on June 1, 2026, billing it as the \"first open-weight model with frontier-grade coding, 1M context, and native multimodal input\" all at once."
   ]
  },
  {
   "url": "https://subagentic.ai/posts/thinking-machines-inkling-975b-open-weights",
   "title": "Thinking Machines Inkling: Mira Murati's 975B Open-Weights Agentic MoE Model Released Under Apache 2.0",
   "published_at": "2026-07-16T15:11:21",
   "organization": "Thinking Machines Lab",
   "model": "Inkling",
   "release_date": "2026-07-15",
   "access_type": "open-weights",
   "license": "Apache 2.0",
   "significance": "975-billion-parameter Mixture-of-Experts model with 1M context, Apache 2.0 license, and full agentic coding support",
   "organization_evidence": [
    "# Thinking Machines Inkling: Mira Murati's 975B Open-Weights Agentic MoE Model Released Under Apache 2.0",
    "Mira Murati's Thinking Machines Lab drops Inkling: a 975B open-weights MoE model with 1M context, Apache 2.0 license, and full agentic coding support.",
    "**Thinking Machines Lab** \u2014 the company founded by former OpenAI CTO Mira Murati \u2014 released **Inkling** yesterday: a 975-billion-parameter Mixture-of-Experts model with full weights available on Hugging Face under Apache 2.0."
   ],
   "model_evidence": [
    "# Thinking Machines Inkling: Mira Murati's 975B Open-Weights Agentic MoE Model Released Under Apache 2.0",
    "Mira Murati's Thinking Machines Lab drops Inkling: a 975B open-weights MoE model with 1M context, Apache 2.0 license, and full agentic coding support."
   ],
   "release_date_evidence": [
    "# Thinking Machines Inkling: Mira Murati's 975B Open-Weights Agentic MoE Model Released Under Apache 2.0",
    "July 16, 2026 \u00b7 4 min \u00b7 767 words \u00b7 Writer Agent (Claude Sonnet 4.6)",
    "1. [Inkling: Our open-weights model \u2014 Thinking Machines Lab, July 15 2026](https://thinkingmachines.ai/news/introducing-inkling/)"
   ],
   "access_type_evidence": [
    "# Thinking Machines Inkling: Mira Murati's 975B Open-Weights Agentic MoE Model Released Under Apache 2.0",
    "Mira Murati's Thinking Machines Lab drops Inkling: a 975B open-weights MoE model with 1M context, Apache 2.0 license, and full agentic coding support.",
    "The open-weights AI space just welcomed its most significant new entrant of 2026."
   ],
   "license_evidence": [
    "# Thinking Machines Inkling: Mira Murati's 975B Open-Weights Agentic MoE Model Released Under Apache 2.0",
    "Mira Murati's Thinking Machines Lab drops Inkling: a 975B open-weights MoE model with 1M context, Apache 2.0 license, and full agentic coding support.",
    "- **Apache 2.0 license** \u2014 commercially permissive, modification allowed"
   ],
   "significance_evidence": [
    "# Thinking Machines Inkling: Mira Murati's 975B Open-Weights Agentic MoE Model Released Under Apache 2.0",
    "Mira Murati's Thinking Machines Lab drops Inkling: a 975B open-weights MoE model with 1M context, Apache 2.0 license, and full agentic coding support."
   ]
  },
  {
   "url": "https://benchr.org/timeline",
   "title": "AI Model Release Timeline 2025\u20132026 \u2014 benchr",
   "published_at": null,
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "1M context, 384K output. V4-Pro and V4-Flash offering very cheap commercial API and self-hosting options.",
   "organization_evidence": [
    "Release Timeline \u00b7 Reviewed July 23, 2026",
    "# AI model release timeline",
    "Major frontier and open-weight releases, newest first, with dated pricing and the key change. Reviewed against provider sources on July 23, 2026.",
    "April 24, 2026",
    "DeepSeek V4-Pro & V4-Flash Open (MIT)",
    "DeepSeek \u00b7 [Review \u2192](articles/deepseek-review)"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Release Timeline \u00b7 Reviewed July 23, 2026",
    "# AI model release timeline",
    "Major frontier and open-weight releases, newest first, with dated pricing and the key change. Reviewed against provider sources on July 23, 2026.",
    "April 24, 2026",
    "DeepSeek V4-Pro & V4-Flash Open (MIT)"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Release Timeline \u00b7 Reviewed July 23, 2026",
    "# AI model release timeline",
    "Major frontier and open-weight releases, newest first, with dated pricing and the key change. Reviewed against provider sources on July 23, 2026.",
    "April 24, 2026",
    "DeepSeek V4-Pro & V4-Flash Open (MIT)",
    "V4-Pro: $0.435/$0.87 per 1M; cache hit $0.003625. V4-Flash: $0.14/$0.28 \u2014 cheapest commercial API. 1M context, 384K output. MIT license. Self-hosted or API."
   ]
  },
  {
   "url": "https://benchr.org/timeline",
   "title": "AI Model Release Timeline 2025\u20132026 \u2014 benchr",
   "published_at": null,
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-04-22",
   "access_type": null,
   "license": null,
   "significance": "Dense 27B and MoE 35B-A3B variants with 262K native context and strong multilingual capabilities.",
   "organization_evidence": [
    "Release Timeline \u00b7 Reviewed July 23, 2026",
    "# AI model release timeline",
    "Major frontier and open-weight releases, newest first, with dated pricing and the key change. Reviewed against provider sources on July 23, 2026.",
    "April 22, 2026",
    "Qwen 3.6 Open (Apache 2.0)",
    "Alibaba \u00b7 [Review \u2192](articles/qwen-review)"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Release Timeline \u00b7 Reviewed July 23, 2026",
    "# AI model release timeline",
    "Major frontier and open-weight releases, newest first, with dated pricing and the key change. Reviewed against provider sources on July 23, 2026.",
    "April 22, 2026",
    "Qwen 3.6 Open (Apache 2.0)"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Release Timeline \u00b7 Reviewed July 23, 2026",
    "# AI model release timeline",
    "Major frontier and open-weight releases, newest first, with dated pricing and the key change. Reviewed against provider sources on July 23, 2026.",
    "April 22, 2026",
    "Qwen 3.6 Open (Apache 2.0)",
    "Apache 2.0. Dense 27B and MoE 35B-A3B variants. 262K native context (~1M extended). Strong multilingual. Self-hosted \u2014 no verified per-token API price."
   ]
  },
  {
   "url": "https://benchr.org/timeline",
   "title": "AI Model Release Timeline 2025\u20132026 \u2014 benchr",
   "published_at": null,
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-04-20",
   "access_type": null,
   "license": null,
   "significance": "1T total / 32B active MoE with 256K context and good multilingual performance.",
   "organization_evidence": [
    "Release Timeline \u00b7 Reviewed July 23, 2026",
    "# AI model release timeline",
    "Major frontier and open-weight releases, newest first, with dated pricing and the key change. Reviewed against provider sources on July 23, 2026.",
    "April 20, 2026",
    "Kimi K2.6 Open (Modified MIT)",
    "Moonshot AI \u00b7 [Review \u2192](articles/kimi-review)"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Release Timeline \u00b7 Reviewed July 23, 2026",
    "# AI model release timeline",
    "Major frontier and open-weight releases, newest first, with dated pricing and the key change. Reviewed against provider sources on July 23, 2026.",
    "April 20, 2026",
    "Kimi K2.6 Open (Modified MIT)"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Release Timeline \u00b7 Reviewed July 23, 2026",
    "# AI model release timeline",
    "Major frontier and open-weight releases, newest first, with dated pricing and the key change. Reviewed against provider sources on July 23, 2026.",
    "April 20, 2026",
    "Kimi K2.6 Open (Modified MIT)",
    "$0.95/$4.00 per 1M; cached $0.16. 256K context. 1T total / 32B active MoE. Good multilingual performance."
   ]
  },
  {
   "url": "https://benchr.org/timeline",
   "title": "AI Model Release Timeline 2025\u20132026 \u2014 benchr",
   "published_at": null,
   "organization": "Mistral",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "Public preview offering multimodal and reasoning capabilities, self-hostable on 4 GPUs with 256K context.",
   "organization_evidence": [
    "Release Timeline \u00b7 Reviewed July 23, 2026",
    "# AI model release timeline",
    "Major frontier and open-weight releases, newest first, with dated pricing and the key change. Reviewed against provider sources on July 23, 2026.",
    "April 2026",
    "Mistral Medium 3.5 Open (Modified MIT)",
    "Mistral"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Release Timeline \u00b7 Reviewed July 23, 2026",
    "# AI model release timeline",
    "Major frontier and open-weight releases, newest first, with dated pricing and the key change. Reviewed against provider sources on July 23, 2026.",
    "April 2026",
    "Mistral Medium 3.5 Open (Modified MIT)"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Release Timeline \u00b7 Reviewed July 23, 2026",
    "# AI model release timeline",
    "Major frontier and open-weight releases, newest first, with dated pricing and the key change. Reviewed against provider sources on July 23, 2026.",
    "April 2026",
    "Mistral Medium 3.5 Open (Modified MIT)",
    "$1.50/$7.50 per 1M. Public preview. 256K context. Multimodal + reasoning. Self-host on 4 GPUs."
   ]
  },
  {
   "url": "https://benchr.org/timeline",
   "title": "AI Model Release Timeline 2025\u20132026 \u2014 benchr",
   "published_at": null,
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-02-16",
   "access_type": null,
   "license": null,
   "significance": "Flagship 397B MoE with 17B active parameters, family of nine sizes, native 262K context.",
   "organization_evidence": [
    "Release Timeline \u00b7 Reviewed July 23, 2026",
    "# AI model release timeline",
    "Major frontier and open-weight releases, newest first, with dated pricing and the key change. Reviewed against provider sources on July 23, 2026.",
    "February 16, 2026",
    "Qwen 3.5 Open",
    "Alibaba"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Release Timeline \u00b7 Reviewed July 23, 2026",
    "# AI model release timeline",
    "Major frontier and open-weight releases, newest first, with dated pricing and the key change. Reviewed against provider sources on July 23, 2026.",
    "February 16, 2026",
    "Qwen 3.5 Open"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Release Timeline \u00b7 Reviewed July 23, 2026",
    "# AI model release timeline",
    "Major frontier and open-weight releases, newest first, with dated pricing and the key change. Reviewed against provider sources on July 23, 2026.",
    "February 16, 2026",
    "Qwen 3.5 Open",
    "Flagship 397B MoE, 17B active; family of nine sizes, most Apache 2.0. Native 262K context. Superseded by Qwen3.6 in April."
   ]
  },
  {
   "url": "https://mer.vin/2026/06/open-weight-ai-release-week-25-models-across-llms-image-audio-video-and-3d-june-2026",
   "title": "Open-Weight AI Release Week: 25+ Models Across LLMs, Image, Audio, Video, and 3D (June 2026)",
   "published_at": "2026-06-08T15:07:56",
   "organization": "NVIDIA",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "First openly weighted 550B hybrid Mamba-Transformer; datacenter agentic scale with ~10% active params",
   "organization_evidence": [
    "## Large language models and edge chat",
    "| Model | Org | Key specs | Why it matters |",
    "| **Nemotron 3 Ultra** | NVIDIA | 550B hybrid Mamba\u2013MoE; **55B active**; **1M** context; **89.1 MMLU**; NVFP4 variant ~5\u00d7 throughput on Blackwell | First openly weighted **550B hybrid Mamba\u2013Transformer**; datacenter agentic scale with ~10% active params |"
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## Large language models and edge chat",
    "| Model | Org | Key specs | Why it matters |",
    "| **Nemotron 3 Ultra** | NVIDIA | 550B hybrid Mamba\u2013MoE; **55B active**; **1M** context; **89.1 MMLU**; NVFP4 variant ~5\u00d7 throughput on Blackwell | First openly weighted **550B hybrid Mamba\u2013Transformer**; datacenter agentic scale with ~10% active params |"
   ]
  },
  {
   "url": "https://mer.vin/2026/06/open-weight-ai-release-week-25-models-across-llms-image-audio-video-and-3d-june-2026",
   "title": "Open-Weight AI Release Week: 25+ Models Across LLMs, Image, Audio, Video, and 3D (June 2026)",
   "published_at": "2026-06-08T15:07:56",
   "organization": "Google",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Most deployable multimodal open model of the week \u2014 laptop-class with Apache 2.0 weights",
   "organization_evidence": [
    "## Large language models and edge chat",
    "| Model | Org | Key specs | Why it matters |",
    "| **Gemma 4 12B** | Google | Encoder-free any-to-any (text/image/audio/video); **256k** context; **140+** languages; **AIME 2026 77.5**; 23-checkpoint QAT wave (mobile ONNX + MLX) | Most deployable multimodal open model of the week \u2014 laptop-class with Apache 2.0 weights |"
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## Large language models and edge chat",
    "| Model | Org | Key specs | Why it matters |",
    "| **Gemma 4 12B** | Google | Encoder-free any-to-any (text/image/audio/video); **256k** context; **140+** languages; **AIME 2026 77.5**; 23-checkpoint QAT wave (mobile ONNX + MLX) | Most deployable multimodal open model of the week \u2014 laptop-class with Apache 2.0 weights |"
   ]
  },
  {
   "url": "https://mer.vin/2026/06/open-weight-ai-release-week-25-models-across-llms-image-audio-video-and-3d-june-2026",
   "title": "Open-Weight AI Release Week: 25+ Models Across LLMs, Image, Audio, Video, and 3D (June 2026)",
   "published_at": "2026-06-08T15:07:56",
   "organization": "Liquid AI",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Strong on-device math/reasoning per active parameter",
   "organization_evidence": [
    "## Large language models and edge chat",
    "| Model | Org | Key specs | Why it matters |",
    "| **LFM2.5-8B-A1B** | Liquid AI | Edge MoE; **~1.5B active**; 128k ctx; **MATH500 88.8**; MLX-ready | Strong on-device math/reasoning per active parameter |"
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## Large language models and edge chat",
    "| Model | Org | Key specs | Why it matters |",
    "| **LFM2.5-8B-A1B** | Liquid AI | Edge MoE; **~1.5B active**; 128k ctx; **MATH500 88.8**; MLX-ready | Strong on-device math/reasoning per active parameter |"
   ]
  },
  {
   "url": "https://mer.vin/2026/06/open-weight-ai-release-week-25-models-across-llms-image-audio-video-and-3d-june-2026",
   "title": "Open-Weight AI Release Week: 25+ Models Across LLMs, Image, Audio, Video, and 3D (June 2026)",
   "published_at": "2026-06-08T15:07:56",
   "organization": "JetBrains",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Near\u2013Qwen3-14B coding quality at much lower active width for IDE/agent tooling",
   "organization_evidence": [
    "## Large language models and edge chat",
    "| Model | Org | Key specs | Why it matters |",
    "| **Mellum2-12B-A2.5B-Thinking** | JetBrains | First open JetBrains MoE; **2.5B active** (8 of 64 experts); 131k ctx; LiveCodeBench v6 **69.9**; Apache 2.0 | Near\u2013Qwen3-14B coding quality at much lower active width for IDE/agent tooling |"
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## Large language models and edge chat",
    "| Model | Org | Key specs | Why it matters |",
    "| **Mellum2-12B-A2.5B-Thinking** | JetBrains | First open JetBrains MoE; **2.5B active** (8 of 64 experts); 131k ctx; LiveCodeBench v6 **69.9**; Apache 2.0 | Near\u2013Qwen3-14B coding quality at much lower active width for IDE/agent tooling |"
   ]
  },
  {
   "url": "https://promptquorum.com/local-llms/local-llm-model-updates-2026",
   "title": "Local LLM Model Updates 2026: Every Major Open-Weight Release This Year",
   "published_at": "2026-04-04T12:56:06",
   "organization": "Google",
   "model": null,
   "release_date": "2026-02-01",
   "access_type": null,
   "license": null,
   "significance": "1B, 4B, 9B, and 27B variants, vision support on all sizes",
   "organization_evidence": [
    "Key Takeaways",
    "- **Biggest Q1 2026 release**: Google Gemma 3 (February 2026) -- 1B, 4B, 9B, and 27B variants, vision support on all sizes, Apache 2.0 licence.",
    "## Which Local LLM Models Were Released in Q1 2026?",
    "As of April 2026, the notable open-weight model releases from January-April 2026. All models below are available in various quantization formats -- see [quantization guide](/local-llms/llm-quantization-explained) for details on Q4 vs Q5 tradeoffs:",
    "| Model | Released | Developer | Key Feature | Ollama |",
    "| Gemma 3 (all sizes) | February 2026 | Google | Vision on all sizes, 128K context, Apache 2.0 | ollama run gemma3:9b |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Key Takeaways",
    "- **Biggest Q1 2026 release**: Google Gemma 3 (February 2026) -- 1B, 4B, 9B, and 27B variants, vision support on all sizes, Apache 2.0 licence.",
    "## Which Local LLM Models Were Released in Q1 2026?",
    "As of April 2026, the notable open-weight model releases from January-April 2026. All models below are available in various quantization formats -- see [quantization guide](/local-llms/llm-quantization-explained) for details on Q4 vs Q5 tradeoffs:",
    "| Model | Released | Developer | Key Feature | Ollama |",
    "| Gemma 3 (all sizes) | February 2026 | Google | Vision on all sizes, 128K context, Apache 2.0 | ollama run gemma3:9b |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Key Takeaways",
    "- **Biggest Q1 2026 release**: Google Gemma 3 (February 2026) -- 1B, 4B, 9B, and 27B variants, vision support on all sizes, Apache 2.0 licence.",
    "## Which Local LLM Models Were Released in Q1 2026?",
    "As of April 2026, the notable open-weight model releases from January-April 2026. All models below are available in various quantization formats -- see [quantization guide](/local-llms/llm-quantization-explained) for details on Q4 vs Q5 tradeoffs:",
    "| Model | Released | Developer | Key Feature | Ollama |",
    "| Gemma 3 (all sizes) | February 2026 | Google | Vision on all sizes, 128K context, Apache 2.0 | ollama run gemma3:9b |"
   ]
  },
  {
   "url": "https://promptquorum.com/local-llms/local-llm-model-updates-2026",
   "title": "Local LLM Model Updates 2026: Every Major Open-Weight Release This Year",
   "published_at": "2026-04-04T12:56:06",
   "organization": "Meta",
   "model": null,
   "release_date": "2026-03-01",
   "access_type": null,
   "license": null,
   "significance": "MoE architecture preview, 10M token context claimed",
   "organization_evidence": [
    "## Which Local LLM Models Were Released in Q1 2026?",
    "As of April 2026, the notable open-weight model releases from January-April 2026. All models below are available in various quantization formats -- see [quantization guide](/local-llms/llm-quantization-explained) for details on Q4 vs Q5 tradeoffs:",
    "| Model | Released | Developer | Key Feature | Ollama |",
    "| Llama 4 Scout (preview) | March 2026 | Meta | MoE architecture preview, 10M token context claimed | Not yet available |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## Which Local LLM Models Were Released in Q1 2026?",
    "As of April 2026, the notable open-weight model releases from January-April 2026. All models below are available in various quantization formats -- see [quantization guide](/local-llms/llm-quantization-explained) for details on Q4 vs Q5 tradeoffs:",
    "| Model | Released | Developer | Key Feature | Ollama |",
    "| Llama 4 Scout (preview) | March 2026 | Meta | MoE architecture preview, 10M token context claimed | Not yet available |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## Which Local LLM Models Were Released in Q1 2026?",
    "As of April 2026, the notable open-weight model releases from January-April 2026. All models below are available in various quantization formats -- see [quantization guide](/local-llms/llm-quantization-explained) for details on Q4 vs Q5 tradeoffs:",
    "| Model | Released | Developer | Key Feature | Ollama |",
    "| Llama 4 Scout (preview) | March 2026 | Meta | MoE architecture preview, 10M token context claimed | Not yet available |"
   ]
  },
  {
   "url": "https://promptquorum.com/local-llms/local-llm-model-updates-2026",
   "title": "Local LLM Model Updates 2026: Every Major Open-Weight Release This Year",
   "published_at": "2026-04-04T12:56:06",
   "organization": "Mistral AI",
   "model": null,
   "release_date": "2026-02-01",
   "access_type": null,
   "license": null,
   "significance": "Improved instruction-following over Small 3.1",
   "organization_evidence": [
    "## Which Local LLM Models Were Released in Q1 2026?",
    "As of April 2026, the notable open-weight model releases from January-April 2026. All models below are available in various quantization formats -- see [quantization guide](/local-llms/llm-quantization-explained) for details on Q4 vs Q5 tradeoffs:",
    "| Model | Released | Developer | Key Feature | Ollama |",
    "| Mistral Small 3.2 | February 2026 | Mistral AI | Improved instruction-following over Small 3.1 | ollama run mistral-small3.2 |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## Which Local LLM Models Were Released in Q1 2026?",
    "As of April 2026, the notable open-weight model releases from January-April 2026. All models below are available in various quantization formats -- see [quantization guide](/local-llms/llm-quantization-explained) for details on Q4 vs Q5 tradeoffs:",
    "| Model | Released | Developer | Key Feature | Ollama |",
    "| Mistral Small 3.2 | February 2026 | Mistral AI | Improved instruction-following over Small 3.1 | ollama run mistral-small3.2 |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## Which Local LLM Models Were Released in Q1 2026?",
    "As of April 2026, the notable open-weight model releases from January-April 2026. All models below are available in various quantization formats -- see [quantization guide](/local-llms/llm-quantization-explained) for details on Q4 vs Q5 tradeoffs:",
    "| Model | Released | Developer | Key Feature | Ollama |",
    "| Mistral Small 3.2 | February 2026 | Mistral AI | Improved instruction-following over Small 3.1 | ollama run mistral-small3.2 |"
   ]
  },
  {
   "url": "https://promptquorum.com/local-llms/local-llm-model-updates-2026",
   "title": "Local LLM Model Updates 2026: Every Major Open-Weight Release This Year",
   "published_at": "2026-04-04T12:56:06",
   "organization": "Microsoft",
   "model": null,
   "release_date": "2026-01-01",
   "access_type": null,
   "license": null,
   "significance": "3.8B, 70% HumanEval, 128K context",
   "organization_evidence": [
    "## Which Local LLM Models Were Released in Q1 2026?",
    "As of April 2026, the notable open-weight model releases from January-April 2026. All models below are available in various quantization formats -- see [quantization guide](/local-llms/llm-quantization-explained) for details on Q4 vs Q5 tradeoffs:",
    "| Model | Released | Developer | Key Feature | Ollama |",
    "| Phi-4 Mini | January 2026 | Microsoft | 3.8B, 70% HumanEval, 128K context | ollama run phi4-mini |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## Which Local LLM Models Were Released in Q1 2026?",
    "As of April 2026, the notable open-weight model releases from January-April 2026. All models below are available in various quantization formats -- see [quantization guide](/local-llms/llm-quantization-explained) for details on Q4 vs Q5 tradeoffs:",
    "| Model | Released | Developer | Key Feature | Ollama |",
    "| Phi-4 Mini | January 2026 | Microsoft | 3.8B, 70% HumanEval, 128K context | ollama run phi4-mini |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## Which Local LLM Models Were Released in Q1 2026?",
    "As of April 2026, the notable open-weight model releases from January-April 2026. All models below are available in various quantization formats -- see [quantization guide](/local-llms/llm-quantization-explained) for details on Q4 vs Q5 tradeoffs:",
    "| Model | Released | Developer | Key Feature | Ollama |",
    "| Phi-4 Mini | January 2026 | Microsoft | 3.8B, 70% HumanEval, 128K context | ollama run phi4-mini |"
   ]
  },
  {
   "url": "https://faq.com.tw/en/ai-ml/2026-07-19-thinking-machines-inkling-open-weight-model-en",
   "title": "Mira Murati's Thinking Machines Releases Inkling: A 975B-Parameter Open-Weight Challenger",
   "published_at": "2026-07-19T00:00:00",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "A 975-billion-parameter mixture-of-experts model featuring controllable reasoning, a 1M-token context window, and native multimodality.",
   "organization_evidence": [
    "# Mira Murati's Thinking Machines Releases Inkling: A 975B-Parameter Open-Weight Challenger",
    "Former OpenAI CTO Mira Murati's Thinking Machines Lab has launched Inkling, a 975-billion-parameter mixture-of-experts model released open-weight under Apache 2.0."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Mira Murati's Thinking Machines Releases Inkling: A 975B-Parameter Open-Weight Challenger",
    "On July 15, 2026, Thinking Machines proved the skeptics wrong by releasing Inkling, a 975-billion-parameter mixture-of-experts model available to anyone under an Apache 2.0 license."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Mira Murati's Thinking Machines Releases Inkling: A 975B-Parameter Open-Weight Challenger",
    "Former OpenAI CTO Mira Murati's Thinking Machines Lab has launched Inkling, a 975-billion-parameter mixture-of-experts model released open-weight under Apache 2.0.",
    "The multimodal model, built in just nine months with a 200-person team, directly challenges the closed-model paradigm with controllable reasoning, a 1M-token context window, and a fine-tuning platform designed to let enterprises train proprietary AI without licensing fees."
   ]
  },
  {
   "url": "https://namiru.ai/blog/glm-5-china-s-open-source-giant-that-rivals-claude-and-gpt",
   "title": "GLM-5: China's Open-Source Giant That Rivals Claude and GPT",
   "published_at": "2026-02-17T18:47:29",
   "organization": "Zhipu AI",
   "model": null,
   "release_date": "2026-02-11",
   "access_type": null,
   "license": null,
   "significance": "744 billion parameters, Mixture-of-Experts architecture, and benchmarks within striking distance of top western proprietary models.",
   "organization_evidence": [
    "# GLM-5: China's Open-Source Giant That Rivals Claude and GPT",
    "Zhipu AI's GLM-5 comes with 744 billion parameters, ships under the MIT license, and benchmarks within striking distance of Claude Opus 4.5 and GPT-5.2."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# GLM-5: China's Open-Source Giant That Rivals Claude and GPT",
    "On February 11, 2026, Chinese AI lab Zhipu AI (now rebranded as Z.ai) released GLM-5, a new open-source large language model that competes directly with Claude Opus 4.5, GPT-5.2, and Gemini 3 Pro on coding, reasoning, and agentic benchmarks."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# GLM-5: China's Open-Source Giant That Rivals Claude and GPT",
    "Zhipu AI's GLM-5 comes with 744 billion parameters, ships under the MIT license, and benchmarks within striking distance of Claude Opus 4.5 and GPT-5.2."
   ]
  },
  {
   "url": "https://theagenttimes.com/articles/thinking-machines-lab-releases-inkling-open-weight-multimoda-ba3abfe1",
   "title": "Thinking Machines Lab Releases Inkling, Open-Weight Multimodal Model Fine-Tunable on Tinker",
   "published_at": "2026-07-15T20:02:02",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "multimodal model trained from scratch that reasons efficiently across text, image, and audio modalities",
   "organization_evidence": [
    "# Thinking Machines Lab Releases Inkling, Open-Weight Multimodal Model Fine-Tunable on Tinker",
    "Thinking Machines Lab, led by former OpenAI CTO Mira Murati, released Inkling \u2014 an open-weight multimodal model trained from scratch, available for fine-tuning on the company's Tinker platform as of July 15, 2026."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Thinking Machines Lab Releases Inkling, Open-Weight Multimodal Model Fine-Tunable on Tinker",
    "Thinking Machines Lab, led by former OpenAI CTO Mira Murati, released Inkling \u2014 an open-weight multimodal model trained from scratch, available for fine-tuning on the company's Tinker platform as of July 15, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Thinking Machines Lab Releases Inkling, Open-Weight Multimodal Model Fine-Tunable on Tinker",
    "The accompanying post from the official Thinking Machines account stated that \u201cInkling reasons efficiently across text, image, and audio modalities\u201d and that the company is \u201cmaking the full weights available,\u201d [according to the same thread](https://x.com/erikbryn/status/2077481873450693119). A link to the company\u2019s blog at thinkingmachines.ai and an \u201cInkling Playground\u201d were included in the thread."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/ling-3-0-flash",
   "title": "Ling 3.0 Flash - Intelligence, Performance & Price Analysis | Artificial Analysis",
   "published_at": null,
   "organization": "InclusionAI",
   "model": null,
   "release_date": "2026-08-04",
   "access_type": null,
   "license": null,
   "significance": "A reasoning model with 124B parameters and a 262k context window.",
   "organization_evidence": [
    "# Ling 3.0 Flash Intelligence, Performance & Price Analysis",
    "Ling 3.0 Flash was created by InclusionAI."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Ling 3.0 Flash Intelligence, Performance & Price Analysis",
    "Ling 3.0 Flash was released on August 4, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Ling 3.0 Flash Intelligence, Performance & Price Analysis",
    "Ling 3.0 Flash is amongst the leading models in intelligence and reasonably priced when comparing to other open weight models of similar size.",
    "Yes, Ling 3.0 Flash is a reasoning model."
   ]
  },
  {
   "url": "https://modeldiplomat.com/story/meituans-16t-ai-model-on-chinese-chips",
   "title": "Meituan's 1.6T AI Model on Chinese Chips",
   "published_at": "2026-07-07T03:39:05",
   "organization": "Meituan",
   "model": null,
   "release_date": "2026-07-07",
   "access_type": null,
   "license": null,
   "significance": "A 1.6-trillion-parameter Mixture-of-Experts language model trained entirely on Chinese chips.",
   "organization_evidence": [
    "Meituan's 1.6T AI Model on Chinese Chips"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Meituan's 1.6T AI Model on Chinese Chips"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Meituan's 1.6T AI Model on Chinese Chips"
   ]
  },
  {
   "url": "https://thedeepfeed.ai/posts/2026-06-24-open-weight-reasoning-gap-three-months",
   "title": "The open-weight reasoning gap is now 3.4 months, and the math is public",
   "published_at": "2026-06-24T00:00:00",
   "organization": "Zhipu",
   "model": null,
   "release_date": "2026-06-16",
   "access_type": null,
   "license": null,
   "significance": "leading open-weight model on the Artificial Analysis Intelligence Index v4.1 at 51, ahead of every Google model",
   "organization_evidence": [
    "# The open-weight reasoning gap is now 3.4 months, and the math is public",
    "On June 16, Z.ai (formerly Zhipu) released GLM-5.2 to open weights under an MIT license."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# The open-weight reasoning gap is now 3.4 months, and the math is public",
    "On June 16, Z.ai (formerly Zhipu) released GLM-5.2 to open weights under an MIT license."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# The open-weight reasoning gap is now 3.4 months, and the math is public",
    "On June 16, GLM-5.2 became the leading open-weight model on the Artificial Analysis Intelligence Index v4.1 at **51**, ahead of every Google model and **3.4 months** behind the equally-capable closed model."
   ]
  },
  {
   "url": "https://thedeepfeed.ai/posts/2026-06-24-open-weight-reasoning-gap-three-months",
   "title": "The open-weight reasoning gap is now 3.4 months, and the math is public",
   "published_at": "2026-06-24T00:00:00",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "1.6-trillion-parameter MoE that sets the record at 2.2 months behind the equally-capable closed model with vendor-reported GPQA Diamond of 90.1% and LiveCodeBench of 93.5%",
   "organization_evidence": [
    "# The open-weight reasoning gap is now 3.4 months, and the math is public",
    "DeepSeek V4 Pro fills in the rest of the reasoning picture."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# The open-weight reasoning gap is now 3.4 months, and the math is public",
    "DeepSeek V4 Pro, released in April, set the record at **2.2 months**, matching Claude Sonnet 4.6 from February."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# The open-weight reasoning gap is now 3.4 months, and the math is public",
    "DeepSeek V4 Pro, released in April, set the record at **2.2 months**, matching Claude Sonnet 4.6 from February.",
    "The [1.6-trillion-parameter MoE](https://arxiv.org/html/2606.19348v1) (49B active, 1M context, MIT) posts vendor-reported **GPQA Diamond of 90.1% and LiveCodeBench of 93.5%**, and Artificial Analysis independently ranks it third among open models on the index."
   ]
  },
  {
   "url": "https://thedeepfeed.ai/posts/2026-06-24-open-weight-reasoning-gap-three-months",
   "title": "The open-weight reasoning gap is now 3.4 months, and the math is public",
   "published_at": "2026-06-24T00:00:00",
   "organization": "MiniMax",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "1M context, multimodal, Intelligence Index v4.1 score of 44",
   "organization_evidence": [
    "# The open-weight reasoning gap is now 3.4 months, and the math is public"
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# The open-weight reasoning gap is now 3.4 months, and the math is public",
    "| Model | Type | License | Intelligence Index v4.1 | Notable reasoning marks | Price (in/out per 1M) | OSS lag |",
    "| MiniMax-M3 | Open | Open weights | 44 | 1M context, multimodal | sub-$1 tier | 3.4 mo |"
   ]
  },
  {
   "url": "https://thedeepfeed.ai/posts/2026-06-24-open-weight-reasoning-gap-three-months",
   "title": "The open-weight reasoning gap is now 3.4 months, and the math is public",
   "published_at": "2026-06-24T00:00:00",
   "organization": "Moonshot",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "trillion-parameter vision-language model that sits right behind top models with an Intelligence Index v4.1 score of 43",
   "organization_evidence": [
    "# The open-weight reasoning gap is now 3.4 months, and the math is public",
    "Kimi K2.6, a trillion-parameter vision-language model from Moonshot, [sits right behind it](https://www.deeplearning.ai/the-batch/kimi-k2-6-matches-open-qwen3-6-max-anddeepseek-v4-falls-just-behind-top-closed-models) and \u201cfalls just behind top closed models\u201d in The Batch\u2019s framing."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# The open-weight reasoning gap is now 3.4 months, and the math is public",
    "Kimi K2.6, a trillion-parameter vision-language model from Moonshot, [sits right behind it](https://www.deeplearning.ai/the-batch/kimi-k2-6-matches-open-qwen3-6-max-anddeepseek-v4-falls-just-behind-top-closed-models) and \u201cfalls just behind top closed models\u201d in The Batch\u2019s framing.",
    "| Model | Type | License | Intelligence Index v4.1 | Notable reasoning marks | Price (in/out per 1M) | OSS lag |",
    "| Kimi K2.6 | Open | Open weights | 43 | Just behind top closed | mid-tier | 2.4 mo |"
   ]
  },
  {
   "url": "https://huggingface.co/blog/FINAL-Bench/darwin-papers",
   "title": "Training-Free Reasoning at 88.89% on GPQA Diamond: How Darwin Family Hit Frontier Scores Without a Single Gradient Step",
   "published_at": "2026-05-15T03:05:43",
   "organization": "VIDRAFT",
   "model": "Darwin Family",
   "release_date": "2026-05",
   "access_type": "open-source",
   "license": null,
   "significance": "recombines weight spaces of existing checkpoints with zero gradient-based training, reaching GPQA Diamond 88.89%",
   "organization_evidence": [
    "Training-Free Reasoning at 88.89% on GPQA Diamond: How Darwin Family Hit Frontier Scores Without a Single Gradient Step",
    "VIDRAFT's **Darwin Family** evolves frontier-level reasoning LLMs by recombining the weight spaces of existing checkpoints, with **zero gradient-based training**."
   ],
   "model_evidence": [
    "Training-Free Reasoning at 88.89% on GPQA Diamond: How Darwin Family Hit Frontier Scores Without a Single Gradient Step",
    "VIDRAFT's **Darwin Family** evolves frontier-level reasoning LLMs by recombining the weight spaces of existing checkpoints, with **zero gradient-based training**."
   ],
   "release_date_evidence": [
    "Training-Free Reasoning at 88.89% on GPQA Diamond: How Darwin Family Hit Frontier Scores Without a Single Gradient Step",
    "\ud83d\udd17 Paper: [https://huggingface.co/papers/2605.14386](https://huggingface.co/papers/2605.14386) \ud83d\udd17 arXiv: [https://arxiv.org/abs/2605.14386](https://arxiv.org/abs/2605.14386) \ud83d\udd17 Model: [https://huggingface.co/FINAL-Bench/Darwin-28B-Opus](https://huggingface.co/FINAL-Bench/Darwin-28B-Opus)"
   ],
   "access_type_evidence": [
    "Training-Free Reasoning at 88.89% on GPQA Diamond: How Darwin Family Hit Frontier Scores Without a Single Gradient Step",
    "Hundreds of strong open-source LLMs already exist."
   ],
   "license_evidence": null,
   "significance_evidence": [
    "Training-Free Reasoning at 88.89% on GPQA Diamond: How Darwin Family Hit Frontier Scores Without a Single Gradient Step",
    "VIDRAFT's **Darwin Family** evolves frontier-level reasoning LLMs by recombining the weight spaces of existing checkpoints, with **zero gradient-based training**."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/glm-5-2",
   "title": "GLM-5.2 (max) - Intelligence, Performance & Price Analysis | Artificial Analysis",
   "published_at": null,
   "organization": "Z AI",
   "model": "GLM-5.2 (max)",
   "release_date": "2026-06-16",
   "access_type": "open weights",
   "license": "Mit",
   "significance": "753B parameter reasoning model with 1M context window scoring 53 on Artificial Analysis Intelligence Index",
   "organization_evidence": [
    "GLM-5.2 (max) Intelligence, Performance & Price Analysis",
    "GLM-5.2 (max) was created by Z AI."
   ],
   "model_evidence": [
    "GLM-5.2 (max) Intelligence, Performance & Price Analysis",
    "\u2022\nGLM-5.2",
    "GLM-5.2 (max) is amongst the leading models in intelligence, but particularly expensive when comparing to other open weight models of similar size."
   ],
   "release_date_evidence": [
    "GLM-5.2 (max) Intelligence, Performance & Price Analysis",
    "\u2022\nReleased June 2026",
    "GLM-5.2 (max) was released on June 16, 2026."
   ],
   "access_type_evidence": [
    "GLM-5.2 (max) Intelligence, Performance & Price Analysis",
    "\u2022\nOpen weights model",
    "Yes, GLM-5.2 (max) is open weights. The model weights are publicly available and can be downloaded for self-hosting."
   ],
   "license_evidence": [
    "GLM-5.2 (max) Intelligence, Performance & Price Analysis",
    "| License | [Mit](https://opensource.org/license/mit) |",
    "GLM-5.2 (max) is released under the Mit license. This license allows commercial use. [View license](https://opensource.org/license/mit)"
   ],
   "significance_evidence": [
    "GLM-5.2 (max) Intelligence, Performance & Price Analysis",
    "| Context window | 1M ~1500 A4 pages of size 12 Arial font |",
    "| Total parameters | 753B |",
    "GLM-5.2 (max) scores 53 on the Artificial Analysis Intelligence Index, placing it well above average among comparable models (median: 27).",
    "Yes, GLM-5.2 (max) is a reasoning model. It uses extended thinking or chain-of-thought reasoning to work through complex problems before providing an answer."
   ]
  },
  {
   "url": "https://news.cgtn.com/news/2026-07-31/China-s-open-source-AI-models-power-real-economy-1PdILcnaQRW/share_amp.html",
   "title": "China's open-source AI models power real economy",
   "published_at": "2026-07-31T00:00:00",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-27",
   "access_type": null,
   "license": null,
   "significance": "world's largest open-source AI model by parameter count with 2.8 trillion parameters",
   "organization_evidence": [
    "China's open-source AI models power real economy",
    "Chinese artificial intelligence (AI) startup Moonshot AI on Monday fully open-sourced its latest flagship large language model, Kimi K3, marking another milestone in China's rapidly expanding open-source AI ecosystem."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "China's open-source AI models power real economy",
    "Updated 2026.07.31 12:32 GMT+8",
    "Chinese artificial intelligence (AI) startup Moonshot AI on Monday fully open-sourced its latest flagship large language model, Kimi K3, marking another milestone in China's rapidly expanding open-source AI ecosystem."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "China's open-source AI models power real economy",
    "With 2.8 trillion parameters, Kimi K3 is currently the world's largest open-source AI model by parameter count."
   ]
  },
  {
   "url": "https://fazm.ai/blog/llm-releases-april-2026",
   "title": "Fazm - AI Computer Agent for macOS | Voice-Controlled Desktop Automation",
   "published_at": "2026-04-14T03:37:23",
   "organization": "Meta",
   "model": null,
   "release_date": "2026-04-05",
   "access_type": null,
   "license": null,
   "significance": "10 million token context window",
   "organization_evidence": [
    "LLM Releases April 2026: Complete Timeline of Every Model Launch",
    "## Complete List of LLM Releases in April 2026",
    "| Model | Organization | Date | Parameters | Type | Pricing (Input/Output per 1M tokens) | Key Feature | |---|---|---|---|---|---|---| | Gemini 2.5 Pro | Google | Apr 1 | Undisclosed | Proprietary | $3.50/$10.50 (under 200K) | 1M token context, native multimodal | | Claude Opus 4 | Anthropic | Apr 2 | Undisclosed | Proprietary | $15/$75 | Top coding benchmark scores, agentic | | Claude Sonnet 4 | Anthropic | Apr 2 | Undisclosed | Proprietary | $3/$15 | Balanced cost and performance | | Gemini 2.5 Flash | Google | Apr 3 | Undisclosed | Proprietary | $0.15/$0.60 | Low latency, high throughput | | Llama 4 Scout | Meta | Apr 5 | 109B (17B active) | Open source (MoE) | Free (self-host) | 10M token context window |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "LLM Releases April 2026: Complete Timeline of Every Model Launch",
    "## Complete List of LLM Releases in April 2026",
    "| Model | Organization | Date | Parameters | Type | Pricing (Input/Output per 1M tokens) | Key Feature | |---|---|---|---|---|---|---| | Gemini 2.5 Pro | Google | Apr 1 | Undisclosed | Proprietary | $3.50/$10.50 (under 200K) | 1M token context, native multimodal | | Claude Opus 4 | Anthropic | Apr 2 | Undisclosed | Proprietary | $15/$75 | Top coding benchmark scores, agentic | | Claude Sonnet 4 | Anthropic | Apr 2 | Undisclosed | Proprietary | $3/$15 | Balanced cost and performance | | Gemini 2.5 Flash | Google | Apr 3 | Undisclosed | Proprietary | $0.15/$0.60 | Low latency, high throughput | | Llama 4 Scout | Meta | Apr 5 | 109B (17B active) | Open source (MoE) | Free (self-host) | 10M token context window |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "LLM Releases April 2026: Complete Timeline of Every Model Launch",
    "## Complete List of LLM Releases in April 2026",
    "| Model | Organization | Date | Parameters | Type | Pricing (Input/Output per 1M tokens) | Key Feature | |---|---|---|---|---|---|---| | Gemini 2.5 Pro | Google | Apr 1 | Undisclosed | Proprietary | $3.50/$10.50 (under 200K) | 1M token context, native multimodal | | Claude Opus 4 | Anthropic | Apr 2 | Undisclosed | Proprietary | $15/$75 | Top coding benchmark scores, agentic | | Claude Sonnet 4 | Anthropic | Apr 2 | Undisclosed | Proprietary | $3/$15 | Balanced cost and performance | | Gemini 2.5 Flash | Google | Apr 3 | Undisclosed | Proprietary | $0.15/$0.60 | Low latency, high throughput | | Llama 4 Scout | Meta | Apr 5 | 109B (17B active) | Open source (MoE) | Free (self-host) | 10M token context window |"
   ]
  },
  {
   "url": "https://fazm.ai/blog/llm-releases-april-2026",
   "title": "Fazm - AI Computer Agent for macOS | Voice-Controlled Desktop Automation",
   "published_at": "2026-04-14T03:37:23",
   "organization": "Meta",
   "model": null,
   "release_date": "2026-04-05",
   "access_type": null,
   "license": null,
   "significance": "Multilingual, strong coding",
   "organization_evidence": [
    "LLM Releases April 2026: Complete Timeline of Every Model Launch",
    "## Complete List of LLM Releases in April 2026",
    "| Model | Organization | Date | Parameters | Type | Pricing (Input/Output per 1M tokens) | Key Feature | |---|---|---|---|---|---|---| | Gemini 2.5 Pro | Google | Apr 1 | Undisclosed | Proprietary | $3.50/$10.50 (under 200K) | 1M token context, native multimodal | | Claude Opus 4 | Anthropic | Apr 2 | Undisclosed | Proprietary | $15/$75 | Top coding benchmark scores, agentic | | Claude Sonnet 4 | Anthropic | Apr 2 | Undisclosed | Proprietary | $3/$15 | Balanced cost and performance | | Gemini 2.5 Flash | Google | Apr 3 | Undisclosed | Proprietary | $0.15/$0.60 | Low latency, high throughput | | Llama 4 Scout | Meta | Apr 5 | 109B (17B active) | Open source (MoE) | Free (self-host) | 10M token context window | | Llama 4 Maverick | Meta | Apr 5 | 400B (17B active) | Open source (MoE) | Free (self-host) | Multilingual, strong coding |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "LLM Releases April 2026: Complete Timeline of Every Model Launch",
    "## Complete List of LLM Releases in April 2026",
    "| Model | Organization | Date | Parameters | Type | Pricing (Input/Output per 1M tokens) | Key Feature | |---|---|---|---|---|---|---| | Gemini 2.5 Pro | Google | Apr 1 | Undisclosed | Proprietary | $3.50/$10.50 (under 200K) | 1M token context, native multimodal | | Claude Opus 4 | Anthropic | Apr 2 | Undisclosed | Proprietary | $15/$75 | Top coding benchmark scores, agentic | | Claude Sonnet 4 | Anthropic | Apr 2 | Undisclosed | Proprietary | $3/$15 | Balanced cost and performance | | Gemini 2.5 Flash | Google | Apr 3 | Undisclosed | Proprietary | $0.15/$0.60 | Low latency, high throughput | | Llama 4 Scout | Meta | Apr 5 | 109B (17B active) | Open source (MoE) | Free (self-host) | 10M token context window | | Llama 4 Maverick | Meta | Apr 5 | 400B (17B active) | Open source (MoE) | Free (self-host) | Multilingual, strong coding |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "LLM Releases April 2026: Complete Timeline of Every Model Launch",
    "## Complete List of LLM Releases in April 2026",
    "| Model | Organization | Date | Parameters | Type | Pricing (Input/Output per 1M tokens) | Key Feature | |---|---|---|---|---|---|---| | Gemini 2.5 Pro | Google | Apr 1 | Undisclosed | Proprietary | $3.50/$10.50 (under 200K) | 1M token context, native multimodal | | Claude Opus 4 | Anthropic | Apr 2 | Undisclosed | Proprietary | $15/$75 | Top coding benchmark scores, agentic | | Claude Sonnet 4 | Anthropic | Apr 2 | Undisclosed | Proprietary | $3/$15 | Balanced cost and performance | | Gemini 2.5 Flash | Google | Apr 3 | Undisclosed | Proprietary | $0.15/$0.60 | Low latency, high throughput | | Llama 4 Scout | Meta | Apr 5 | 109B (17B active) | Open source (MoE) | Free (self-host) | 10M token context window | | Llama 4 Maverick | Meta | Apr 5 | 400B (17B active) | Open source (MoE) | Free (self-host) | Multilingual, strong coding |"
   ]
  },
  {
   "url": "https://fazm.ai/blog/llm-releases-april-2026",
   "title": "Fazm - AI Computer Agent for macOS | Voice-Controlled Desktop Automation",
   "published_at": "2026-04-14T03:37:23",
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-04-08",
   "access_type": null,
   "license": null,
   "significance": "Dual-mode thinking, runs locally",
   "organization_evidence": [
    "LLM Releases April 2026: Complete Timeline of Every Model Launch",
    "## Complete List of LLM Releases in April 2026",
    "| Model | Organization | Date | Parameters | Type | Pricing (Input/Output per 1M tokens) | Key Feature | |---|---|---|---|---|---|---| | Gemini 2.5 Pro | Google | Apr 1 | Undisclosed | Proprietary | $3.50/$10.50 (under 200K) | 1M token context, native multimodal | | Claude Opus 4 | Anthropic | Apr 2 | Undisclosed | Proprietary | $15/$75 | Top coding benchmark scores, agentic | | Claude Sonnet 4 | Anthropic | Apr 2 | Undisclosed | Proprietary | $3/$15 | Balanced cost and performance | | Gemini 2.5 Flash | Google | Apr 3 | Undisclosed | Proprietary | $0.15/$0.60 | Low latency, high throughput | | Llama 4 Scout | Meta | Apr 5 | 109B (17B active) | Open source (MoE) | Free (self-host) | 10M token context window | | Llama 4 Maverick | Meta | Apr 5 | 400B (17B active) | Open source (MoE) | Free (self-host) | Multilingual, strong coding | | GPT-5 Turbo | OpenAI | Apr 7 | Undisclosed | Proprietary | $10/$30 | Native image + audio generation | | Qwen 3 (0.6B-72B) | Alibaba | Apr 8 | 0.6B to 72B | Open source | Free (Apache 2.0) | Dual-mode thinking, runs locally |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "LLM Releases April 2026: Complete Timeline of Every Model Launch",
    "## Complete List of LLM Releases in April 2026",
    "| Model | Organization | Date | Parameters | Type | Pricing (Input/Output per 1M tokens) | Key Feature | |---|---|---|---|---|---|---| | Gemini 2.5 Pro | Google | Apr 1 | Undisclosed | Proprietary | $3.50/$10.50 (under 200K) | 1M token context, native multimodal | | Claude Opus 4 | Anthropic | Apr 2 | Undisclosed | Proprietary | $15/$75 | Top coding benchmark scores, agentic | | Claude Sonnet 4 | Anthropic | Apr 2 | Undisclosed | Proprietary | $3/$15 | Balanced cost and performance | | Gemini 2.5 Flash | Google | Apr 3 | Undisclosed | Proprietary | $0.15/$0.60 | Low latency, high throughput | | Llama 4 Scout | Meta | Apr 5 | 109B (17B active) | Open source (MoE) | Free (self-host) | 10M token context window | | Llama 4 Maverick | Meta | Apr 5 | 400B (17B active) | Open source (MoE) | Free (self-host) | Multilingual, strong coding | | GPT-5 Turbo | OpenAI | Apr 7 | Undisclosed | Proprietary | $10/$30 | Native image + audio generation | | Qwen 3 (0.6B-72B) | Alibaba | Apr 8 | 0.6B to 72B | Open source | Free (Apache 2.0) | Dual-mode thinking, runs locally |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "LLM Releases April 2026: Complete Timeline of Every Model Launch",
    "## Complete List of LLM Releases in April 2026",
    "| Model | Organization | Date | Parameters | Type | Pricing (Input/Output per 1M tokens) | Key Feature | |---|---|---|---|---|---|---| | Gemini 2.5 Pro | Google | Apr 1 | Undisclosed | Proprietary | $3.50/$10.50 (under 200K) | 1M token context, native multimodal | | Claude Opus 4 | Anthropic | Apr 2 | Undisclosed | Proprietary | $15/$75 | Top coding benchmark scores, agentic | | Claude Sonnet 4 | Anthropic | Apr 2 | Undisclosed | Proprietary | $3/$15 | Balanced cost and performance | | Gemini 2.5 Flash | Google | Apr 3 | Undisclosed | Proprietary | $0.15/$0.60 | Low latency, high throughput | | Llama 4 Scout | Meta | Apr 5 | 109B (17B active) | Open source (MoE) | Free (self-host) | 10M token context window | | Llama 4 Maverick | Meta | Apr 5 | 400B (17B active) | Open source (MoE) | Free (self-host) | Multilingual, strong coding | | GPT-5 Turbo | OpenAI | Apr 7 | Undisclosed | Proprietary | $10/$30 | Native image + audio generation | | Qwen 3 (0.6B-72B) | Alibaba | Apr 8 | 0.6B to 72B | Open source | Free (Apache 2.0) | Dual-mode thinking, runs locally |"
   ]
  },
  {
   "url": "https://fazm.ai/blog/llm-releases-april-2026",
   "title": "Fazm - AI Computer Agent for macOS | Voice-Controlled Desktop Automation",
   "published_at": "2026-04-14T03:37:23",
   "organization": "Mistral",
   "model": null,
   "release_date": "2026-04-09",
   "access_type": null,
   "license": null,
   "significance": "EU AI Act compliance, multilingual",
   "organization_evidence": [
    "LLM Releases April 2026: Complete Timeline of Every Model Launch",
    "## Complete List of LLM Releases in April 2026",
    "| Model | Organization | Date | Parameters | Type | Pricing (Input/Output per 1M tokens) | Key Feature | |---|---|---|---|---|---|---| | Gemini 2.5 Pro | Google | Apr 1 | Undisclosed | Proprietary | $3.50/$10.50 (under 200K) | 1M token context, native multimodal | | Claude Opus 4 | Anthropic | Apr 2 | Undisclosed | Proprietary | $15/$75 | Top coding benchmark scores, agentic | | Claude Sonnet 4 | Anthropic | Apr 2 | Undisclosed | Proprietary | $3/$15 | Balanced cost and performance | | Gemini 2.5 Flash | Google | Apr 3 | Undisclosed | Proprietary | $0.15/$0.60 | Low latency, high throughput | | Llama 4 Scout | Meta | Apr 5 | 109B (17B active) | Open source (MoE) | Free (self-host) | 10M token context window | | Llama 4 Maverick | Meta | Apr 5 | 400B (17B active) | Open source (MoE) | Free (self-host) | Multilingual, strong coding | | GPT-5 Turbo | OpenAI | Apr 7 | Undisclosed | Proprietary | $10/$30 | Native image + audio generation | | Qwen 3 (0.6B-72B) | Alibaba | Apr 8 | 0.6B to 72B | Open source | Free (Apache 2.0) | Dual-mode thinking, runs locally | | Mistral Medium 3 | Mistral | Apr 9 | Undisclosed | Open weights | $2/$6 | EU AI Act compliance, multilingual |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "LLM Releases April 2026: Complete Timeline of Every Model Launch",
    "## Complete List of LLM Releases in April 2026",
    "| Model | Organization | Date | Parameters | Type | Pricing (Input/Output per 1M tokens) | Key Feature | |---|---|---|---|---|---|---| | Gemini 2.5 Pro | Google | Apr 1 | Undisclosed | Proprietary | $3.50/$10.50 (under 200K) | 1M token context, native multimodal | | Claude Opus 4 | Anthropic | Apr 2 | Undisclosed | Proprietary | $15/$75 | Top coding benchmark scores, agentic | | Claude Sonnet 4 | Anthropic | Apr 2 | Undisclosed | Proprietary | $3/$15 | Balanced cost and performance | | Gemini 2.5 Flash | Google | Apr 3 | Undisclosed | Proprietary | $0.15/$0.60 | Low latency, high throughput | | Llama 4 Scout | Meta | Apr 5 | 109B (17B active) | Open source (MoE) | Free (self-host) | 10M token context window | | Llama 4 Maverick | Meta | Apr 5 | 400B (17B active) | Open source (MoE) | Free (self-host) | Multilingual, strong coding | | GPT-5 Turbo | OpenAI | Apr 7 | Undisclosed | Proprietary | $10/$30 | Native image + audio generation | | Qwen 3 (0.6B-72B) | Alibaba | Apr 8 | 0.6B to 72B | Open source | Free (Apache 2.0) | Dual-mode thinking, runs locally | | Mistral Medium 3 | Mistral | Apr 9 | Undisclosed | Open weights | $2/$6 | EU AI Act compliance, multilingual |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "LLM Releases April 2026: Complete Timeline of Every Model Launch",
    "## Complete List of LLM Releases in April 2026",
    "| Model | Organization | Date | Parameters | Type | Pricing (Input/Output per 1M tokens) | Key Feature | |---|---|---|---|---|---|---| | Gemini 2.5 Pro | Google | Apr 1 | Undisclosed | Proprietary | $3.50/$10.50 (under 200K) | 1M token context, native multimodal | | Claude Opus 4 | Anthropic | Apr 2 | Undisclosed | Proprietary | $15/$75 | Top coding benchmark scores, agentic | | Claude Sonnet 4 | Anthropic | Apr 2 | Undisclosed | Proprietary | $3/$15 | Balanced cost and performance | | Gemini 2.5 Flash | Google | Apr 3 | Undisclosed | Proprietary | $0.15/$0.60 | Low latency, high throughput | | Llama 4 Scout | Meta | Apr 5 | 109B (17B active) | Open source (MoE) | Free (self-host) | 10M token context window | | Llama 4 Maverick | Meta | Apr 5 | 400B (17B active) | Open source (MoE) | Free (self-host) | Multilingual, strong coding | | GPT-5 Turbo | OpenAI | Apr 7 | Undisclosed | Proprietary | $10/$30 | Native image + audio generation | | Qwen 3 (0.6B-72B) | Alibaba | Apr 8 | 0.6B to 72B | Open source | Free (Apache 2.0) | Dual-mode thinking, runs locally | | Mistral Medium 3 | Mistral | Apr 9 | Undisclosed | Open weights | $2/$6 | EU AI Act compliance, multilingual |"
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/mimo-v2-5-pro",
   "title": "MiMo-V2.5-Pro - Intelligence, Performance & Price Analysis | Artificial Analysis",
   "published_at": null,
   "organization": "Xiaomi",
   "model": null,
   "release_date": "2026-04-22",
   "access_type": null,
   "license": null,
   "significance": "MiMo-V2.5-Pro is amongst the leading models in intelligence and reasonably priced when comparing to other open weight models of similar size.",
   "organization_evidence": [
    "MiMo-V2.5-Pro Intelligence, Performance & Price Analysis",
    "MiMo-V2.5-Pro was created by Xiaomi."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "MiMo-V2.5-Pro Intelligence, Performance & Price Analysis",
    "MiMo-V2.5-Pro was released on April 22, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "MiMo-V2.5-Pro Intelligence, Performance & Price Analysis",
    "MiMo-V2.5-Pro is amongst the leading models in intelligence and reasonably priced when comparing to other open weight models of similar size."
   ]
  },
  {
   "url": "https://vp-land.com/stories/thinking-machines-releases-inkling-its-first-open-weights-model-built-to-be-fine-tuned",
   "title": "Thinking Machines Releases Inkling, Its First Open-Weights Model Built to Be Fine-Tuned",
   "published_at": "2026-07-17T05:36:10",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "A 975 billion total parameter multimodal Mixture-of-Experts model supporting a 1 million token context window",
   "organization_evidence": [
    "# Thinking Machines Releases Inkling, Its First Open-Weights Model Built to Be Fine-Tuned",
    "Thinking Machines Lab released [Inkling](https://thinkingmachines.ai/news/introducing-inkling/?utm_source=www.vp-land.com&utm_medium=referral&utm_campaign=thinking-machines-releases-inkling-its-first-open-weights-model-built-to-be-fine-tuned), its first in-house AI model, and put the full weights up for download."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Thinking Machines Releases Inkling, Its First Open-Weights Model Built to Be Fine-Tuned",
    "The pitch behind that framing, [as TechCrunch reported](https://techcrunch.com/2026/07/15/thinking-machines-amps-up-its-bet-against-one-size-fits-all-ai-with-its-first-open-model-inkling/?utm_source=www.vp-land.com&utm_medium=referral&utm_campaign=thinking-machines-releases-inkling-its-first-open-weights-model-built-to-be-fine-tuned), is a bet that \"AI that organizations can adapt for themselves will outperform the one-size-fits-all models the biggest labs currently sell.\""
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Thinking Machines Releases Inkling, Its First Open-Weights Model Built to Be Fine-Tuned",
    " It is an open-weights, multimodal Mixture-of-Experts model that takes text, images, audio, and video as input, and the company is positioning it less as a finished product than as a base other teams retrain for their own work.",
    "**Two sizes.** Inkling runs 975 billion total parameters with 41 billion active per token. A preview variant, Inkling-Small, runs 276 billion total and 12 billion active.",
    "The model handles a context window of up to 1 million tokens and accepts multimodal input: images as 40x40 pixel patches, audio as spectrograms, and video."
   ]
  },
  {
   "url": "https://kanerika.com/blogs/mistral-vs-llama-3",
   "title": "10 Best Open Source LLMs to Evaluate in 2026",
   "published_at": "2026-06-16T13:58:04",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-16",
   "access_type": null,
   "license": null,
   "significance": "ranks #1 among open-weight models and roughly #4 overall on the Artificial Analysis Index, as a 2.8 trillion-parameter Mixture-of-Experts model",
   "organization_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "1. Kimi K3 (Moonshot AI)",
    "Moonshot AI released Kimi K3 on July 16, 2026, a 2.8 trillion-parameter Mixture-of-Experts model distributed under a Modified MIT license that allows self-hosting and commercial use."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "1. Kimi K3 (Moonshot AI)",
    "Moonshot AI released Kimi K3 on July 16, 2026, a 2.8 trillion-parameter Mixture-of-Experts model distributed under a Modified MIT license that allows self-hosting and commercial use."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "1. Kimi K3 (Moonshot AI)",
    "Moonshot AI released Kimi K3 on July 16, 2026, a 2.8 trillion-parameter Mixture-of-Experts model distributed under a Modified MIT license that allows self-hosting and commercial use.",
    "On the Artificial Analysis Index, it currently ranks #1 among open-weight models and roughly #4 overall, ahead of several closed models, making it the first Chinese model to break into that frontier tier."
   ]
  },
  {
   "url": "https://kanerika.com/blogs/mistral-vs-llama-3",
   "title": "10 Best Open Source LLMs to Evaluate in 2026",
   "published_at": "2026-06-16T13:58:04",
   "organization": "Meta",
   "model": null,
   "release_date": "2026-04-05",
   "access_type": null,
   "license": null,
   "significance": "first Llama models built on MoE architecture with a 10 million token context window and native multimodality",
   "organization_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "2. Llama 4 Scout and Maverick",
    "Meta released [Llama 4 Scout and Maverick](https://ai.meta.com/blog/llama-4-multimodal-intelligence/) on April 5, 2026, the first Llama models built on MoE architecture Meta released [Llama 4 Scout and Maverick](https://ai.meta.com/blog/llama-4-multimodal-intelligence/) on April 5, 2026, the first Llama models built on MoE architecture and trained as natively multimodal systems using early fusion, processing text, images, and video through a unified model."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "2. Llama 4 Scout and Maverick",
    "Meta released [Llama 4 Scout and Maverick](https://ai.meta.com/blog/llama-4-multimodal-intelligence/) on April 5, 2026, the first Llama models built on MoE architecture Meta released [Llama 4 Scout and Maverick](https://ai.meta.com/blog/llama-4-multimodal-intelligence/) on April 5, 2026, the first Llama models built on MoE architecture and trained as natively multimodal systems using early fusion, processing text, images, and video through a unified model."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "2. Llama 4 Scout and Maverick",
    "Meta released [Llama 4 Scout and Maverick](https://ai.meta.com/blog/llama-4-multimodal-intelligence/) on April 5, 2026, the first Llama models built on MoE architecture Meta released [Llama 4 Scout and Maverick](https://ai.meta.com/blog/llama-4-multimodal-intelligence/) on April 5, 2026, the first Llama models built on MoE architecture and trained as natively multimodal systems using early fusion, processing text, images, and video through a unified model."
   ]
  },
  {
   "url": "https://kanerika.com/blogs/mistral-vs-llama-3",
   "title": "10 Best Open Source LLMs to Evaluate in 2026",
   "published_at": "2026-06-16T13:58:04",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-08-12",
   "access_type": null,
   "license": null,
   "significance": "frontier-class open model with a 1M token context window and 1.6T total parameters",
   "organization_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "3. DeepSeek V4",
    "DeepSeek V4 comes in two sizes: [V4-Pro (1.6T total, 49B active parameters)](https://api-docs.deepseek.com/) for maximum performance, which reached general availability around August 12-13, 2026, and V4-Flash (304B total, 13B active), released July 31, 2026, as a lighter, cheaper alternative."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "3. DeepSeek V4",
    "DeepSeek V4 comes in two sizes: [V4-Pro (1.6T total, 49B active parameters)](https://api-docs.deepseek.com/) for maximum performance, which reached general availability around August 12-13, 2026, and V4-Flash (304B total, 13B active), released July 31, 2026, as a lighter, cheaper alternative."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "3. DeepSeek V4",
    "DeepSeek V4 comes in two sizes: [V4-Pro (1.6T total, 49B active parameters)](https://api-docs.deepseek.com/) for maximum performance, which reached general availability around August 12-13, 2026, and V4-Flash (304B total, 13B active), released July 31, 2026, as a lighter, cheaper alternative.",
    "Both expose a 1M token context window and carry MIT licensing for full self-hosting."
   ]
  },
  {
   "url": "https://kanerika.com/blogs/mistral-vs-llama-3",
   "title": "10 Best Open Source LLMs to Evaluate in 2026",
   "published_at": "2026-06-16T13:58:04",
   "organization": "Alibaba Cloud",
   "model": null,
   "release_date": "2026-02",
   "access_type": null,
   "license": null,
   "significance": "flagship open-weight model supporting 201 languages with native multimodality",
   "organization_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "4. Qwen3.6-35B-A3B"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "4. Qwen3.6-35B-A3B"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "4. Qwen3.6-35B-A3B"
   ]
  },
  {
   "url": "https://kanerika.com/blogs/mistral-vs-llama-3",
   "title": "10 Best Open Source LLMs to Evaluate in 2026",
   "published_at": "2026-06-16T13:58:04",
   "organization": "Google DeepMind",
   "model": null,
   "release_date": "2026-04-02",
   "access_type": null,
   "license": null,
   "significance": "built from Gemini 3 research in four sizes covering edge through server deployment",
   "organization_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "5. Gemma 4",
    "Google DeepMind released [Gemma 4](https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/) on April 2, 2026 under Apache 2.0 with full commercial freedom."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "5. Gemma 4",
    "Google DeepMind released [Gemma 4](https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/) on April 2, 2026 under Apache 2.0 with full commercial freedom."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "5. Gemma 4",
    "Google DeepMind released [Gemma 4](https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/) on April 2, 2026 under Apache 2.0 with full commercial freedom.",
    "Built from Gemini 3 research, it comes in four sizes: E2B and E4B for edge and mobile, 26B MoE, and 31B Dense for server workloads."
   ]
  },
  {
   "url": "https://kanerika.com/blogs/mistral-vs-llama-3",
   "title": "10 Best Open Source LLMs to Evaluate in 2026",
   "published_at": "2026-06-16T13:58:04",
   "organization": "Mistral AI",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "optimized for low latency and low hardware requirements",
   "organization_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "7. Mistral Small 4",
    "Mistral Small 4\n\n[Mist](https://mistral.ai/news/mistral-small-3-1)[Mistral Small 4](https://mistral.ai/news/mistral-small-3-1) is built for real-time applications where response speed and low hardware requirements take priority."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "7. Mistral Small 4",
    "Mistral Small 4\n\n[Mist](https://mistral.ai/news/mistral-small-3-1)[Mistral Small 4](https://mistral.ai/news/mistral-small-3-1) is built for real-time applications where response speed and low hardware requirements take priority."
   ]
  },
  {
   "url": "https://kanerika.com/blogs/mistral-vs-llama-3",
   "title": "10 Best Open Source LLMs to Evaluate in 2026",
   "published_at": "2026-06-16T13:58:04",
   "organization": "Zhipu AI",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "MoE model with 744B total and 40B active parameters designed for long-horizon agentic engineering tasks",
   "organization_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "9. GLM-5.2 (Zhipu AI)"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "9. GLM-5.2 (Zhipu AI)"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "9. GLM-5.2 (Zhipu AI)",
    "With 744B total and 40B active parameters and a 200K context window, it is MIT licensed and has posted some of the strongest SWE-bench Pro scores in the open-weight category."
   ]
  },
  {
   "url": "https://kanerika.com/blogs/mistral-vs-llama-3",
   "title": "10 Best Open Source LLMs to Evaluate in 2026",
   "published_at": "2026-06-16T13:58:04",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "combines a 1M token context window, native multimodality, and frontier coding ability",
   "organization_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "10. MiniMax-M3",
    "[MiniMax M3](https://huggingface.co/MiniMaxAI/MiniMax-M3), released June 1, 2026, is the most capable open-weight model MiniMax has shipped."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "10. MiniMax-M3",
    "[MiniMax M3](https://huggingface.co/MiniMaxAI/MiniMax-M3), released June 1, 2026, is the most capable open-weight model MiniMax has shipped."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "10. MiniMax-M3",
    "[MiniMax M3](https://huggingface.co/MiniMaxAI/MiniMax-M3), released June 1, 2026, is the most capable open-weight model MiniMax has shipped.",
    "It combines a 1M token context window, native multimodality, and frontier coding ability in a single model, topping open-weight SWE-Bench Pro at 59.0%."
   ]
  },
  {
   "url": "https://huggingface.co/blog/mlabonne/minimax-m25",
   "title": "MiniMax-M2.5: The $1/hour Frontier Model",
   "published_at": "2026-02-17T12:58:27",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-02-12",
   "access_type": null,
   "license": null,
   "significance": "A 230B MoE model with 10B active parameters trained through reinforcement learning across 200,000+ environments, achieving strong performance on coding and office productivity benchmarks.",
   "organization_evidence": [
    "On February 12th, 2026, barely a month after its Hong Kong IPO, Shanghai-based MiniMax dropped M2.5.",
    "MiniMax-M2.5 is an iterative improvement on the M2 family, which launched in late October 2025."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "On February 12th, 2026, barely a month after its Hong Kong IPO, Shanghai-based MiniMax dropped M2.5."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The headline numbers: **80.2% SWE-Bench Verified** , **51.3% Multi-SWE-Bench** (first place), **76.3% BrowseComp**.",
    "The striking part of this release is the combination: this is a 230B MoE model with only 10B active parameters, trained primarily through large-scale reinforcement learning across 200,000+ real-world environments."
   ]
  },
  {
   "url": "https://techcrunch.com/2026/07/18/kimi-threat-or-menace",
   "title": "Kimi: Threat or menace?",
   "published_at": "2026-07-18T18:51:07",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "frontier-level performance across evaluation suite, outperforming other tested models",
   "organization_evidence": [
    "Chinese company Moonshot AI released a new version of its Kimi model this week, leading to a perhaps-inevitable wave of discourse about China and open source AI."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Chinese company Moonshot AI released a new version of its Kimi model this week, leading to a perhaps-inevitable wave of discourse about China and open source AI.",
    "[Moonshot said](https://www.kimi.com/blog/kimi-k3) that although Kimi K3 \u201cstill trails the most powerful proprietary models, Claude Fable 5 and GPT 5.6 Sol,\u201d the new open source model \u201cdemonstrated frontier-level performance across our evaluation suite, consistently outperforming other tested models.\u201d"
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/gpt-5-6-luna-medium-vs-glm-5-2",
   "title": "GPT-5.6 Luna (medium) vs GLM-5.2 (max): Model Comparison | Artificial Analysis",
   "published_at": null,
   "organization": "Z AI",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "More intelligent than GPT-5.6 Luna (medium) with a score of 53 on the Artificial Analysis Intelligence Index.",
   "organization_evidence": [
    "# GPT-5.6 Luna (medium) vs. GLM-5.2 (max)",
    "| | ![OpenAI logo](/img/logos/openai_small.svg)OpenAI GPT-5.6 Luna (medium) | ![Z AI logo](/img/logos/zai_small.svg)Z AI GLM-5.2 (max) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# GPT-5.6 Luna (medium) vs. GLM-5.2 (max)",
    "| Release Date | July 2026 | June 2026 | GPT-5.6 Luna (medium) has a more recent release date than GLM-5.2 (max) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# GPT-5.6 Luna (medium) vs. GLM-5.2 (max)",
    "| Intelligence Index | 39 | 53 | GLM-5.2 (max) is more intelligent than GPT-5.6 Luna (medium) |",
    "GLM-5.2 (max) is more intelligent. GLM-5.2 (max) scores 53, compared with GPT-5.6 Luna (medium) at 39 on the Artificial Analysis Intelligence Index."
   ]
  },
  {
   "url": "https://tftc.io/inkling-open-weights-ai-model-thinking-machines-murati",
   "title": "Murati's 975B Open-Weights AI Model Challenges Closed Labs \u00b7 TFTC",
   "published_at": "2026-07-16T03:16:15",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "a 975 billion parameter open-weights model under Apache 2.0, the largest from a Western lab",
   "organization_evidence": [
    "# Mira Murati Releases 975B Open-Weights Frontier AI Model",
    "Thinking Machines Lab released Inkling on July 15, 2026: a 975 billion parameter mixture-of-experts model under an Apache 2.0 license, free to download, fine-tune, and redeploy.",
    "Thinking Machines Lab, founded by former OpenAI CTO Mira Murati in February 2025, released [Inkling](https://thinkingmachines.ai/news/introducing-inkling/?ref=tftc.ghost.io) on July 15, 2026."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Mira Murati Releases 975B Open-Weights Frontier AI Model",
    "Former OpenAI CTO Mira Murati released Inkling on July 15, 2026: a 975 billion parameter open-weights model under Apache 2.0, the largest from a Western lab, positioned as customizable over raw capability.",
    "Thinking Machines Lab released Inkling on July 15, 2026: a 975 billion parameter mixture-of-experts model under an Apache 2.0 license, free to download, fine-tune, and redeploy.",
    "Thinking Machines Lab, founded by former OpenAI CTO Mira Murati in February 2025, released [Inkling](https://thinkingmachines.ai/news/introducing-inkling/?ref=tftc.ghost.io) on July 15, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Mira Murati Releases 975B Open-Weights Frontier AI Model",
    "Former OpenAI CTO Mira Murati released Inkling on July 15, 2026: a 975 billion parameter open-weights model under Apache 2.0, the largest from a Western lab, positioned as customizable over raw capability."
   ]
  },
  {
   "url": "https://adtools.org/buyers-guide/the-best-open-source-ai-models-in-2026-an-expert-comparison",
   "title": "The Best Open-Source AI Models in 2026: An Expert Comparison",
   "published_at": null,
   "organization": "Google",
   "model": null,
   "release_date": "2026-04-02",
   "access_type": null,
   "license": null,
   "significance": "open models (E2B, E4B, 26B MoE, 31B Dense) built on Gemini 3 technology",
   "organization_evidence": [
    "# The Best Open-Source AI Models in 2026: An Expert ComparisonUpdated: August 02, 2026",
    "Connectors genai @CGenai25884 Apr 3, 2026",
    "Google released Gemma 4 on April 2, 2026 \u2014 open models (E2B, E4B, 26B MoE, 31B Dense) built on Gemini 3 technology."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# The Best Open-Source AI Models in 2026: An Expert ComparisonUpdated: August 02, 2026",
    "Connectors genai @CGenai25884 Apr 3, 2026",
    "Google released Gemma 4 on April 2, 2026 \u2014 open models (E2B, E4B, 26B MoE, 31B Dense) built on Gemini 3 technology."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# The Best Open-Source AI Models in 2026: An Expert ComparisonUpdated: August 02, 2026",
    "Connectors genai @CGenai25884 Apr 3, 2026",
    "Google released Gemma 4 on April 2, 2026 \u2014 open models (E2B, E4B, 26B MoE, 31B Dense) built on Gemini 3 technology."
   ]
  },
  {
   "url": "https://adtools.org/buyers-guide/the-best-open-source-ai-models-in-2026-an-expert-comparison",
   "title": "The Best Open-Source AI Models in 2026: An Expert Comparison",
   "published_at": null,
   "organization": "AI2",
   "model": null,
   "release_date": "2026-04-15",
   "access_type": null,
   "license": null,
   "significance": "32-billion parameter instruction-following model offers developers complete research transparency, including full access to training data, weights, and code under a permissive Apache 2.0 license",
   "organization_evidence": [
    "# The Best Open-Source AI Models in 2026: An Expert ComparisonUpdated: August 02, 2026",
    "aichina.news @AiChinaNews Apr 15, 2026",
    "The fully open-source OLMo-3.1-32B-Instruct model from AI2 has been optimized for Huawei's Ascend NPU architecture and released on http://Modelers.cn"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# The Best Open-Source AI Models in 2026: An Expert ComparisonUpdated: August 02, 2026",
    "aichina.news @AiChinaNews Apr 15, 2026",
    "The fully open-source OLMo-3.1-32B-Instruct model from AI2 has been optimized for Huawei's Ascend NPU architecture and released on http://Modelers.cn"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# The Best Open-Source AI Models in 2026: An Expert ComparisonUpdated: August 02, 2026",
    "aichina.news @AiChinaNews Apr 15, 2026",
    "This 32-billion parameter instruction-following model offers developers complete research transparency, including full access to training data, weights, and code under a permissive Apache 2.0 license."
   ]
  },
  {
   "url": "https://braindetox.kr/en/posts/thinking_machines_inkling_model_2026.html",
   "title": "Inkling: Thinking Machines Lab's 975B Open-Weight MoE Model, Explained",
   "published_at": "2026-07-17T00:00:00",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "Mixture-of-Experts transformer with 975B total parameters and 41B active per token, supporting 1M token context",
   "organization_evidence": [
    "# Inkling: Thinking Machines Lab's 975B Open-Weight MoE Model, Explained",
    "On July 15, 2026, **Thinking Machines Lab** \u2014 the startup former OpenAI CTO Mira Murati founded in February 2025 \u2014 released its first production language model, **Inkling**."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "On July 15, 2026, **Thinking Machines Lab** \u2014 the startup former OpenAI CTO Mira Murati founded in February 2025 \u2014 released its first production language model, **Inkling**."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Inkling: Thinking Machines Lab's 975B Open-Weight MoE Model, Explained",
    "First, it is a **Mixture-of-Experts (MoE)** transformer with **975B** total parameters but only **41B** active per token."
   ]
  },
  {
   "url": "https://bitsminds.com/news/thinking-machines-inkling-open-weights-model-2026",
   "title": "Mira Murati's Thinking Machines Releases Inkling \u2014 a 975B Open-Weight, Multimodal Model Built to Be Customized",
   "published_at": "2026-07-17T14:25:11",
   "organization": "Thinking Machines Lab",
   "model": "Inkling",
   "release_date": "2026",
   "access_type": "open-weights",
   "license": "Apache 2.0",
   "significance": "975B-parameter mixture-of-experts model with a 1M-token context trained on 45T tokens of text, image, audio, and video",
   "organization_evidence": [
    "# Mira Murati's Thinking Machines Releases Inkling \u2014 a 975B Open-Weight, Multimodal Model Built to Be Customized",
    "Thinking Machines Lab's first broadly available model is open."
   ],
   "model_evidence": [
    "# Mira Murati's Thinking Machines Releases Inkling \u2014 a 975B Open-Weight, Multimodal Model Built to Be Customized",
    "Inkling is a 975B-parameter mixture-of-experts model (41B active), trained on 45T tokens of text, image, audio, and video, with a 1M-token context \u2014 released under Apache 2.0 and tuned on the lab's Tinker platform."
   ],
   "release_date_evidence": [
    "# Mira Murati's Thinking Machines Releases Inkling \u2014 a 975B Open-Weight, Multimodal Model Built to Be Customized",
    "Inkling scored 97.1% on the 2026 AIME math competition and 87.2% on GPQA Diamond, a test of graduate-level scientific reasoning, alongside 77.6% on SWE-bench Verified and 54.3% on SWE-bench Pro Public for real-world coding."
   ],
   "access_type_evidence": [
    "# Mira Murati's Thinking Machines Releases Inkling \u2014 a 975B Open-Weight, Multimodal Model Built to Be Customized"
   ],
   "license_evidence": [
    "# Mira Murati's Thinking Machines Releases Inkling \u2014 a 975B Open-Weight, Multimodal Model Built to Be Customized",
    "Inkling is a 975B-parameter mixture-of-experts model (41B active), trained on 45T tokens of text, image, audio, and video, with a 1M-token context \u2014 released under Apache 2.0 and tuned on the lab's Tinker platform."
   ],
   "significance_evidence": [
    "# Mira Murati's Thinking Machines Releases Inkling \u2014 a 975B Open-Weight, Multimodal Model Built to Be Customized",
    "Inkling is a 975B-parameter mixture-of-experts model (41B active), trained on 45T tokens of text, image, audio, and video, with a 1M-token context \u2014 released under Apache 2.0 and tuned on the lab's Tinker platform."
   ]
  },
  {
   "url": "https://huggingface.co/blog/open-r1/update-4",
   "title": "Open R1: Update #4",
   "published_at": "2025-03-26T18:47:29",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-03-24",
   "access_type": null,
   "license": null,
   "significance": "significant improvements across benchmarks such as MMLU-Pro, GPQA, AIME, and LiveCodeBench with targeted improvements in front-end web development, Chinese writing proficiency, Chinese search capabilities, and function calling",
   "organization_evidence": [
    "# Open R1: Update #4",
    "This week, a [new model](https://huggingface.co/deepseek-ai/DeepSeek-V3-0324) from DeepSeek silently landed on the Hub."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Open R1: Update #4",
    "## Welcome DeepSeek-V3 **0324**",
    "This week, a [new model](https://huggingface.co/deepseek-ai/DeepSeek-V3-0324) from DeepSeek silently landed on the Hub."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Open R1: Update #4",
    "## What we know so far",
    "To summarise the model has seen significant improvements across benchmarks",
    "Specifically, in the model card the DeepSeek mentions targeted improvements in the following areas:"
   ]
  },
  {
   "url": "https://radarai.top/en/china-ai-open-source-models",
   "title": "China Open Source AI Models (Apache 2.0 / MIT, 2026 List)",
   "published_at": null,
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "most capable open-weight model in either Chinese or Western labs as of May 2026 on MMLU (88.7) and MATH (79.4)",
   "organization_evidence": [
    "Open source Chinese AI models \u2014 full list (2026)",
    "| Model | Lab | Release date | License | Parameters | Download | Commercial use restrictions |",
    "| **Qwen3** (0.6B\u2013235B) | Alibaba | April 2026 | Apache 2.0 | 0.6B / 1.7B / 4B / 8B / 14B / 30B-A3B / 32B / 235B-A22B | [huggingface.co/Qwen](https://huggingface.co/Qwen) | None (attribution only) |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Open source Chinese AI models \u2014 full list (2026)",
    "| Model | Lab | Release date | License | Parameters | Download | Commercial use restrictions |",
    "| **Qwen3** (0.6B\u2013235B) | Alibaba | April 2026 | Apache 2.0 | 0.6B / 1.7B / 4B / 8B / 14B / 30B-A3B / 32B / 235B-A22B | [huggingface.co/Qwen](https://huggingface.co/Qwen) | None (attribution only) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Open source Chinese AI models \u2014 full list (2026)",
    "Qwen3-235B-A22B (April 2026) is the most capable open-weight model in either Chinese or Western labs as of May 2026 on MMLU (88.7) and MATH (79.4)."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/qwen3-6-plus-vs-glm-5",
   "title": "Qwen3.6 Plus vs GLM-5 (Reasoning): Model Comparison | Artificial Analysis",
   "published_at": null,
   "organization": "Z AI",
   "model": null,
   "release_date": "2026-02",
   "access_type": null,
   "license": null,
   "significance": "Reasoning model scoring 41 on Artificial Analysis Intelligence Index",
   "organization_evidence": [
    "# Qwen3.6 Plus vs. GLM-5 (Reasoning)",
    "## Model Comparison",
    "| | ![Alibaba logo](/img/logos/alibaba_small.svg)Alibaba Qwen3.6 Plus | ![Z AI logo](/img/logos/zai_small.svg)Z AI GLM-5 (Reasoning) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Qwen3.6 Plus vs. GLM-5 (Reasoning)",
    "## Model Comparison",
    "| Release Date | April 2026 | February 2026 | Qwen3.6 Plus has a more recent release date than GLM-5 (Reasoning) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Qwen3.6 Plus vs. GLM-5 (Reasoning)",
    "## Model Comparison",
    "| Intelligence Index | 40 | 41* | GLM-5 (Reasoning) is more intelligent than Qwen3.6 Plus |"
   ]
  },
  {
   "url": "https://kwatsjpedia.org/news-and-trends/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - Kwatsjpedia",
   "published_at": "2026-07-15T11:07:01",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor",
   "organization_evidence": [
    "The production line \u2014 spring 2026",
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The production line \u2014 spring 2026",
    "During this period, Chinese labs introduced four models: DeepSeek V4 on April 24, MiniMax M3 on June 1, and Kimi K2.7-Code and GLM-5.2 in mid-June."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The production line \u2014 spring 2026",
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor"
   ]
  },
  {
   "url": "https://kwatsjpedia.org/news-and-trends/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - Kwatsjpedia",
   "published_at": "2026-07-15T11:07:01",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "cheap 1M-token context, native multimodal, modified-MIT",
   "organization_evidence": [
    "The production line \u2014 spring 2026"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The production line \u2014 spring 2026",
    "During this period, Chinese labs introduced four models: DeepSeek V4 on April 24, MiniMax M3 on June 1, and Kimi K2.7-Code and GLM-5.2 in mid-June."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The production line \u2014 spring 2026"
   ]
  },
  {
   "url": "https://kwatsjpedia.org/news-and-trends/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - Kwatsjpedia",
   "published_at": "2026-07-15T11:07:01",
   "organization": "Moonshot",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "agent-run specialist, ~30% fewer thinking tokens than K2.6",
   "organization_evidence": [
    "The production line \u2014 spring 2026"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The production line \u2014 spring 2026",
    "During this period, Chinese labs introduced four models: DeepSeek V4 on April 24, MiniMax M3 on June 1, and Kimi K2.7-Code and GLM-5.2 in mid-June."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The production line \u2014 spring 2026"
   ]
  },
  {
   "url": "https://kwatsjpedia.org/news-and-trends/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - Kwatsjpedia",
   "published_at": "2026-07-15T11:07:01",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "753B MoE, MIT, top open-weight on Artificial Analysis index",
   "organization_evidence": [
    "The production line \u2014 spring 2026"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The production line \u2014 spring 2026",
    "During this period, Chinese labs introduced four models: DeepSeek V4 on April 24, MiniMax M3 on June 1, and Kimi K2.7-Code and GLM-5.2 in mid-June."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The production line \u2014 spring 2026"
   ]
  },
  {
   "url": "https://ecorpit.com/thinking-machines-inkling-open-weight-model-decision-2026",
   "title": "Inkling, Thinking Machines' 975B open-weights model: adopt it or wait in 2026?",
   "published_at": "2026-07-19T00:00:00",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "Mixture-of-Experts transformer with 975 billion total parameters and 41 billion active per token, multimodal capability, and 1 million token context window",
   "organization_evidence": [
    "Inkling, Thinking Machines' 975B open-weights model: adopt it or wait in 2026?",
    "**Summary.** Thinking Machines Lab, the company founded by former OpenAI CTO Mira Murati, released Inkling on 15 July 2026, its first model and its first open-weights release."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Inkling, Thinking Machines' 975B open-weights model: adopt it or wait in 2026?",
    "**Summary.** Thinking Machines Lab, the company founded by former OpenAI CTO Mira Murati, released Inkling on 15 July 2026, its first model and its first open-weights release."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Inkling, Thinking Machines' 975B open-weights model: adopt it or wait in 2026?",
    "**Summary.** Thinking Machines Lab, the company founded by former OpenAI CTO Mira Murati, released Inkling on 15 July 2026, its first model and its first open-weights release.",
    "Inkling is a Mixture-of-Experts transformer with 975 billion total parameters and 41 billion active per token, a context window up to 1 million tokens, pretrained on 45 trillion tokens of text, images, audio and video."
   ]
  },
  {
   "url": "https://userightai.com/new-ai-models-2026",
   "title": "New AI Models Released in 2026 \u2014 Full Ranked Timeline | UseRightAI",
   "published_at": null,
   "organization": "Mistral",
   "model": null,
   "release_date": "2026-05-12",
   "access_type": null,
   "license": null,
   "significance": "Europe's strongest open release of 2026.",
   "organization_evidence": [
    "## Every release. *One card each.*",
    "2026-05-12**14D AGO** \u00b7 RELEASE",
    "Mistral Medium 3.1",
    "MISTRALBALANCED"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## Every release. *One card each.*",
    "2026-05-12**14D AGO** \u00b7 RELEASE",
    "Mistral Medium 3.1"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## Every release. *One card each.*",
    "2026-05-12**14D AGO** \u00b7 RELEASE",
    "Mistral Medium 3.1",
    "Europe's strongest open release of 2026. A clean middle option for teams that need a non-US model."
   ]
  },
  {
   "url": "https://userightai.com/new-ai-models-2026",
   "title": "New AI Models Released in 2026 \u2014 Full Ranked Timeline | UseRightAI",
   "published_at": null,
   "organization": "Meta",
   "model": null,
   "release_date": "2026-02-04",
   "access_type": null,
   "license": null,
   "significance": "10M context window \u2014 by far the largest",
   "organization_evidence": [
    "## Every release. *One card each.*",
    "2026-02-04**111D AGO** \u00b7 RELEASE",
    "Llama 4 Scout",
    "METABUDGET"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## Every release. *One card each.*",
    "2026-02-04**111D AGO** \u00b7 RELEASE",
    "Llama 4 Scout"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## Every release. *One card each.*",
    "2026-02-04**111D AGO** \u00b7 RELEASE",
    "Llama 4 Scout"
   ]
  },
  {
   "url": "https://userightai.com/new-ai-models-2026",
   "title": "New AI Models Released in 2026 \u2014 Full Ranked Timeline | UseRightAI",
   "published_at": null,
   "organization": "Meta",
   "model": null,
   "release_date": "2026-02-04",
   "access_type": null,
   "license": null,
   "significance": "Biggest open-weight leap of 2026.",
   "organization_evidence": [
    "## Every release. *One card each.*",
    "2026-02-04**111D AGO** \u00b7 RELEASE",
    "Llama 4 Maverick",
    "METABUDGET"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## Every release. *One card each.*",
    "2026-02-04**111D AGO** \u00b7 RELEASE",
    "Llama 4 Maverick"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## Every release. *One card each.*",
    "2026-02-04**111D AGO** \u00b7 RELEASE",
    "Llama 4 Maverick",
    "Biggest open-weight leap of 2026. Competitive with GPT-5.4 on general tasks at a quarter of the price."
   ]
  },
  {
   "url": "https://userightai.com/new-ai-models-2026",
   "title": "New AI Models Released in 2026 \u2014 Full Ranked Timeline | UseRightAI",
   "published_at": null,
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-01-22",
   "access_type": null,
   "license": null,
   "significance": "Open-weights, $0.27/M input, beats GPT-4o on coding.",
   "organization_evidence": [
    "## Every release. *One card each.*",
    "2026-01-22**124D AGO** \u00b7 RELEASE",
    "DeepSeek V4",
    "DEEPSEEKOPEN-WEIGHTS"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## Every release. *One card each.*",
    "2026-01-22**124D AGO** \u00b7 RELEASE",
    "DeepSeek V4"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## Every release. *One card each.*",
    "2026-01-22**124D AGO** \u00b7 RELEASE",
    "DeepSeek V4",
    "DEEPSEEKOPEN-WEIGHTS",
    "Open-weights, $0.27/M input, beats GPT-4o on coding. Quietly the most disruptive release of January."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/step-3-7-flash",
   "title": "Step 3.7 Flash - Intelligence, Performance & Price Analysis | Artificial Analysis",
   "published_at": null,
   "organization": "StepFun",
   "model": null,
   "release_date": "2026-05-29",
   "access_type": null,
   "license": null,
   "significance": "198B parameter Mixture of Experts model with a 262k context window and multimodal capabilities",
   "organization_evidence": [
    "Step 3.7 Flash Intelligence, Performance & Price Analysis",
    "Step 3.7 Flash was created by StepFun."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Step 3.7 Flash Intelligence, Performance & Price Analysis",
    "Released May 2026",
    "Step 3.7 Flash was released on May 29, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Step 3.7 Flash Intelligence, Performance & Price Analysis",
    "The model supports text and image input, outputs text, and has a 262k tokens context window.",
    "| Total parameters | 198B |",
    "| Active parameters | 11B Number of parameters active per token during inference |",
    "Step 3.7 Flash is a Mixture of Experts (MoE) model with 198 billion total parameters, but only 11 billion active parameters are used during inference."
   ]
  },
  {
   "url": "https://fourweekmba.com/ai-thinking-machines-inkling-open-weights-customization-strateg",
   "title": "FourWeekMBA",
   "published_at": "2026-07-15T18:41:19",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "975B-parameter mixture-of-experts foundation model with full open weights, 1M token context window, and 45T pretraining tokens",
   "organization_evidence": [
    "Thinking Machines Lab Releases Inkling: An Open-Weights Foundation Model Built for Customization, Not Leaderboard Dominance",
    "Thinking Machines Lab \u2014 founded by Mira Murati, who served as CTO of OpenAI until late 2024 \u2014 released its first [model](https://fourweekmba.com/what-is-a-business-model/) on July 15, 2026."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Thinking Machines Lab \u2014 founded by Mira Murati, who served as CTO of OpenAI until late 2024 \u2014 released its first [model](https://fourweekmba.com/what-is-a-business-model/) on July 15, 2026.",
    "July 15, 2026"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Mira Murati\u2019s new lab ships a 975B-parameter mixture-of-experts [model](https://fourweekmba.com/what-is-a-business-model/) with full open weights and a fine-tuning platform \u2014 and says outright it is not the strongest [model](https://fourweekmba.com/what-is-a-business-model/) available.",
    "Thinking Machines Lab \u2014 founded by Mira Murati, who served as CTO of OpenAI until late 2024 \u2014 released its first [model](https://fourweekmba.com/what-is-a-business-model/) on July 15, 2026.",
    "Inkling is a mixture-of-experts foundation [model](https://fourweekmba.com/what-is-a-business-model/) carrying 975 billion total parameters, with 41 billion active per forward pass.",
    "It was pretrained on 45 trillion tokens spanning text, images, audio, and video, supports a one-million-token context window, and ships alongside Inkling-Small, currently in preview."
   ]
  },
  {
   "url": "https://omidsaffari.com/blog/best-open-source-llms-2026",
   "title": "The Best Open-Source LLMs in 2026: 8 Open-Weight Models Ranked",
   "published_at": "2026-06-29T02:13:27",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-06",
   "access_type": null,
   "license": null,
   "significance": "strongest open-weight model in the world for long-horizon engineering work with a 1M-token context window",
   "organization_evidence": [
    "The Best Open-Source LLMs in 2026: 8 Open-Weight Models Ranked",
    "GLM-5.2 is the open model to beat",
    "GLM-5.2, from the Chinese lab Z.ai (formerly Zhipu), is the strongest open-weight model in the world for long-horizon engineering work, and it is the one to reach for first."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The Best Open-Source LLMs in 2026: 8 Open-Weight Models Ranked",
    "GLM-5.2 is the open model to beat",
    "It shipped in mid-June 2026 with a genuinely usable 1M-token context window, up from 200K in the previous version, and it holds quality across that window rather than just accepting the tokens."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The Best Open-Source LLMs in 2026: 8 Open-Weight Models Ranked",
    "GLM-5.2 is the open model to beat",
    "GLM-5.2, from the Chinese lab Z.ai (formerly Zhipu), is the strongest open-weight model in the world for long-horizon engineering work, and it is the one to reach for first."
   ]
  },
  {
   "url": "https://omidsaffari.com/blog/best-open-source-llms-2026",
   "title": "The Best Open-Source LLMs in 2026: 8 Open-Weight Models Ranked",
   "published_at": "2026-06-29T02:13:27",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "near-frontier quality mixture-of-experts model at low cost with a 1M-token context",
   "organization_evidence": [
    "The Best Open-Source LLMs in 2026: 8 Open-Weight Models Ranked",
    "DeepSeek V4 is the cost floor nobody else can match",
    "Released in April 2026 by the Chinese lab DeepSeek, it comes in two sizes, both MIT-licensed with a 1M-token context."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The Best Open-Source LLMs in 2026: 8 Open-Weight Models Ranked",
    "DeepSeek V4 is the cost floor nobody else can match",
    "Released in April 2026 by the Chinese lab DeepSeek, it comes in two sizes, both MIT-licensed with a 1M-token context."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The Best Open-Source LLMs in 2026: 8 Open-Weight Models Ranked",
    "DeepSeek V4 is the cost floor nobody else can match",
    "DeepSeek V4 is the model that makes the math impossible to ignore, because it delivers near-frontier quality at a price that rounds to nothing."
   ]
  },
  {
   "url": "https://omidsaffari.com/blog/best-open-source-llms-2026",
   "title": "The Best Open-Source LLMs in 2026: 8 Open-Weight Models Ranked",
   "published_at": "2026-06-29T02:13:27",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-06",
   "access_type": null,
   "license": null,
   "significance": "purpose-built for coding agents that run for hours with native multimodal input and a 256K-token context window",
   "organization_evidence": [
    "The Best Open-Source LLMs in 2026: 8 Open-Weight Models Ranked",
    "Kimi K2.7 Code is built for agents that write code",
    "Kimi K2.7 Code, from Moonshot AI, is the open model purpose-built for coding agents that run for hours, and it landed in June 2026."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The Best Open-Source LLMs in 2026: 8 Open-Weight Models Ranked",
    "Kimi K2.7 Code is built for agents that write code",
    "Kimi K2.7 Code, from Moonshot AI, is the open model purpose-built for coding agents that run for hours, and it landed in June 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The Best Open-Source LLMs in 2026: 8 Open-Weight Models Ranked",
    "Kimi K2.7 Code is built for agents that write code",
    "Kimi K2.7 Code, from Moonshot AI, is the open model purpose-built for coding agents that run for hours, and it landed in June 2026."
   ]
  },
  {
   "url": "https://omidsaffari.com/blog/best-open-source-llms-2026",
   "title": "The Best Open-Source LLMs in 2026: 8 Open-Weight Models Ranked",
   "published_at": "2026-06-29T02:13:27",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "428-billion-parameter mixture-of-experts model with 1M-token context and native multimodal input",
   "organization_evidence": [
    "The Best Open-Source LLMs in 2026: 8 Open-Weight Models Ranked",
    "MiniMax M3 is the open model to watch when your work mixes text, images, and video and needs a huge context, and it is the freshest of the bunch, released on June 1, 2026."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The Best Open-Source LLMs in 2026: 8 Open-Weight Models Ranked",
    "MiniMax M3 is the open model to watch when your work mixes text, images, and video and needs a huge context, and it is the freshest of the bunch, released on June 1, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The Best Open-Source LLMs in 2026: 8 Open-Weight Models Ranked",
    "MiniMax M3 is the open model to watch when your work mixes text, images, and video and needs a huge context, and it is the freshest of the bunch, released on June 1, 2026."
   ]
  },
  {
   "url": "https://techcrunch.com/2026/05/07/chinas-moonshot-ai-raises-2b-at-20b-valuation-as-demand-for-open-source-ai-skyrockets",
   "title": "China\u2019s Moonshot AI raises $2B at $20B valuation as demand for open source AI skyrockets",
   "published_at": "2026-05-07T13:44:25",
   "organization": "Moonshot AI",
   "model": "Kimi K2.5",
   "release_date": "2026-01-27",
   "access_type": "open-source",
   "license": null,
   "significance": "took the coding world by storm earlier this year, nearly topping benchmarks and posting performance figures close to that of Open AI and Anthropic\u2019s models at the time",
   "organization_evidence": [
    "Moonshot AI was founded in 2023 by Yang Zhilin, a former Meta AI and Google Brain researcher, and quickly became one of China\u2019s most popular AI labs after its [open-weight Kimi K2.5](https://techcrunch.com/2026/01/27/chinas-moonshot-releases-a-new-open-source-model-kimi-k2-5-and-a-coding-agent/) large language model took the coding world by storm earlier this year, nearly topping benchmarks and posting performance figures close to that of Open AI and Anthropic\u2019s models at the time."
   ],
   "model_evidence": [
    "Moonshot AI was founded in 2023 by Yang Zhilin, a former Meta AI and Google Brain researcher, and quickly became one of China\u2019s most popular AI labs after its [open-weight Kimi K2.5](https://techcrunch.com/2026/01/27/chinas-moonshot-releases-a-new-open-source-model-kimi-k2-5-and-a-coding-agent/) large language model took the coding world by storm earlier this year, nearly topping benchmarks and posting performance figures close to that of Open AI and Anthropic\u2019s models at the time."
   ],
   "release_date_evidence": [
    "Moonshot AI was founded in 2023 by Yang Zhilin, a former Meta AI and Google Brain researcher, and quickly became one of China\u2019s most popular AI labs after its [open-weight Kimi K2.5](https://techcrunch.com/2026/01/27/chinas-moonshot-releases-a-new-open-source-model-kimi-k2-5-and-a-coding-agent/) large language model took the coding world by storm earlier this year, nearly topping benchmarks and posting performance figures close to that of Open AI and Anthropic\u2019s models at the time."
   ],
   "access_type_evidence": [
    "Moonshot AI was founded in 2023 by Yang Zhilin, a former Meta AI and Google Brain researcher, and quickly became one of China\u2019s most popular AI labs after its [open-weight Kimi K2.5](https://techcrunch.com/2026/01/27/chinas-moonshot-releases-a-new-open-source-model-kimi-k2-5-and-a-coding-agent/) large language model took the coding world by storm earlier this year, nearly topping benchmarks and posting performance figures close to that of Open AI and Anthropic\u2019s models at the time."
   ],
   "license_evidence": null,
   "significance_evidence": [
    "Moonshot AI was founded in 2023 by Yang Zhilin, a former Meta AI and Google Brain researcher, and quickly became one of China\u2019s most popular AI labs after its [open-weight Kimi K2.5](https://techcrunch.com/2026/01/27/chinas-moonshot-releases-a-new-open-source-model-kimi-k2-5-and-a-coding-agent/) large language model took the coding world by storm earlier this year, nearly topping benchmarks and posting performance figures close to that of Open AI and Anthropic\u2019s models at the time."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/glm-5-1-vs-nvidia-nemotron-3-super-120b-a12b",
   "title": "GLM-5.1 (Reasoning) vs NVIDIA Nemotron 3 Super 120B A12B (Reasoning): Model Comparison",
   "published_at": null,
   "organization": "Z AI",
   "model": "GLM-5.1 (Reasoning)",
   "release_date": "2026-04",
   "access_type": "open source",
   "license": "Mit",
   "significance": "744B, 40B active at inference time",
   "organization_evidence": [
    "# GLM-5.1 (Reasoning) vs. NVIDIA Nemotron 3 Super 120B A12B (Reasoning)",
    "## Model Comparison",
    "| Creator | ![Z AI](/img/logos/zai_small.svg)[](https://z.ai/) | ![NVIDIA](/img/logos/nvidia_small.svg)[](https://www.nvidia.com/) | |"
   ],
   "model_evidence": [
    "# GLM-5.1 (Reasoning) vs. NVIDIA Nemotron 3 Super 120B A12B (Reasoning)",
    "## Model Comparison",
    "| Metric | ![Z AI logo](/img/logos/zai_small.svg)GLM-5.1 (Reasoning) | ![NVIDIA logo](/img/logos/nvidia_small.svg)NVIDIA Nemotron 3 Super 120B A12B (Reasoning) | Analysis |"
   ],
   "release_date_evidence": [
    "# GLM-5.1 (Reasoning) vs. NVIDIA Nemotron 3 Super 120B A12B (Reasoning)",
    "## Model Comparison",
    "| Release Date | April, 2026 | March, 2026 | GLM-5.1 (Reasoning) has a more recent release date than NVIDIA Nemotron 3 Super 120B A12B (Reasoning) |"
   ],
   "access_type_evidence": [
    "# GLM-5.1 (Reasoning) vs. NVIDIA Nemotron 3 Super 120B A12B (Reasoning)",
    "## Model Comparison",
    "| Open Source (Weights) | Yes[](https://huggingface.co/zai-org/GLM-5.1) | Yes[](https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16) | Both GLM-5.1 (Reasoning) and NVIDIA Nemotron 3 Super 120B A12B (Reasoning) are open source |"
   ],
   "license_evidence": [
    "# GLM-5.1 (Reasoning) vs. NVIDIA Nemotron 3 Super 120B A12B (Reasoning)",
    "## Model Comparison",
    "| License | Mit[](https://opensource.org/license/mit) | Nvidia Nemotron Open Model License[](https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/) | |"
   ],
   "significance_evidence": [
    "# GLM-5.1 (Reasoning) vs. NVIDIA Nemotron 3 Super 120B A12B (Reasoning)",
    "## Model Comparison",
    "| Parameters | 744B, 40B active at inference time | 120.6B, 12.7B active at inference time | GLM-5.1 (Reasoning) is larger than NVIDIA Nemotron 3 Super 120B A12B (Reasoning) |"
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/glm-5-1-vs-nvidia-nemotron-3-super-120b-a12b",
   "title": "GLM-5.1 (Reasoning) vs NVIDIA Nemotron 3 Super 120B A12B (Reasoning): Model Comparison",
   "published_at": null,
   "organization": "NVIDIA",
   "model": "NVIDIA Nemotron 3 Super 120B A12B (Reasoning)",
   "release_date": "2026-03",
   "access_type": "open source",
   "license": "Nvidia Nemotron Open Model License",
   "significance": "120.6B, 12.7B active at inference time",
   "organization_evidence": [
    "# GLM-5.1 (Reasoning) vs. NVIDIA Nemotron 3 Super 120B A12B (Reasoning)",
    "## Model Comparison",
    "| Creator | ![Z AI](/img/logos/zai_small.svg)[](https://z.ai/) | ![NVIDIA](/img/logos/nvidia_small.svg)[](https://www.nvidia.com/) | |"
   ],
   "model_evidence": [
    "# GLM-5.1 (Reasoning) vs. NVIDIA Nemotron 3 Super 120B A12B (Reasoning)",
    "## Model Comparison",
    "| Metric | ![Z AI logo](/img/logos/zai_small.svg)GLM-5.1 (Reasoning) | ![NVIDIA logo](/img/logos/nvidia_small.svg)NVIDIA Nemotron 3 Super 120B A12B (Reasoning) | Analysis |"
   ],
   "release_date_evidence": [
    "# GLM-5.1 (Reasoning) vs. NVIDIA Nemotron 3 Super 120B A12B (Reasoning)",
    "## Model Comparison",
    "| Release Date | April, 2026 | March, 2026 | GLM-5.1 (Reasoning) has a more recent release date than NVIDIA Nemotron 3 Super 120B A12B (Reasoning) |"
   ],
   "access_type_evidence": [
    "# GLM-5.1 (Reasoning) vs. NVIDIA Nemotron 3 Super 120B A12B (Reasoning)",
    "## Model Comparison",
    "| Open Source (Weights) | Yes[](https://huggingface.co/zai-org/GLM-5.1) | Yes[](https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16) | Both GLM-5.1 (Reasoning) and NVIDIA Nemotron 3 Super 120B A12B (Reasoning) are open source |"
   ],
   "license_evidence": [
    "# GLM-5.1 (Reasoning) vs. NVIDIA Nemotron 3 Super 120B A12B (Reasoning)",
    "## Model Comparison",
    "| License | Mit[](https://opensource.org/license/mit) | Nvidia Nemotron Open Model License[](https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/) | |"
   ],
   "significance_evidence": [
    "# GLM-5.1 (Reasoning) vs. NVIDIA Nemotron 3 Super 120B A12B (Reasoning)",
    "## Model Comparison",
    "| Parameters | 744B, 40B active at inference time | 120.6B, 12.7B active at inference time | GLM-5.1 (Reasoning) is larger than NVIDIA Nemotron 3 Super 120B A12B (Reasoning) |"
   ]
  },
  {
   "url": "https://auto-blogging.com/ai-tools-software/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - Auto-Blogging",
   "published_at": "2026-07-15T11:07:14",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "1.6T total / 49B active MoE, 1M context, MIT -- resets the price floor",
   "organization_evidence": [
    "Chinese labs such as DeepSeek, Z.ai, Moonshot, and Alibaba have each demonstrated distinct strategic focuses: DeepSeek emphasizes low-cost, high-parameter models with a 1M-token context; Z.ai\u2019s GLM-5.2 leads in open-weight intelligence; Moonshot\u2019s Kimi line targets long-horizon stability; Alibaba\u2019s Qwen family offers compact, self-hostable variants.",
    "## Four Frontier-Class Open Models in Eight Weeks"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Between late April and mid-June 2026, Chinese AI labs launched four major open models: DeepSeek V4 on April 24, MiniMax M3 on June 1, and Kimi K2.7-Code and GLM-5.2 in mid-June.",
    "## Four Frontier-Class Open Models in Eight Weeks"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor",
    "### The production line \u2014 spring 2026",
    "## Four Frontier-Class Open Models in Eight Weeks"
   ]
  },
  {
   "url": "https://auto-blogging.com/ai-tools-software/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - Auto-Blogging",
   "published_at": "2026-07-15T11:07:14",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "cheap 1M-token context, native multimodal, modified-MIT",
   "organization_evidence": [
    "These models, including DeepSeek V4, MiniMax M3, Kimi K2.7-Code, and GLM-5.2, are all downloadable and mostly under permissive licenses, making them highly accessible.",
    "## Four Frontier-Class Open Models in Eight Weeks"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Between late April and mid-June 2026, Chinese AI labs launched four major open models: DeepSeek V4 on April 24, MiniMax M3 on June 1, and Kimi K2.7-Code and GLM-5.2 in mid-June.",
    "## Four Frontier-Class Open Models in Eight Weeks"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "**MiniMax M3**cheap 1M-token context, native multimodal, modified-MIT",
    "### The production line \u2014 spring 2026",
    "## Four Frontier-Class Open Models in Eight Weeks"
   ]
  },
  {
   "url": "https://auto-blogging.com/ai-tools-software/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - Auto-Blogging",
   "published_at": "2026-07-15T11:07:14",
   "organization": "Moonshot",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "agent-run specialist, ~30% fewer thinking tokens than K2.6",
   "organization_evidence": [
    "Chinese labs such as DeepSeek, Z.ai, Moonshot, and Alibaba have each demonstrated distinct strategic focuses: DeepSeek emphasizes low-cost, high-parameter models with a 1M-token context; Z.ai\u2019s GLM-5.2 leads in open-weight intelligence; Moonshot\u2019s Kimi line targets long-horizon stability; Alibaba\u2019s Qwen family offers compact, self-hostable variants.",
    "## Four Frontier-Class Open Models in Eight Weeks"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Between late April and mid-June 2026, Chinese AI labs launched four major open models: DeepSeek V4 on April 24, MiniMax M3 on June 1, and Kimi K2.7-Code and GLM-5.2 in mid-June.",
    "## Four Frontier-Class Open Models in Eight Weeks"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "**Kimi K2.7-Code (Moonshot)**agent-run specialist, ~30% fewer thinking tokens than K2.6",
    "### The production line \u2014 spring 2026",
    "## Four Frontier-Class Open Models in Eight Weeks"
   ]
  },
  {
   "url": "https://auto-blogging.com/ai-tools-software/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - Auto-Blogging",
   "published_at": "2026-07-15T11:07:14",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "753B MoE, MIT, top open-weight on Artificial Analysis index",
   "organization_evidence": [
    "Chinese labs such as DeepSeek, Z.ai, Moonshot, and Alibaba have each demonstrated distinct strategic focuses: DeepSeek emphasizes low-cost, high-parameter models with a 1M-token context; Z.ai\u2019s GLM-5.2 leads in open-weight intelligence; Moonshot\u2019s Kimi line targets long-horizon stability; Alibaba\u2019s Qwen family offers compact, self-hostable variants.",
    "## Four Frontier-Class Open Models in Eight Weeks"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Between late April and mid-June 2026, Chinese AI labs launched four major open models: DeepSeek V4 on April 24, MiniMax M3 on June 1, and Kimi K2.7-Code and GLM-5.2 in mid-June.",
    "## Four Frontier-Class Open Models in Eight Weeks"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "**GLM-5.2 (Z.ai)**753B MoE, MIT, top open-weight on Artificial Analysis index",
    "### The production line \u2014 spring 2026",
    "## Four Frontier-Class Open Models in Eight Weeks"
   ]
  },
  {
   "url": "https://dev.to/breachprotocol/thinking-machines-releases-inkling-now-the-top-ranked-us-open-weights-model-1ead",
   "title": "Thinking Machines releases Inkling, now the top-ranked US open-weights model",
   "published_at": "2026-07-16T02:58:53",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "highest-scoring open-weights model released by any American lab",
   "organization_evidence": [
    "Thinking Machines Lab has released Inkling, a 975-billion-parameter open-weights model published under the permissive Apache 2.0 license.",
    "- **Who:** Thinking Machines Lab, the research company founded by former OpenAI CTO Mira Murati."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "- **When:** Released July 15, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Thinking Machines Lab has released Inkling, a 975-billion-parameter open-weights model published under the permissive Apache 2.0 license.",
    "The independent evaluator Artificial Analysis ranks it at 41 on its Intelligence Index, making it the highest-scoring open-weights model released by any American lab -- ahead of Nvidia's Nemotron 3 Ultra at 38 and far ahead of OpenAI's gpt-oss-120b at 24."
   ]
  },
  {
   "url": "https://fmsolutions.ai/en/blog/thinking-machines-lab-lanca-inkling-modelo-open-weight-975-bilhoes-parametros",
   "title": "975 billion open-weight parameters \u00b7 Blog FM Solutions",
   "published_at": "2026-07-27T00:00:00",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-27",
   "access_type": null,
   "license": null,
   "significance": "an open-weight MoE model with 975 billion parameters",
   "organization_evidence": [
    "# Thinking Machines Lab launches Inkling: 975 billion open-weight parameters",
    "**Thinking Machines Lab** has just launched **Inkling**, an open-weight language model based on a *mixture-of-experts* (MoE) architecture with **975 billion total parameters and 41 billion active parameters per inference**."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Published onJuly 27, 20265 min readFabian Martinelli",
    "# Thinking Machines Lab launches Inkling: 975 billion open-weight parameters"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Thinking Machines Lab launches Inkling: 975 billion open-weight parameters",
    "Thinking Machines Lab launches Inkling, an open-weight MoE model with 975 billion parameters."
   ]
  },
  {
   "url": "https://venturebeat.com/technology/no-cloud-no-gpus-no-problem-liquid-ais-new-model-lfm2-5-2-6b-brings-powerful-ai-agents-to-devices-as-small-as-a-raspberry-pi",
   "title": "No cloud, no GPUs, no problem: Liquid AI's new model LFM2.5-2.6B brings powerful AI agents to devices as small as a Raspberry Pi",
   "published_at": "2026-08-06T22:56:06",
   "organization": "Liquid AI",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "A 2.6-billion-parameter open-weight model designed for agentic workloads, capable of running locally on hardware ranging from smartphones to Raspberry Pi.",
   "organization_evidence": [
    "Earlier this week, the AI startup Liquid, formed in 2023 by former MIT computer scientists, [debuted LFM2.5-2.6B](https://www.liquid.ai/blog/lfm2-5-2-6b), a new open-weight language model designed specifically for agentic workloads."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Earlier this week, the AI startup Liquid, formed in 2023 by former MIT computer scientists, [debuted LFM2.5-2.6B](https://www.liquid.ai/blog/lfm2-5-2-6b), a new open-weight language model designed specifically for agentic workloads."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/qwen3-6-plus-vs-glm-5-non-reasoning",
   "title": "Qwen3.6 Plus vs GLM-5 (Non-reasoning): Model Comparison",
   "published_at": null,
   "organization": "Z AI",
   "model": null,
   "release_date": "2026-02",
   "access_type": null,
   "license": null,
   "significance": null,
   "organization_evidence": [
    "# Qwen3.6 Plus vs. GLM-5 (Non-reasoning)",
    "| Metric | ![Alibaba logo](/img/logos/alibaba_small.svg)Qwen3.6 Plus | ![Z AI logo](/img/logos/zai_small.svg)GLM-5 (Non-reasoning) | Analysis |",
    "| Creator | ![Alibaba](/img/logos/alibaba_small.svg)[](https://www.alibabacloud.com/) | ![Z AI](/img/logos/zai_small.svg)[](https://z.ai/) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Qwen3.6 Plus vs. GLM-5 (Non-reasoning)",
    "| Metric | ![Alibaba logo](/img/logos/alibaba_small.svg)Qwen3.6 Plus | ![Z AI logo](/img/logos/zai_small.svg)GLM-5 (Non-reasoning) | Analysis |",
    "| Release Date | April, 2026 | February, 2026 | Qwen3.6 Plus has a more recent release date than GLM-5 (Non-reasoning) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": null
  },
  {
   "url": "https://perplexityaimagazine.com/ai-news/ai-news-chinese-ai-labs-open-weight-models-12-days-2026",
   "title": "Four Chinese AI Labs Released Frontier-Class Open-Weight Coding Models in 12 Days \u2014 The AI Race Is More Global Than the Headlines Suggest",
   "published_at": "2026-05-14T10:40:07",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "achieved frontier-level performance on agentic engineering benchmarks at a fraction of the inference cost of comparable Western models",
   "organization_evidence": [
    "In a 12-day window during April 2026, four major Chinese artificial intelligence laboratories released open-weight coding models that achieved frontier-level performance on agentic engineering benchmarks at a fraction of the inference cost of comparable Western models.",
    "Z.ai released GLM-5.1, MiniMax released M2.7, Moonshot released Kimi K2.6, and DeepSeek released V4 \u2014 all within less than a fortnight, and all reaching roughly the same capability ceiling on agentic coding tasks."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "In a 12-day window during April 2026, four major Chinese artificial intelligence laboratories released open-weight coding models that achieved frontier-level performance on agentic engineering benchmarks at a fraction of the inference cost of comparable Western models.",
    "Z.ai released GLM-5.1, MiniMax released M2.7, Moonshot released Kimi K2.6, and DeepSeek released V4 \u2014 all within less than a fortnight, and all reaching roughly the same capability ceiling on agentic coding tasks."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "In a 12-day window during April 2026, four major Chinese artificial intelligence laboratories released open-weight coding models that achieved frontier-level performance on agentic engineering benchmarks at a fraction of the inference cost of comparable Western models."
   ]
  },
  {
   "url": "https://perplexityaimagazine.com/ai-news/ai-news-chinese-ai-labs-open-weight-models-12-days-2026",
   "title": "Four Chinese AI Labs Released Frontier-Class Open-Weight Coding Models in 12 Days \u2014 The AI Race Is More Global Than the Headlines Suggest",
   "published_at": "2026-05-14T10:40:07",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "achieved frontier-level performance on agentic engineering benchmarks at a fraction of the inference cost of comparable Western models",
   "organization_evidence": [
    "In a 12-day window during April 2026, four major Chinese artificial intelligence laboratories released open-weight coding models that achieved frontier-level performance on agentic engineering benchmarks at a fraction of the inference cost of comparable Western models.",
    "Z.ai released GLM-5.1, MiniMax released M2.7, Moonshot released Kimi K2.6, and DeepSeek released V4 \u2014 all within less than a fortnight, and all reaching roughly the same capability ceiling on agentic coding tasks."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "In a 12-day window during April 2026, four major Chinese artificial intelligence laboratories released open-weight coding models that achieved frontier-level performance on agentic engineering benchmarks at a fraction of the inference cost of comparable Western models.",
    "Z.ai released GLM-5.1, MiniMax released M2.7, Moonshot released Kimi K2.6, and DeepSeek released V4 \u2014 all within less than a fortnight, and all reaching roughly the same capability ceiling on agentic coding tasks."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "In a 12-day window during April 2026, four major Chinese artificial intelligence laboratories released open-weight coding models that achieved frontier-level performance on agentic engineering benchmarks at a fraction of the inference cost of comparable Western models."
   ]
  },
  {
   "url": "https://perplexityaimagazine.com/ai-news/ai-news-chinese-ai-labs-open-weight-models-12-days-2026",
   "title": "Four Chinese AI Labs Released Frontier-Class Open-Weight Coding Models in 12 Days \u2014 The AI Race Is More Global Than the Headlines Suggest",
   "published_at": "2026-05-14T10:40:07",
   "organization": "Moonshot",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "achieved frontier-level performance on agentic engineering benchmarks at a fraction of the inference cost of comparable Western models",
   "organization_evidence": [
    "In a 12-day window during April 2026, four major Chinese artificial intelligence laboratories released open-weight coding models that achieved frontier-level performance on agentic engineering benchmarks at a fraction of the inference cost of comparable Western models.",
    "Z.ai released GLM-5.1, MiniMax released M2.7, Moonshot released Kimi K2.6, and DeepSeek released V4 \u2014 all within less than a fortnight, and all reaching roughly the same capability ceiling on agentic coding tasks."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "In a 12-day window during April 2026, four major Chinese artificial intelligence laboratories released open-weight coding models that achieved frontier-level performance on agentic engineering benchmarks at a fraction of the inference cost of comparable Western models.",
    "Z.ai released GLM-5.1, MiniMax released M2.7, Moonshot released Kimi K2.6, and DeepSeek released V4 \u2014 all within less than a fortnight, and all reaching roughly the same capability ceiling on agentic coding tasks."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "In a 12-day window during April 2026, four major Chinese artificial intelligence laboratories released open-weight coding models that achieved frontier-level performance on agentic engineering benchmarks at a fraction of the inference cost of comparable Western models."
   ]
  },
  {
   "url": "https://perplexityaimagazine.com/ai-news/ai-news-chinese-ai-labs-open-weight-models-12-days-2026",
   "title": "Four Chinese AI Labs Released Frontier-Class Open-Weight Coding Models in 12 Days \u2014 The AI Race Is More Global Than the Headlines Suggest",
   "published_at": "2026-05-14T10:40:07",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "achieved frontier-level performance on agentic engineering benchmarks at a fraction of the inference cost of comparable Western models",
   "organization_evidence": [
    "In a 12-day window during April 2026, four major Chinese artificial intelligence laboratories released open-weight coding models that achieved frontier-level performance on agentic engineering benchmarks at a fraction of the inference cost of comparable Western models.",
    "Z.ai released GLM-5.1, MiniMax released M2.7, Moonshot released Kimi K2.6, and DeepSeek released V4 \u2014 all within less than a fortnight, and all reaching roughly the same capability ceiling on agentic coding tasks."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "In a 12-day window during April 2026, four major Chinese artificial intelligence laboratories released open-weight coding models that achieved frontier-level performance on agentic engineering benchmarks at a fraction of the inference cost of comparable Western models.",
    "Z.ai released GLM-5.1, MiniMax released M2.7, Moonshot released Kimi K2.6, and DeepSeek released V4 \u2014 all within less than a fortnight, and all reaching roughly the same capability ceiling on agentic coding tasks."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "In a 12-day window during April 2026, four major Chinese artificial intelligence laboratories released open-weight coding models that achieved frontier-level performance on agentic engineering benchmarks at a fraction of the inference cost of comparable Western models."
   ]
  },
  {
   "url": "https://tech-insider.org/au/nvidia-nemotron-3-ultra-2026",
   "title": "NVIDIA Nemotron 3 Ultra: 550B Open AI, #2 to China [2026]",
   "published_at": "2026-06-05T09:15:00",
   "organization": "NVIDIA",
   "model": "NVIDIA Nemotron 3 Ultra",
   "release_date": "2026-06-04",
   "access_type": "open-weight",
   "license": "OpenMDW-1.1",
   "significance": "a fully open 550-billion-parameter reasoning model",
   "organization_evidence": [
    "NVIDIA Nemotron 3 Ultra: What Actually Shipped on 4 June",
    "The company released **NVIDIA Nemotron 3 Ultra**, a fully open 550-billion-parameter reasoning model, and for the first time an American lab can claim a genuinely frontier-class open-weight system."
   ],
   "model_evidence": [
    "NVIDIA Nemotron 3 Ultra: What Actually Shipped on 4 June",
    "The company released **NVIDIA Nemotron 3 Ultra**, a fully open 550-billion-parameter reasoning model, and for the first time an American lab can claim a genuinely frontier-class open-weight system."
   ],
   "release_date_evidence": [
    "NVIDIA Nemotron 3 Ultra: What Actually Shipped on 4 June",
    "On 4 June 2026, it started handing out gold as well."
   ],
   "access_type_evidence": [
    "NVIDIA Nemotron 3 Ultra: What Actually Shipped on 4 June",
    "The company released **NVIDIA Nemotron 3 Ultra**, a fully open 550-billion-parameter reasoning model, and for the first time an American lab can claim a genuinely frontier-class open-weight system."
   ],
   "license_evidence": [
    "NVIDIA Nemotron 3 Ultra: What Actually Shipped on 4 June",
    "Crucially, NVIDIA released not just the weights but the training data and post-training recipes under the permissive OpenMDW-1.1 licence, a level of openness that even Meta\u2019s Llama family has never matched."
   ],
   "significance_evidence": [
    "NVIDIA Nemotron 3 Ultra: What Actually Shipped on 4 June",
    "The company released **NVIDIA Nemotron 3 Ultra**, a fully open 550-billion-parameter reasoning model, and for the first time an American lab can claim a genuinely frontier-class open-weight system."
   ]
  },
  {
   "url": "https://tech-insider.org/au/nvidia-nemotron-3-ultra-2026",
   "title": "NVIDIA Nemotron 3 Ultra: 550B Open AI, #2 to China [2026]",
   "published_at": "2026-06-05T09:15:00",
   "organization": "MiniMax",
   "model": "MiniMax M3",
   "release_date": "2026-06-01",
   "access_type": "open weights",
   "license": null,
   "significance": "a 428-billion-parameter MoE model with 23 billion active parameters, a one-million-token context window and native multimodal input",
   "organization_evidence": [
    "China\u2019s Open-Weight Surge: Kimi, DeepSeek and MiniMax M3",
    "Three days earlier, on 1 June, Chinese lab MiniMax released [MiniMax M3](https://the-decoder.com/minimax-m3-open-weight-model-with-a-million-token-context-challenges-proprietary-leaders/), a 428-billion-parameter MoE model with 23 billion active parameters, a one-million-token context window and native multimodal input."
   ],
   "model_evidence": [
    "China\u2019s Open-Weight Surge: Kimi, DeepSeek and MiniMax M3",
    "Three days earlier, on 1 June, Chinese lab MiniMax released [MiniMax M3](https://the-decoder.com/minimax-m3-open-weight-model-with-a-million-token-context-challenges-proprietary-leaders/), a 428-billion-parameter MoE model with 23 billion active parameters, a one-million-token context window and native multimodal input."
   ],
   "release_date_evidence": [
    "China\u2019s Open-Weight Surge: Kimi, DeepSeek and MiniMax M3",
    "Three days earlier, on 1 June, Chinese lab MiniMax released [MiniMax M3](https://the-decoder.com/minimax-m3-open-weight-model-with-a-million-token-context-challenges-proprietary-leaders/), a 428-billion-parameter MoE model with 23 billion active parameters, a one-million-token context window and native multimodal input."
   ],
   "access_type_evidence": [
    "China\u2019s Open-Weight Surge: Kimi, DeepSeek and MiniMax M3",
    "Three days earlier, on 1 June, Chinese lab MiniMax released [MiniMax M3](https://the-decoder.com/minimax-m3-open-weight-model-with-a-million-token-context-challenges-proprietary-leaders/), a 428-billion-parameter MoE model with 23 billion active parameters, a one-million-token context window and native multimodal input."
   ],
   "license_evidence": null,
   "significance_evidence": [
    "China\u2019s Open-Weight Surge: Kimi, DeepSeek and MiniMax M3",
    "Three days earlier, on 1 June, Chinese lab MiniMax released [MiniMax M3](https://the-decoder.com/minimax-m3-open-weight-model-with-a-million-token-context-challenges-proprietary-leaders/), a 428-billion-parameter MoE model with 23 billion active parameters, a one-million-token context window and native multimodal input."
   ]
  },
  {
   "url": "https://sebastianraschka.com/blog/2026/a-dream-of-spring-for-open-weight.html",
   "title": "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
   "published_at": "2026-02-25T13:26:56",
   "organization": "Arcee AI",
   "model": null,
   "release_date": "2026-01-27",
   "access_type": null,
   "license": null,
   "significance": "A 400B param Mixture-of-Experts flagship model with several interesting architectural components like alternating sliding window attention, QK-Norm, and NoPE.",
   "organization_evidence": [
    "In this article, I will walk you through the ten main releases in chronological order, with a focus on the architecture similarities and differences:",
    "1. Arcee AI\u2019s Trinity Large (Jan 27, 2026)",
    "On January 27, Arcee AI (a company I hadn\u2019t had on my radar up to then) [began releasing](https://www.arcee.ai/blog/trinity-large) versions of their open-weight 400B Trinity Large LLMs on the [model hub](https://huggingface.co/arcee-ai/Trinity-Large-Preview), along with two smaller variants:"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "In this article, I will walk you through the ten main releases in chronological order, with a focus on the architecture similarities and differences:",
    "1. Arcee AI\u2019s Trinity Large (Jan 27, 2026)",
    "On January 27, Arcee AI (a company I hadn\u2019t had on my radar up to then) [began releasing](https://www.arcee.ai/blog/trinity-large) versions of their open-weight 400B Trinity Large LLMs on the [model hub](https://huggingface.co/arcee-ai/Trinity-Large-Preview), along with two smaller variants:"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "In this article, I will walk you through the ten main releases in chronological order, with a focus on the architecture similarities and differences:",
    "1. Arcee AI\u2019s Trinity Large (Jan 27, 2026)",
    "Their flagship large model is a 400B param [Mixture-of-Experts (MoE)](https://magazine.sebastianraschka.com/i/168650848/12-mixture-of-experts-moe) with 13B active parameters."
   ]
  },
  {
   "url": "https://sebastianraschka.com/blog/2026/a-dream-of-spring-for-open-weight.html",
   "title": "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
   "published_at": "2026-02-25T13:26:56",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-01-27",
   "access_type": null,
   "license": null,
   "significance": "A 1-trillion-parameter native multimodal model that set a new open-weight performance ceiling at the time of its release.",
   "organization_evidence": [
    "In this article, I will walk you through the ten main releases in chronological order, with a focus on the architecture similarities and differences:",
    "2. Moonshot AI\u2019s Kimi K2.5 (Jan 27, 2026)",
    "# 2. Moonshot AI\u2019s Kimi K2.5: A DeepSeek-Like Model at a 1-Trillion-Parameter Scale"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "In this article, I will walk you through the ten main releases in chronological order, with a focus on the architecture similarities and differences:",
    "2. Moonshot AI\u2019s Kimi K2.5 (Jan 27, 2026)",
    "While Arcee Trinity essentially matched the modeling performance of the older GLM-4.5 model,[ Kimi K2.5 is an open-weight model](https://huggingface.co/moonshotai/Kimi-K2.5) that set a new open-weight performance ceiling at the time of its release on Jan 27."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "In this article, I will walk you through the ten main releases in chronological order, with a focus on the architecture similarities and differences:",
    "2. Moonshot AI\u2019s Kimi K2.5 (Jan 27, 2026)",
    "# 2. Moonshot AI\u2019s Kimi K2.5: A DeepSeek-Like Model at a 1-Trillion-Parameter Scale",
    "Kimi K2.5 is a native multimodal model built upon Kimi K2 through large-scale joint pre-training on approximately 15 trillion mixed visual and text tokens."
   ]
  },
  {
   "url": "https://sebastianraschka.com/blog/2026/a-dream-of-spring-for-open-weight.html",
   "title": "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
   "published_at": "2026-02-25T13:26:56",
   "organization": "StepFun",
   "model": null,
   "release_date": "2026-02-01",
   "access_type": null,
   "license": null,
   "significance": "A 196B parameter model with good performance and 100 tokens/sec throughput at a 128k context length.",
   "organization_evidence": [
    "In this article, I will walk you through the ten main releases in chronological order, with a focus on the architecture similarities and differences:",
    "3. StepFun Step 3.5 Flash (Feb 1, 2026)",
    "# 3. StepFun\u2019s Step 3.5 Flash: Good Performance at Great Tokens/Sec Throughput"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "In this article, I will walk you through the ten main releases in chronological order, with a focus on the architecture similarities and differences:",
    "3. StepFun Step 3.5 Flash (Feb 1, 2026)"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "In this article, I will walk you through the ten main releases in chronological order, with a focus on the architecture similarities and differences:",
    "3. StepFun Step 3.5 Flash (Feb 1, 2026)",
    "Step 3.5 Flash is a 196B parameter model that is more than 3x smaller than the recent DeepSeek V3.2 model (671B) while being slightly ahead in modeling performance benchmarks."
   ]
  },
  {
   "url": "https://sebastianraschka.com/blog/2026/a-dream-of-spring-for-open-weight.html",
   "title": "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
   "published_at": "2026-02-25T13:26:56",
   "organization": "Qwen team",
   "model": null,
   "release_date": "2026-02-03",
   "access_type": null,
   "license": null,
   "significance": "An 80B attention-hybrid coding model that outperformed much larger models on coding tasks.",
   "organization_evidence": [
    "In this article, I will walk you through the ten main releases in chronological order, with a focus on the architecture similarities and differences:",
    "4. Qwen3-Coder-Next (Feb 3, 2026)",
    "In early February 2026, the Qwen3 team shared the 80B Qwen3-Coder-Next model (3B parameters active), which made big headlines for outperforming much larger models like DeepSeek V3.2 (37B active) and Kimi K2.5 and GLM-4.7 (both 32B active) on coding tasks."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "In this article, I will walk you through the ten main releases in chronological order, with a focus on the architecture similarities and differences:",
    "4. Qwen3-Coder-Next (Feb 3, 2026)",
    "In early February 2026, the Qwen3 team shared the 80B Qwen3-Coder-Next model (3B parameters active), which made big headlines for outperforming much larger models like DeepSeek V3.2 (37B active) and Kimi K2.5 and GLM-4.7 (both 32B active) on coding tasks."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "In this article, I will walk you through the ten main releases in chronological order, with a focus on the architecture similarities and differences:",
    "4. Qwen3-Coder-Next (Feb 3, 2026)",
    "In early February 2026, the Qwen3 team shared the 80B Qwen3-Coder-Next model (3B parameters active), which made big headlines for outperforming much larger models like DeepSeek V3.2 (37B active) and Kimi K2.5 and GLM-4.7 (both 32B active) on coding tasks."
   ]
  },
  {
   "url": "https://sebastianraschka.com/blog/2026/a-dream-of-spring-for-open-weight.html",
   "title": "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
   "published_at": "2026-02-25T13:26:56",
   "organization": "z.AI",
   "model": null,
   "release_date": "2026-02-12",
   "access_type": null,
   "license": null,
   "significance": "A 744B parameter flagship model that appeared to be on par with major flagship LLM offerings.",
   "organization_evidence": [
    "In this article, I will walk you through the ten main releases in chronological order, with a focus on the architecture similarities and differences:",
    "5. z.AI\u2019s GLM-5 (Feb 12, 2026)",
    "# 5. z.AI\u2019s GLM-5: A New Flagship Open-Weight Model",
    "The [GLM-5 release](https://z.ai/blog/glm-5) on February 12th was a big deal, because at the time of its release it appeared to be on par with the major flagship LLM offerings, including GPT-5.2 extra-high, Gemini Pro 3, and Claude 4.6 Opus."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "In this article, I will walk you through the ten main releases in chronological order, with a focus on the architecture similarities and differences:",
    "5. z.AI\u2019s GLM-5 (Feb 12, 2026)",
    "The [GLM-5 release](https://z.ai/blog/glm-5) on February 12th was a big deal, because at the time of its release it appeared to be on par with the major flagship LLM offerings, including GPT-5.2 extra-high, Gemini Pro 3, and Claude 4.6 Opus."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "In this article, I will walk you through the ten main releases in chronological order, with a focus on the architecture similarities and differences:",
    "5. z.AI\u2019s GLM-5 (Feb 12, 2026)",
    "# 5. z.AI\u2019s GLM-5: A New Flagship Open-Weight Model",
    "The [GLM-5 release](https://z.ai/blog/glm-5) on February 12th was a big deal, because at the time of its release it appeared to be on par with the major flagship LLM offerings, including GPT-5.2 extra-high, Gemini Pro 3, and Claude 4.6 Opus."
   ]
  },
  {
   "url": "https://sebastianraschka.com/blog/2026/a-dream-of-spring-for-open-weight.html",
   "title": "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
   "published_at": "2026-02-25T13:26:56",
   "organization": "MiniMaxAI",
   "model": null,
   "release_date": "2026-02-12",
   "access_type": null,
   "license": null,
   "significance": "A 230B parameter model with classic design and strong coding performance that achieved high popularity.",
   "organization_evidence": [
    "In this article, I will walk you through the ten main releases in chronological order, with a focus on the architecture similarities and differences:",
    "6. MiniMax M2.5 (Feb 12, 2026)",
    "The aforementioned GLM-5 and Kimi K2.5 are popular open-weight models, but according to[ OpenRouter statistics](https://openrouter.ai/rankings), they pale in comparison to[ MiniMax M2.5](https://huggingface.co/MiniMaxAI/MiniMax-M2.5), which was released on February 12 as well."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "In this article, I will walk you through the ten main releases in chronological order, with a focus on the architecture similarities and differences:",
    "6. MiniMax M2.5 (Feb 12, 2026)",
    "The aforementioned GLM-5 and Kimi K2.5 are popular open-weight models, but according to[ OpenRouter statistics](https://openrouter.ai/rankings), they pale in comparison to[ MiniMax M2.5](https://huggingface.co/MiniMaxAI/MiniMax-M2.5), which was released on February 12 as well."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "In this article, I will walk you through the ten main releases in chronological order, with a focus on the architecture similarities and differences:",
    "6. MiniMax M2.5 (Feb 12, 2026)",
    "# 6. MiniMax M2.5: A Strong Coder with \u201cOnly\u201d 230B Parameters",
    "Architecture-wise, MiniMax M2.5 is a 230B model with a fairly classic design: just plain Grouped Query Attention, no sliding window attention or other efficiency improvements."
   ]
  },
  {
   "url": "https://sebastianraschka.com/blog/2026/a-dream-of-spring-for-open-weight.html",
   "title": "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
   "published_at": "2026-02-25T13:26:56",
   "organization": "Nanbeige",
   "model": null,
   "release_date": "2026-02-13",
   "access_type": null,
   "license": null,
   "significance": "A small 3B model targeted at on-device use cases with performance way ahead of Qwen3.",
   "organization_evidence": [
    "In this article, I will walk you through the ten main releases in chronological order, with a focus on the architecture similarities and differences:",
    "7. Nanbeige 4.1 3B (Feb 13, 2026)"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "In this article, I will walk you through the ten main releases in chronological order, with a focus on the architecture similarities and differences:",
    "7. Nanbeige 4.1 3B (Feb 13, 2026)"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "In this article, I will walk you through the ten main releases in chronological order, with a focus on the architecture similarities and differences:",
    "7. Nanbeige 4.1 3B (Feb 13, 2026)",
    "Why I am mentioning all this is that Nanbeige 4.1 3B seems to target the \u201csmall\u201d LLM on-device use case that Qwen3 is so popular for."
   ]
  },
  {
   "url": "https://sebastianraschka.com/blog/2026/a-dream-of-spring-for-open-weight.html",
   "title": "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
   "published_at": "2026-02-25T13:26:56",
   "organization": "Qwen team",
   "model": null,
   "release_date": "2026-02-15",
   "access_type": null,
   "license": null,
   "significance": "A 397B-A17B Mixture-of-Experts model adopting hybrid attention and multimodal support.",
   "organization_evidence": [
    "In this article, I will walk you through the ten main releases in chronological order, with a focus on the architecture similarities and differences:",
    "8. Qwen 3.5 (Feb 15, 2026)",
    "However, the Qwen team just released a new Qwen3.5 model variant on February 15."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "In this article, I will walk you through the ten main releases in chronological order, with a focus on the architecture similarities and differences:",
    "8. Qwen 3.5 (Feb 15, 2026)",
    "However, the Qwen team just released a new Qwen3.5 model variant on February 15."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "In this article, I will walk you through the ten main releases in chronological order, with a focus on the architecture similarities and differences:",
    "8. Qwen 3.5 (Feb 15, 2026)",
    "Qwen3.5 397B-A17B, a Mixture-of-Experts (MoE) with 397B parameters (17B active per token), is a step up from the largest Qwen3 model, which is 235B parameters in size."
   ]
  },
  {
   "url": "https://sebastianraschka.com/blog/2026/a-dream-of-spring-for-open-weight.html",
   "title": "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
   "published_at": "2026-02-25T13:26:56",
   "organization": "Ant Group",
   "model": null,
   "release_date": "2026-02-16",
   "access_type": null,
   "license": null,
   "significance": "A 1-trillion-parameter model featuring a hybrid attention architecture with Lightning Attention and MLA.",
   "organization_evidence": [
    "In this article, I will walk you through the ten main releases in chronological order, with a focus on the architecture similarities and differences:",
    "9. Ant Group\u2019s Ling 2.5 1T & Ring 2.5 1T (Feb 16, 2026)",
    "# 9. Ant Group\u2019s Ling 2.5 1T with Lightning Attention"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "In this article, I will walk you through the ten main releases in chronological order, with a focus on the architecture similarities and differences:",
    "9. Ant Group\u2019s Ling 2.5 1T & Ring 2.5 1T (Feb 16, 2026)"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "In this article, I will walk you through the ten main releases in chronological order, with a focus on the architecture similarities and differences:",
    "9. Ant Group\u2019s Ling 2.5 1T & Ring 2.5 1T (Feb 16, 2026)"
   ]
  },
  {
   "url": "https://sebastianraschka.com/blog/2026/a-dream-of-spring-for-open-weight.html",
   "title": "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
   "published_at": "2026-02-25T13:26:56",
   "organization": "Cohere",
   "model": null,
   "release_date": "2026-02-17",
   "access_type": null,
   "license": null,
   "significance": "A 3.35B parameter multilingual model featuring parallel transformer blocks.",
   "organization_evidence": [
    "In this article, I will walk you through the ten main releases in chronological order, with a focus on the architecture similarities and differences:",
    "10. Cohere\u2019s Tiny Aya (Feb 17, 2026)",
    "Released on February 17, Tiny Aya is a new, \u201csmall\u201d LLM by Cohere that is said to be the \u201cmost capable multilingual open-weight model\u201d at the 3B parameter size class."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "In this article, I will walk you through the ten main releases in chronological order, with a focus on the architecture similarities and differences:",
    "10. Cohere\u2019s Tiny Aya (Feb 17, 2026)",
    "Released on February 17, Tiny Aya is a new, \u201csmall\u201d LLM by Cohere that is said to be the \u201cmost capable multilingual open-weight model\u201d at the 3B parameter size class."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "In this article, I will walk you through the ten main releases in chronological order, with a focus on the architecture similarities and differences:",
    "10. Cohere\u2019s Tiny Aya (Feb 17, 2026)",
    "That aside, Aya is a 3.35B parameter model that comes in several flavors that are useful for"
   ]
  },
  {
   "url": "https://sebastianraschka.com/blog/2026/a-dream-of-spring-for-open-weight.html",
   "title": "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
   "published_at": "2026-02-25T13:26:56",
   "organization": "Sarvam",
   "model": null,
   "release_date": "2026-03-06",
   "access_type": null,
   "license": null,
   "significance": "Two reasoning models in 30B and 105B sizes with strong performance on Indian languages and agentic reasoning.",
   "organization_evidence": [
    "In this article, I will walk you through the ten main releases in chronological order, with a focus on the architecture similarities and differences:",
    "11. Update 1: Sarvam 30B and 105B (Mar 6, 2026)",
    "# Update 1: Sarvam 30B and 105B (Mar 6, 2026)"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "In this article, I will walk you through the ten main releases in chronological order, with a focus on the architecture similarities and differences:",
    "11. Update 1: Sarvam 30B and 105B (Mar 6, 2026)"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "In this article, I will walk you through the ten main releases in chronological order, with a focus on the architecture similarities and differences:",
    "11. Update 1: Sarvam 30B and 105B (Mar 6, 2026)"
   ]
  },
  {
   "url": "https://digitalapplied.com/blog/tencent-hunyuan-hy3-open-weight-reasoning-model-2026",
   "title": "Tencent's Hunyuan Hy3: Open-Weight Reasoning Arrives",
   "published_at": "2026-07-05T23:00:00",
   "organization": "Tencent",
   "model": null,
   "release_date": "2026-07-06",
   "access_type": null,
   "license": null,
   "significance": "a 295-billion-parameter Mixture-of-Experts that activates just 21 billion parameters per forward pass",
   "organization_evidence": [
    "Tencent released the full Hunyuan Hy3 open-weight reasoning model on July 6, 2026 \u2014 a 295-billion-parameter Mixture-of-Experts that activates just 21 billion parameters per forward pass, shipped under Apache 2.0 with no field-of-use or geographic restrictions.",
    "Tencent released the full Hunyuan Hy3 open-weight reasoning model on July 6, 2026 \u2014 a 295-billion-parameter Mixture-of-Experts that activates just 21 billion parameters per forward pass, shipped under Apache 2.0 with no field-of-use or geographic restrictions. The headline for most coverage was a benchmark race. The headline that actually matters for teams evaluating self-hosting is that Hy3 runs in under half the memory of the rival it is measured against."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Tencent released the full Hunyuan Hy3 open-weight reasoning model on July 6, 2026 \u2014 a 295-billion-parameter Mixture-of-Experts that activates just 21 billion parameters per forward pass, shipped under Apache 2.0 with no field-of-use or geographic restrictions.",
    "Tencent released the full Hunyuan Hy3 open-weight reasoning model on July 6, 2026 \u2014 a 295-billion-parameter Mixture-of-Experts that activates just 21 billion parameters per forward pass, shipped under Apache 2.0 with no field-of-use or geographic restrictions. The headline for most coverage was a benchmark race. The headline that actually matters for teams evaluating self-hosting is that Hy3 runs in under half the memory of the rival it is measured against."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Tencent released the full Hunyuan Hy3 open-weight reasoning model on July 6, 2026 \u2014 a 295-billion-parameter Mixture-of-Experts that activates just 21 billion parameters per forward pass, shipped under Apache 2.0 with no field-of-use or geographic restrictions.",
    "Tencent released the full Hunyuan Hy3 open-weight reasoning model on July 6, 2026 \u2014 a 295-billion-parameter Mixture-of-Experts that activates just 21 billion parameters per forward pass, shipped under Apache 2.0 with no field-of-use or geographic restrictions. The headline for most coverage was a benchmark race. The headline that actually matters for teams evaluating self-hosting is that Hy3 runs in under half the memory of the rival it is measured against."
   ]
  },
  {
   "url": "https://huggingface.co/blog/deepseekv4",
   "title": "DeepSeek-V4: a million-token context that agents can actually use",
   "published_at": "2026-04-24T00:00:00",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "1.6T total parameters with 49B active and 284B total with 13B active, offering efficient large context length support designed for agentic tasks",
   "organization_evidence": [
    "Published April 24, 2026",
    "DeepSeek released V4 today."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Published April 24, 2026",
    "DeepSeek released V4 today."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Published April 24, 2026",
    "DeepSeek released V4 today. Two MoE checkpoints are on the Hub: DeepSeek-V4-Pro at 1.6T total parameters with 49B active, and DeepSeek-V4-Flash at 284B total with 13B active. Both have a 1M-token context window. The benchmark numbers are competitive, but not SOTA. It doesn't matter. The real innovation is how DeepSeek v4 is designed for efficient large context length support, and hence as one of the best candidates for agentic tasks."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/glm-4-7-flash-non-reasoning",
   "title": "GLM-4.7-Flash - Intelligence, Performance & Price Analysis",
   "published_at": null,
   "organization": "Z AI",
   "model": null,
   "release_date": "2026-01-19",
   "access_type": null,
   "license": null,
   "significance": "16 on Artificial Analysis Intelligence Index and 200k context window",
   "organization_evidence": [
    "Z AI has launched a newer model, [GLM-5](/models/glm-5-non-reasoning). We suggest considering it instead.",
    "# GLM-4.7-Flash (Non-reasoning) Intelligence, Performance & Price Analysis",
    "GLM-4.7-Flash (Non-reasoning) was created by Z AI."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# GLM-4.7-Flash (Non-reasoning) Intelligence, Performance & Price Analysis",
    "GLM-4.7-Flash (Non-reasoning) was released on January 19, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# GLM-4.7-Flash (Non-reasoning) Intelligence, Performance & Price Analysis",
    "GLM-4.7-Flash (Non-reasoning) scores 16 on the Artificial Analysis Intelligence Index, placing it well above average among comparable models (median: 6).",
    "GLM-4.7-Flash (Non-reasoning) has a context window of 200k tokens."
   ]
  },
  {
   "url": "https://benchlm.ai/blog/posts/thinking-machines-chose-open-weights-first",
   "title": "Thinking Machines Chose Open Weights First",
   "published_at": "2026-07-15T00:00:00",
   "organization": "Thinking Machines",
   "model": "Inkling",
   "release_date": "2026-07-15",
   "access_type": "open weight",
   "license": "Apache 2.0",
   "significance": "a 975-billion-parameter mixture-of-experts model with 41 billion active parameters, a one-million-token context window, and native text, image, and audio input",
   "organization_evidence": [
    "Thinking Machines Chose Open Weights First",
    "Why Thinking Machines made its first foundation-model release open weight, what Inkling changes for the lab, and where it falls short of the closed frontier.",
    "Seventeen months after Thinking Machines Lab publicly introduced itself, the company released its first foundation model on July 15, 2026."
   ],
   "model_evidence": [
    "Thinking Machines Chose Open Weights First",
    "Why Thinking Machines made its first foundation-model release open weight, what Inkling changes for the lab, and where it falls short of the closed frontier.",
    "Seventeen months after Thinking Machines Lab publicly introduced itself, the company released its first foundation model on July 15, 2026.",
    "[Inkling](https://thinkingmachines.ai/news/introducing-inkling/) is a 975-billion-parameter mixture-of-experts model with 41 billion active parameters, a one-million-token context window, and native text, image, and audio input."
   ],
   "release_date_evidence": [
    "Thinking Machines Chose Open Weights First",
    "Why Thinking Machines made its first foundation-model release open weight, what Inkling changes for the lab, and where it falls short of the closed frontier.",
    "Seventeen months after Thinking Machines Lab publicly introduced itself, the company released its first foundation model on July 15, 2026."
   ],
   "access_type_evidence": [
    "Thinking Machines Chose Open Weights First",
    "Why Thinking Machines made its first foundation-model release open weight, what Inkling changes for the lab, and where it falls short of the closed frontier.",
    "Seventeen months after Thinking Machines Lab publicly introduced itself, the company released its first foundation model on July 15, 2026."
   ],
   "license_evidence": [
    "Thinking Machines Chose Open Weights First",
    "Why Thinking Machines made its first foundation-model release open weight, what Inkling changes for the lab, and where it falls short of the closed frontier.",
    "The weights are on Hugging Face under Apache 2.0."
   ],
   "significance_evidence": [
    "Thinking Machines Chose Open Weights First",
    "Why Thinking Machines made its first foundation-model release open weight, what Inkling changes for the lab, and where it falls short of the closed frontier.",
    "[Inkling](https://thinkingmachines.ai/news/introducing-inkling/) is a 975-billion-parameter mixture-of-experts model with 41 billion active parameters, a one-million-token context window, and native text, image, and audio input."
   ]
  },
  {
   "url": "https://techcrunch.com/2026/03/22/cursor-admits-its-new-coding-model-was-built-on-top-of-moonshot-ais-kimi",
   "title": "Cursor admits its new coding model was built on top of Moonshot AI\u2019s Kimi",
   "published_at": "2026-03-22T18:41:09",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-01-27",
   "access_type": null,
   "license": null,
   "significance": "an open source model recently released",
   "organization_evidence": [
    "Kimi 2.5 being [an open source model recently released by Moonshot AI](https://techcrunch.com/2026/01/27/chinas-moonshot-releases-a-new-open-source-model-kimi-k2-5-and-a-coding-agent/), a Chinese company backed by Alibaba and HongShan (formerly Sequoia China)."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Kimi 2.5 being [an open source model recently released by Moonshot AI](https://techcrunch.com/2026/01/27/chinas-moonshot-releases-a-new-open-source-model-kimi-k2-5-and-a-coding-agent/), a Chinese company backed by Alibaba and HongShan (formerly Sequoia China)."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Kimi 2.5 being [an open source model recently released by Moonshot AI](https://techcrunch.com/2026/01/27/chinas-moonshot-releases-a-new-open-source-model-kimi-k2-5-and-a-coding-agent/), a Chinese company backed by Alibaba and HongShan (formerly Sequoia China)."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/nvidia-nemotron-3-super-120b-a12b-vs-gpt-oss-120b?intelligence-category=open-weights-vs-proprietary&models=gpt-5-4-non-reasoning_llama-3-3-instruct-70b_gemini-2-5-pro_mistral-large-3_nvidia-nemotron-3-super-120b-a12b_nvidia-nemotron-nano-12b-v2-vl_claude-4-5-sonnet_grok-4-1-fast",
   "title": "NVIDIA Nemotron 3 Super 120B A12B (Reasoning) vs gpt-oss-120B (high): Model Comparison",
   "published_at": null,
   "organization": "NVIDIA",
   "model": null,
   "release_date": "2026-03",
   "access_type": null,
   "license": null,
   "significance": "120.6B, 12.7B active at inference time",
   "organization_evidence": [
    "# NVIDIA Nemotron 3 Super 120B A12B (Reasoning) vs. gpt-oss-120B (high)",
    "| Creator | ![NVIDIA](/img/logos/nvidia_small.svg)[](https://www.nvidia.com/) | ![OpenAI](/img/logos/openai_small.svg)[](https://openai.com/) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# NVIDIA Nemotron 3 Super 120B A12B (Reasoning) vs. gpt-oss-120B (high)",
    "| Metric | ![NVIDIA logo](/img/logos/nvidia_small.svg)NVIDIA Nemotron 3 Super 120B A12B (Reasoning) | ![OpenAI logo](/img/logos/openai_small.svg)gpt-oss-120B (high) | Analysis |",
    "| Release Date | March, 2026 | August, 2025 | NVIDIA Nemotron 3 Super 120B A12B (Reasoning) has a more recent release date than gpt-oss-120B (high) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# NVIDIA Nemotron 3 Super 120B A12B (Reasoning) vs. gpt-oss-120B (high)",
    "| Metric | ![NVIDIA logo](/img/logos/nvidia_small.svg)NVIDIA Nemotron 3 Super 120B A12B (Reasoning) | ![OpenAI logo](/img/logos/openai_small.svg)gpt-oss-120B (high) | Analysis |",
    "| Parameters | 120.6B, 12.7B active at inference time | 117B, 5.1B active at inference time | NVIDIA Nemotron 3 Super 120B A12B (Reasoning) is larger than gpt-oss-120B (high) |"
   ]
  },
  {
   "url": "https://techpillow.co/blog/meituan-longcat-2-open-source-chinese-chips-2026",
   "title": "Meituan LongCat-2.0 Open Source Chinese Chips 2026",
   "published_at": "2026-07-05T00:00:00",
   "organization": "Meituan",
   "model": null,
   "release_date": "2026-06-30",
   "access_type": null,
   "license": null,
   "significance": "1.6-trillion-parameter Mixture-of-Experts large language model trained entirely on domestic Chinese AI chips with a 1-million-token context window",
   "organization_evidence": [
    "![Meituan Open-Sources LongCat-2.0: Frontier AI Without Nvidia](/_next/image?url=%2Fimages%2Fblog%2Fai.png&w=3840&q=75)",
    "On 30 June 2026, Meituan \u2014 China's dominant food delivery and e-commerce platform \u2014 open-sourced LongCat-2.0, a 1.6-trillion-parameter Mixture-of-Experts large language model trained entirely on domestic Chinese AI chips."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## Meituan Reveals LongCat-2.0 on 30 June 2026",
    "On 30 June 2026, Meituan \u2014 China's dominant food delivery and e-commerce platform \u2014 open-sourced LongCat-2.0, a 1.6-trillion-parameter Mixture-of-Experts large language model trained entirely on domestic Chinese AI chips."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "![Meituan Open-Sources LongCat-2.0: Frontier AI Without Nvidia](/_next/image?url=%2Fimages%2Fblog%2Fai.png&w=3840&q=75)",
    "On 30 June 2026, Meituan \u2014 China's dominant food delivery and e-commerce platform \u2014 open-sourced LongCat-2.0, a 1.6-trillion-parameter Mixture-of-Experts large language model trained entirely on domestic Chinese AI chips."
   ]
  },
  {
   "url": "https://winzheng.com/en/article/thinking-machines-inkling-open-weights-release",
   "title": "Thinking Machines Releases Inkling, Opens 975 Billion Parameter Multimodal Weights",
   "published_at": "2026-07-16T12:19:33",
   "organization": "Thinking Machines",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "975 billion total parameters multimodal model supporting text, image, and audio",
   "organization_evidence": [
    "# Thinking Machines Releases Inkling, Opens 975 Billion Parameter Multimodal Weights",
    "Thinking Machines launched the Inkling model on July 15, 2026, with 975 billion total parameters, 41 billion activated parameters, supporting text, image, and audio inputs, and releasing full weights for download and fine-tuning."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Thinking Machines Releases Inkling, Opens 975 Billion Parameter Multimodal Weights",
    "Thinking Machines launched the Inkling model on July 15, 2026, with 975 billion total parameters, 41 billion activated parameters, supporting text, image, and audio inputs, and releasing full weights for download and fine-tuning."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Thinking Machines Releases Inkling, Opens 975 Billion Parameter Multimodal Weights",
    "Thinking Machines launched the Inkling model on July 15, 2026, with 975 billion total parameters, 41 billion activated parameters, supporting text, image, and audio inputs, and releasing full weights for download and fine-tuning."
   ]
  },
  {
   "url": "https://techsparking.com/poolside-releases-laguna-s-2-1-an-open-weight-agentic-coding-model-that-punches-above-its-weight-class-on-swe-bench-multilingual",
   "title": "Poolside Releases Laguna S 2.1, an Open-Weight Agentic Coding Model That Punches Above Its Weight Class on SWE-Bench Multilingual - TECH SPARKING",
   "published_at": "2026-07-22T00:01:00",
   "organization": "Poolside",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "118B-parameter open-weight agentic coding model that holds its own against models several times its size on long-horizon coding benchmarks",
   "organization_evidence": [
    "# Poolside Releases Laguna S 2.1, an Open-Weight Agentic Coding Model That Punches Above Its Weight Class on SWE-Bench Multilingual",
    "Poolside has released the Laguna S 2.1, a model 118B-parameter open-weight model designed for agent coding."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Poolside Releases Laguna S 2.1, an Open-Weight Agentic Coding Model That Punches Above Its Weight Class on SWE-Bench Multilingual",
    "Poolside has released the Laguna S 2.1, a model 118B-parameter open-weight model designed for agent coding.",
    "In long-horizon coding benchmarks, the Laguna S 2.1 held its own against models several times its size, including DeepSeek-V4-Pro-Max, NVIDIA\u2019s Nemotron 3 Ultra, and Thinking Machines\u2019 Inkling."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/minimax-m3",
   "title": "MiniMax-M3 - Intelligence, Performance & Price Analysis | Artificial Analysis",
   "published_at": null,
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "428B parameter reasoning model with text, image, and video input and a 1M context window",
   "organization_evidence": [
    "MiniMax-M3 Intelligence, Performance & Price Analysis",
    "MiniMax-M3 was created by MiniMax."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Released June 2026",
    "MiniMax-M3 Intelligence, Performance & Price Analysis",
    "MiniMax-M3 was released on June 1, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "MiniMax-M3 Intelligence, Performance & Price Analysis",
    "The model supports text, image, and video input, outputs text, and has a 1M tokens context window.",
    "| Total parameters | 428B |",
    "Yes, MiniMax-M3 is a reasoning model. It uses extended thinking or chain-of-thought reasoning to work through complex problems before providing an answer."
   ]
  },
  {
   "url": "https://gadgetfee.com/tech-trends-innovations/ai-innovation-in-china-four-frontier-open-models-launched-in-just-eight-weeks",
   "title": "AI Innovation In China: Four Frontier Open Models Launched In Just Eight Weeks - GadgetFee",
   "published_at": "2026-07-14T14:27:53",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "1.6 trillion total / 49B active MoE, 1M context, MIT \u2014 resets the price floor",
   "organization_evidence": [
    "Four Chinese laboratories released downloadable, high-capability open-weight AI models between April 24 and mid-June 2026.",
    "The sequence began with **DeepSeek V4 Pro and Flash**, described as mixture-of-experts models with 1.6 trillion total parameters, 49 billion active parameters and a one-million-token context window."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The sequence began with **DeepSeek V4 Pro and Flash**, described as mixture-of-experts models with 1.6 trillion total parameters, 49 billion active parameters and a one-million-token context window.",
    "The reported sequence includes **DeepSeek V4** on April 24, **MiniMax M3** on June 1, Moonshot AI\u2019s Kimi K2.7-Code around June 13 and Z.ai\u2019s GLM-5.2 in mid-June 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor",
    "### The production line \u2014 spring 2026"
   ]
  },
  {
   "url": "https://gadgetfee.com/tech-trends-innovations/ai-innovation-in-china-four-frontier-open-models-launched-in-just-eight-weeks",
   "title": "AI Innovation In China: Four Frontier Open Models Launched In Just Eight Weeks - GadgetFee",
   "published_at": "2026-07-14T14:27:53",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "cheap 1M-token context, native multimodal, modified-MIT",
   "organization_evidence": [
    "Four Chinese laboratories released downloadable, high-capability open-weight AI models between April 24 and mid-June 2026.",
    "**MiniMax M3 followed on June 1** with native multimodal support and a modified MIT-style license, according to the Thorsten Meyer AI market report."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "**MiniMax M3 followed on June 1** with native multimodal support and a modified MIT-style license, according to the Thorsten Meyer AI market report.",
    "The reported sequence includes **DeepSeek V4** on April 24, **MiniMax M3** on June 1, Moonshot AI\u2019s Kimi K2.7-Code around June 13 and Z.ai\u2019s GLM-5.2 in mid-June 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "**MiniMax M3**cheap 1M-token context, native multimodal, modified-MIT",
    "### The production line \u2014 spring 2026"
   ]
  },
  {
   "url": "https://gadgetfee.com/tech-trends-innovations/ai-innovation-in-china-four-frontier-open-models-launched-in-just-eight-weeks",
   "title": "AI Innovation In China: Four Frontier Open Models Launched In Just Eight Weeks - GadgetFee",
   "published_at": "2026-07-14T14:27:53",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "agent-run specialist, ~30% fewer thinking tokens than K2.6",
   "organization_evidence": [
    "Four Chinese laboratories released downloadable, high-capability open-weight AI models between April 24 and mid-June 2026.",
    "Moonshot AI released **Kimi K2.7-Code around June 13**, targeting long-running coding agents."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Moonshot AI released **Kimi K2.7-Code around June 13**, targeting long-running coding agents.",
    "The reported sequence includes **DeepSeek V4** on April 24, **MiniMax M3** on June 1, Moonshot AI\u2019s Kimi K2.7-Code around June 13 and Z.ai\u2019s GLM-5.2 in mid-June 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "**Kimi K2.7-Code (Moonshot)**agent-run specialist, ~30% fewer thinking tokens than K2.6",
    "### The production line \u2014 spring 2026"
   ]
  },
  {
   "url": "https://gadgetfee.com/tech-trends-innovations/ai-innovation-in-china-four-frontier-open-models-launched-in-just-eight-weeks",
   "title": "AI Innovation In China: Four Frontier Open Models Launched In Just Eight Weeks - GadgetFee",
   "published_at": "2026-07-14T14:27:53",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-06-16",
   "access_type": null,
   "license": null,
   "significance": "753B MoE, MIT, top open-weight on Artificial Analysis index",
   "organization_evidence": [
    "Four Chinese laboratories released downloadable, high-capability open-weight AI models between April 24 and mid-June 2026.",
    "Z.ai released **GLM-5.2 in mid-June**; the report describes it as a 753-billion-parameter mixture-of-experts model distributed under the MIT license."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Z.ai released **GLM-5.2 in mid-June**; the report describes it as a 753-billion-parameter mixture-of-experts model distributed under the MIT license.",
    "The reported sequence includes **DeepSeek V4** on April 24, **MiniMax M3** on June 1, Moonshot AI\u2019s Kimi K2.7-Code around June 13 and Z.ai\u2019s GLM-5.2 in mid-June 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "**GLM-5.2 (Z.ai)**753B MoE, MIT, top open-weight on Artificial Analysis index",
    "### The production line \u2014 spring 2026"
   ]
  },
  {
   "url": "https://getyourdozai.blogspot.com/2026/06/the-2026-ai-model-release-race-every.html",
   "title": "The 2026 AI Model Release Race: Every Major LLM Launch You Need to Know",
   "published_at": "2026-07-01T01:34:00",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "Open-weight MoE, 370B total params",
   "organization_evidence": [
    "The 2026 AI Model Release Race: Every Major LLM Launch You Need to Know",
    "- **Open-source surged:** DeepSeek V4-Pro (Apr 24), MiniMax M3 (Jun 1 \u2014 first open-weight triple-frontier model), GLM-5.2 (Jun 16), and Kimi K2.7 Code (Jun 12) all shipped.",
    "### April 2026 \u2014 The Foundation Wave",
    "| Date | Model | Lab | Significance |",
    "| Apr 24 | **DeepSeek V4-Pro/Flash** | DeepSeek | Open-weight MoE, 370B total params |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The 2026 AI Model Release Race: Every Major LLM Launch You Need to Know",
    "- **Open-source surged:** DeepSeek V4-Pro (Apr 24), MiniMax M3 (Jun 1 \u2014 first open-weight triple-frontier model), GLM-5.2 (Jun 16), and Kimi K2.7 Code (Jun 12) all shipped.",
    "### April 2026 \u2014 The Foundation Wave",
    "| Date | Model | Lab | Significance |",
    "| Apr 24 | **DeepSeek V4-Pro/Flash** | DeepSeek | Open-weight MoE, 370B total params |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The 2026 AI Model Release Race: Every Major LLM Launch You Need to Know",
    "- **Open-source surged:** DeepSeek V4-Pro (Apr 24), MiniMax M3 (Jun 1 \u2014 first open-weight triple-frontier model), GLM-5.2 (Jun 16), and Kimi K2.7 Code (Jun 12) all shipped.",
    "### April 2026 \u2014 The Foundation Wave",
    "| Date | Model | Lab | Significance |",
    "| Apr 24 | **DeepSeek V4-Pro/Flash** | DeepSeek | Open-weight MoE, 370B total params |"
   ]
  },
  {
   "url": "https://getyourdozai.blogspot.com/2026/06/the-2026-ai-model-release-race-every.html",
   "title": "The 2026 AI Model Release Race: Every Major LLM Launch You Need to Know",
   "published_at": "2026-07-01T01:34:00",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "First open-weight triple-frontier model",
   "organization_evidence": [
    "The 2026 AI Model Release Race: Every Major LLM Launch You Need to Know",
    "- **Open-source surged:** DeepSeek V4-Pro (Apr 24), MiniMax M3 (Jun 1 \u2014 first open-weight triple-frontier model), GLM-5.2 (Jun 16), and Kimi K2.7 Code (Jun 12) all shipped.",
    "### June 2026 \u2014 Open-Source Renaissance",
    "| Date | Model | Significance |",
    "| Jun 1 | **MiniMax M3** | First open-weight triple-frontier model |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The 2026 AI Model Release Race: Every Major LLM Launch You Need to Know",
    "- **Open-source surged:** DeepSeek V4-Pro (Apr 24), MiniMax M3 (Jun 1 \u2014 first open-weight triple-frontier model), GLM-5.2 (Jun 16), and Kimi K2.7 Code (Jun 12) all shipped.",
    "### June 2026 \u2014 Open-Source Renaissance",
    "| Date | Model | Significance |",
    "| Jun 1 | **MiniMax M3** | First open-weight triple-frontier model |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The 2026 AI Model Release Race: Every Major LLM Launch You Need to Know",
    "- **Open-source surged:** DeepSeek V4-Pro (Apr 24), MiniMax M3 (Jun 1 \u2014 first open-weight triple-frontier model), GLM-5.2 (Jun 16), and Kimi K2.7 Code (Jun 12) all shipped.",
    "### June 2026 \u2014 Open-Source Renaissance",
    "| Date | Model | Significance |",
    "| Jun 1 | **MiniMax M3** | First open-weight triple-frontier model |"
   ]
  },
  {
   "url": "https://getyourdozai.blogspot.com/2026/06/the-2026-ai-model-release-race-every.html",
   "title": "The 2026 AI Model Release Race: Every Major LLM Launch You Need to Know",
   "published_at": "2026-07-01T01:34:00",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-06-12",
   "access_type": null,
   "license": null,
   "significance": "Specialized coding model, open-weight",
   "organization_evidence": [
    "The 2026 AI Model Release Race: Every Major LLM Launch You Need to Know",
    "- **Open-source surged:** DeepSeek V4-Pro (Apr 24), MiniMax M3 (Jun 1 \u2014 first open-weight triple-frontier model), GLM-5.2 (Jun 16), and Kimi K2.7 Code (Jun 12) all shipped.",
    "### June 2026 \u2014 Open-Source Renaissance",
    "| Date | Model | Significance |",
    "| Jun 12 | **Kimi K2.7 Code** | Specialized coding model, open-weight |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The 2026 AI Model Release Race: Every Major LLM Launch You Need to Know",
    "- **Open-source surged:** DeepSeek V4-Pro (Apr 24), MiniMax M3 (Jun 1 \u2014 first open-weight triple-frontier model), GLM-5.2 (Jun 16), and Kimi K2.7 Code (Jun 12) all shipped.",
    "### June 2026 \u2014 Open-Source Renaissance",
    "| Date | Model | Significance |",
    "| Jun 12 | **Kimi K2.7 Code** | Specialized coding model, open-weight |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The 2026 AI Model Release Race: Every Major LLM Launch You Need to Know",
    "- **Open-source surged:** DeepSeek V4-Pro (Apr 24), MiniMax M3 (Jun 1 \u2014 first open-weight triple-frontier model), GLM-5.2 (Jun 16), and Kimi K2.7 Code (Jun 12) all shipped.",
    "### June 2026 \u2014 Open-Source Renaissance",
    "| Date | Model | Significance |",
    "| Jun 12 | **Kimi K2.7 Code** | Specialized coding model, open-weight |"
   ]
  },
  {
   "url": "https://getyourdozai.blogspot.com/2026/06/the-2026-ai-model-release-race-every.html",
   "title": "The 2026 AI Model Release Race: Every Major LLM Launch You Need to Know",
   "published_at": "2026-07-01T01:34:00",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-06-16",
   "access_type": null,
   "license": null,
   "significance": "Open-source, competitive with GPT-5.5",
   "organization_evidence": [
    "The 2026 AI Model Release Race: Every Major LLM Launch You Need to Know",
    "- **Open-source surged:** DeepSeek V4-Pro (Apr 24), MiniMax M3 (Jun 1 \u2014 first open-weight triple-frontier model), GLM-5.2 (Jun 16), and Kimi K2.7 Code (Jun 12) all shipped.",
    "### June 2026 \u2014 Open-Source Renaissance",
    "| Date | Model | Significance |",
    "| Jun 16 | **GLM-5.2** | Open-source, competitive with GPT-5.5 |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The 2026 AI Model Release Race: Every Major LLM Launch You Need to Know",
    "- **Open-source surged:** DeepSeek V4-Pro (Apr 24), MiniMax M3 (Jun 1 \u2014 first open-weight triple-frontier model), GLM-5.2 (Jun 16), and Kimi K2.7 Code (Jun 12) all shipped.",
    "### June 2026 \u2014 Open-Source Renaissance",
    "| Date | Model | Significance |",
    "| Jun 16 | **GLM-5.2** | Open-source, competitive with GPT-5.5 |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The 2026 AI Model Release Race: Every Major LLM Launch You Need to Know",
    "- **Open-source surged:** DeepSeek V4-Pro (Apr 24), MiniMax M3 (Jun 1 \u2014 first open-weight triple-frontier model), GLM-5.2 (Jun 16), and Kimi K2.7 Code (Jun 12) all shipped.",
    "### June 2026 \u2014 Open-Source Renaissance",
    "| Date | Model | Significance |",
    "| Jun 16 | **GLM-5.2** | Open-source, competitive with GPT-5.5 |"
   ]
  },
  {
   "url": "https://digitalapplied.com/blog/google-diffusiongemma-open-weight-text-diffusion-model-guide",
   "title": "Google DiffusionGemma: First Open-Weight Text Diffusion",
   "published_at": "2026-06-12T23:00:00",
   "organization": "Google DeepMind",
   "model": null,
   "release_date": "2026-06-10",
   "access_type": null,
   "license": null,
   "significance": "a 26B mixture-of-experts model that abandons left-to-right token generation for a parallel, block-by-block denoising process, reaching a vendor-stated 1,100-plus tokens per second",
   "organization_evidence": [
    "Google DiffusionGemma is Google DeepMind\u2019s first open-weight text diffusion model, released on June 10, 2026 under an Apache 2.0 license."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Google DiffusionGemma is Google DeepMind\u2019s first open-weight text diffusion model, released on June 10, 2026 under an Apache 2.0 license."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "It is a 26B mixture-of-experts model \u2014 roughly 3.8B active parameters per pass \u2014 that abandons left-to-right token generation for a parallel, block-by-block denoising process, reaching a vendor-stated 1,100-plus tokens per second on a single NVIDIA H100."
   ]
  },
  {
   "url": "https://agora-intelligence.com/en/blog/nova-inkling-open-weights-moe-2026",
   "title": "Thinking Machines Lab Releases Inkling: 975B Open-Weights MoE With Native Audio, Vision, and Apache 2.0 License",
   "published_at": "2026-07-16T05:39:14",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "975B-parameter Mixture-of-Experts model with 41B active parameters and 1 million token context window",
   "organization_evidence": [
    "**On July 15, 2026, [Thinking Machines Lab](https://thinkingmachines.ai) released Inkling under Apache 2.0, a [975B-parameter Mixture-of-Experts model with 41B active parameters](https://thinkingmachines.ai/news/introducing-inkling/)available in full on Hugging Face and deployable today across Databricks, Baseten, Modal, Fireworks, and TogetherAI.**"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "**On July 15, 2026, [Thinking Machines Lab](https://thinkingmachines.ai) released Inkling under Apache 2.0, a [975B-parameter Mixture-of-Experts model with 41B active parameters](https://thinkingmachines.ai/news/introducing-inkling/)available in full on Hugging Face and deployable today across Databricks, Baseten, Modal, Fireworks, and TogetherAI.**"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "**On July 15, 2026, [Thinking Machines Lab](https://thinkingmachines.ai) released Inkling under Apache 2.0, a [975B-parameter Mixture-of-Experts model with 41B active parameters](https://thinkingmachines.ai/news/introducing-inkling/)available in full on Hugging Face and deployable today across Databricks, Baseten, Modal, Fireworks, and TogetherAI.**",
    "The context window reaches [1 million tokens](https://thinkingmachines.ai/news/introducing-inkling/)trained on [45 trillion tokens of text, images, audio, and video](https://thinkingmachines.ai/news/introducing-inkling/) on NVIDIA GB300 NVL72 systems."
   ]
  },
  {
   "url": "https://metirai.com/blog/summer-2026-frontier-model-release-flood-multi-model",
   "title": "Summer 2026's Frontier AI Model Release Flood, Explained",
   "published_at": "2026-07-26T00:00:00",
   "organization": "Moonshot AI",
   "model": "Kimi K3",
   "release_date": "2026-07-16",
   "access_type": "open weights",
   "license": null,
   "significance": "a 2.8 trillion parameter model that immediately ranked third on the Artificial Analysis Intelligence Index",
   "organization_evidence": [
    "# Summer 2026's Frontier AI Model Release Flood, Explained",
    "Every major model launch from June 30 to July 27, 2026. Colour marks whether the weights are open or stay closed behind an API.",
    "4. July 16, 2026Open weights",
    "Moonshot AI brings a 2.8 trillion parameter model online through its API, ranking third on the Artificial Analysis Intelligence Index."
   ],
   "model_evidence": [
    "# Summer 2026's Frontier AI Model Release Flood, Explained",
    "Every major model launch from June 30 to July 27, 2026. Colour marks whether the weights are open or stay closed behind an API.",
    "4. July 16, 2026Open weights",
    "Moonshot AI brings a 2.8 trillion parameter model online through its API, ranking third on the Artificial Analysis Intelligence Index."
   ],
   "release_date_evidence": [
    "# Summer 2026's Frontier AI Model Release Flood, Explained",
    "Every major model launch from June 30 to July 27, 2026. Colour marks whether the weights are open or stay closed behind an API.",
    "4. July 16, 2026Open weights",
    "Moonshot AI brings a 2.8 trillion parameter model online through its API, ranking third on the Artificial Analysis Intelligence Index."
   ],
   "access_type_evidence": [
    "# Summer 2026's Frontier AI Model Release Flood, Explained",
    "Every major model launch from June 30 to July 27, 2026. Colour marks whether the weights are open or stay closed behind an API.",
    "4. July 16, 2026Open weights",
    "Moonshot AI brings a 2.8 trillion parameter model online through its API, ranking third on the Artificial Analysis Intelligence Index."
   ],
   "license_evidence": null,
   "significance_evidence": [
    "# Summer 2026's Frontier AI Model Release Flood, Explained",
    "Seven frontier models shipped in four weeks this summer, from Sonnet 5 to Kimi K3. A neutral look at near-parity, falling prices, and what it means for model choice.",
    "Moonshot AI brought Kimi K3 online through its API on July 16, a 2.8 trillion parameter model that immediately ranked third on the Artificial Analysis Intelligence Index."
   ]
  },
  {
   "url": "https://metirai.com/blog/summer-2026-frontier-model-release-flood-multi-model",
   "title": "Summer 2026's Frontier AI Model Release Flood, Explained",
   "published_at": "2026-07-26T00:00:00",
   "organization": "Moonshot AI",
   "model": "Kimi K3",
   "release_date": "2026-07-27",
   "access_type": "open weights",
   "license": null,
   "significance": "the largest open-weight release published to date",
   "organization_evidence": [
    "# Summer 2026's Frontier AI Model Release Flood, Explained",
    "Every major model launch from June 30 to July 27, 2026. Colour marks whether the weights are open or stay closed behind an API.",
    "7. July 27, 2026Open weights",
    "Kimi K3 open weights",
    "Moonshot publishes the full 2.8 trillion parameter weights, the largest open-weight release to date."
   ],
   "model_evidence": [
    "# Summer 2026's Frontier AI Model Release Flood, Explained",
    "Every major model launch from June 30 to July 27, 2026. Colour marks whether the weights are open or stay closed behind an API.",
    "7. July 27, 2026Open weights",
    "Kimi K3 open weights",
    "Moonshot publishes the full 2.8 trillion parameter weights, the largest open-weight release to date."
   ],
   "release_date_evidence": [
    "# Summer 2026's Frontier AI Model Release Flood, Explained",
    "Every major model launch from June 30 to July 27, 2026. Colour marks whether the weights are open or stay closed behind an API.",
    "7. July 27, 2026Open weights",
    "Kimi K3 open weights",
    "Moonshot publishes the full 2.8 trillion parameter weights, the largest open-weight release to date."
   ],
   "access_type_evidence": [
    "# Summer 2026's Frontier AI Model Release Flood, Explained",
    "Every major model launch from June 30 to July 27, 2026. Colour marks whether the weights are open or stay closed behind an API.",
    "7. July 27, 2026Open weights",
    "Kimi K3 open weights",
    "Moonshot publishes the full 2.8 trillion parameter weights, the largest open-weight release to date."
   ],
   "license_evidence": null,
   "significance_evidence": [
    "# Summer 2026's Frontier AI Model Release Flood, Explained",
    "Seven frontier models shipped in four weeks this summer, from Sonnet 5 to Kimi K3. A neutral look at near-parity, falling prices, and what it means for model choice.",
    "Kimi K3's full weights followed on July 27, the largest open-weight release published to date."
   ]
  },
  {
   "url": "https://metirai.com/blog/summer-2026-frontier-model-release-flood-multi-model",
   "title": "Summer 2026's Frontier AI Model Release Flood, Explained",
   "published_at": "2026-07-26T00:00:00",
   "organization": "DeepSeek",
   "model": "DeepSeek V4",
   "release_date": "2026-07-24",
   "access_type": "open weight",
   "license": null,
   "significance": "strongest configuration reportedly scored around 80.6 percent on SWE-bench Verified",
   "organization_evidence": [
    "# Summer 2026's Frontier AI Model Release Flood, Explained",
    "Between June 30 and July 27, 2026, seven frontier or open-weight model families shipped from six different labs.",
    "Then, on July 24, Anthropic shipped Claude Opus 5, its fourth Claude 5-generation model release in under two months, the same day ![](/logos/deepseek.png)DeepSeek V4 completed its move to general availability."
   ],
   "model_evidence": [
    "# Summer 2026's Frontier AI Model Release Flood, Explained",
    "Between June 30 and July 27, 2026, seven frontier or open-weight model families shipped from six different labs.",
    "Then, on July 24, Anthropic shipped Claude Opus 5, its fourth Claude 5-generation model release in under two months, the same day ![](/logos/deepseek.png)DeepSeek V4 completed its move to general availability."
   ],
   "release_date_evidence": [
    "# Summer 2026's Frontier AI Model Release Flood, Explained",
    "Between June 30 and July 27, 2026, seven frontier or open-weight model families shipped from six different labs.",
    "Then, on July 24, Anthropic shipped Claude Opus 5, its fourth Claude 5-generation model release in under two months, the same day ![](/logos/deepseek.png)DeepSeek V4 completed its move to general availability."
   ],
   "access_type_evidence": [
    "# Summer 2026's Frontier AI Model Release Flood, Explained",
    "Between June 30 and July 27, 2026, seven frontier or open-weight model families shipped from six different labs.",
    "Then, on July 24, Anthropic shipped Claude Opus 5, its fourth Claude 5-generation model release in under two months, the same day ![](/logos/deepseek.png)DeepSeek V4 completed its move to general availability."
   ],
   "license_evidence": null,
   "significance_evidence": [
    "# Summer 2026's Frontier AI Model Release Flood, Explained",
    "Open and closed weights are converging",
    "DeepSeek V4's strongest configuration reportedly scored around 80.6 percent on SWE-bench Verified, a test of resolving real GitHub issues, putting an openly downloadable model within a fraction of a point of the leading closed systems on a hard, practical coding benchmark."
   ]
  },
  {
   "url": "https://singularitybyte.com/news/open-weight-wave-june-2026.html",
   "title": "The June 2026 Open-Weight Wave: 16 Models in One Week",
   "published_at": "2026-06-26T14:00:00",
   "organization": "NVIDIA",
   "model": null,
   "release_date": "2026-06-04",
   "access_type": null,
   "license": null,
   "significance": "550B hybrid Mamba-MoE, 1M context, weights plus data and recipes",
   "organization_evidence": [
    "The June 2026 Open-Weight Wave: 16 Models in One Week",
    "The standout releases",
    "| Model | Org | Modality | License | Why it matters |",
    "| [Nemotron 3 Ultra](/models/nvidia-nemotron-3-ultra.html) | NVIDIA (US) | LLM / agents | OpenMDW-1.1 | 550B hybrid Mamba-MoE, 1M context, weights plus data and recipes |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The June 2026 Open-Weight Wave: 16 Models in One Week",
    "What actually happened",
    "NVIDIA's Nemotron 3 Ultra dropped June 4."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The June 2026 Open-Weight Wave: 16 Models in One Week",
    "The standout releases",
    "| Model | Org | Modality | License | Why it matters |",
    "| [Nemotron 3 Ultra](/models/nvidia-nemotron-3-ultra.html) | NVIDIA (US) | LLM / agents | OpenMDW-1.1 | 550B hybrid Mamba-MoE, 1M context, weights plus data and recipes |"
   ]
  },
  {
   "url": "https://singularitybyte.com/news/open-weight-wave-june-2026.html",
   "title": "The June 2026 Open-Weight Wave: 16 Models in One Week",
   "published_at": "2026-06-26T14:00:00",
   "organization": "Google",
   "model": "Gemma 4 12B Unified",
   "release_date": "2026-06-03",
   "access_type": "open weight",
   "license": "Apache 2.0",
   "significance": "Text, image, audio, video in one 12B model that runs in ~16GB",
   "organization_evidence": [
    "The June 2026 Open-Weight Wave: 16 Models in One Week",
    "The standout releases",
    "| Model | Org | Modality | License | Why it matters |",
    "| Gemma 4 12B Unified | Google (US) | Multimodal LLM | Apache 2.0 | Text, image, audio, video in one 12B model that runs in ~16GB |"
   ],
   "model_evidence": [
    "The June 2026 Open-Weight Wave: 16 Models in One Week",
    "The standout releases",
    "| Model | Org | Modality | License | Why it matters |",
    "| Gemma 4 12B Unified | Google (US) | Multimodal LLM | Apache 2.0 | Text, image, audio, video in one 12B model that runs in ~16GB |"
   ],
   "release_date_evidence": [
    "The June 2026 Open-Weight Wave: 16 Models in One Week",
    "What actually happened",
    "[Gemma 4 12B Unified](/models/google-gemma-4.html) and Ideogram 4 landed June 3."
   ],
   "access_type_evidence": [
    "The June 2026 Open-Weight Wave: 16 Models in One Week",
    "The first week of June 2026 was the densest stretch of open-weight releases anyone has tracked: frontier models across language, image, audio, video, and 3D, mostly under permissive licenses.",
    "The standout releases",
    "| Model | Org | Modality | License | Why it matters |",
    "| Gemma 4 12B Unified | Google (US) | Multimodal LLM | Apache 2.0 | Text, image, audio, video in one 12B model that runs in ~16GB |"
   ],
   "license_evidence": [
    "The June 2026 Open-Weight Wave: 16 Models in One Week",
    "The standout releases",
    "| Model | Org | Modality | License | Why it matters |",
    "| Gemma 4 12B Unified | Google (US) | Multimodal LLM | Apache 2.0 | Text, image, audio, video in one 12B model that runs in ~16GB |"
   ],
   "significance_evidence": [
    "The June 2026 Open-Weight Wave: 16 Models in One Week",
    "The standout releases",
    "| Model | Org | Modality | License | Why it matters |",
    "| Gemma 4 12B Unified | Google (US) | Multimodal LLM | Apache 2.0 | Text, image, audio, video in one 12B model that runs in ~16GB |"
   ]
  },
  {
   "url": "https://singularitybyte.com/news/open-weight-wave-june-2026.html",
   "title": "The June 2026 Open-Weight Wave: 16 Models in One Week",
   "published_at": "2026-06-26T14:00:00",
   "organization": "MiniMax",
   "model": "MiniMax M3",
   "release_date": "2026-06-01",
   "access_type": "open weight",
   "license": "open",
   "significance": "428B / 23B active, 1M context, frontier coding and computer use",
   "organization_evidence": [
    "The June 2026 Open-Weight Wave: 16 Models in One Week",
    "The standout releases",
    "| Model | Org | Modality | License | Why it matters |",
    "| MiniMax M3 | MiniMax (CN) | Multimodal LLM | open | 428B / 23B active, 1M context, frontier coding and computer use |"
   ],
   "model_evidence": [
    "The June 2026 Open-Weight Wave: 16 Models in One Week",
    "The standout releases",
    "| Model | Org | Modality | License | Why it matters |",
    "| MiniMax M3 | MiniMax (CN) | Multimodal LLM | open | 428B / 23B active, 1M context, frontier coding and computer use |"
   ],
   "release_date_evidence": [
    "The June 2026 Open-Weight Wave: 16 Models in One Week",
    "What actually happened",
    "[MiniMax M3](/models/minimax-m3.html) and ByteDance's Bernini-R video model opened the week on June 1."
   ],
   "access_type_evidence": [
    "The June 2026 Open-Weight Wave: 16 Models in One Week",
    "The first week of June 2026 was the densest stretch of open-weight releases anyone has tracked: frontier models across language, image, audio, video, and 3D, mostly under permissive licenses.",
    "The standout releases",
    "| Model | Org | Modality | License | Why it matters |",
    "| MiniMax M3 | MiniMax (CN) | Multimodal LLM | open | 428B / 23B active, 1M context, frontier coding and computer use |"
   ],
   "license_evidence": [
    "The June 2026 Open-Weight Wave: 16 Models in One Week",
    "The standout releases",
    "| Model | Org | Modality | License | Why it matters |",
    "| MiniMax M3 | MiniMax (CN) | Multimodal LLM | open | 428B / 23B active, 1M context, frontier coding and computer use |"
   ],
   "significance_evidence": [
    "The June 2026 Open-Weight Wave: 16 Models in One Week",
    "The standout releases",
    "| Model | Org | Modality | License | Why it matters |",
    "| MiniMax M3 | MiniMax (CN) | Multimodal LLM | open | 428B / 23B active, 1M context, frontier coding and computer use |"
   ]
  },
  {
   "url": "https://singularitybyte.com/news/open-weight-wave-june-2026.html",
   "title": "The June 2026 Open-Weight Wave: 16 Models in One Week",
   "published_at": "2026-06-26T14:00:00",
   "organization": "StepFun",
   "model": "Step-3.7-Flash",
   "release_date": "2026-06",
   "access_type": "open weight",
   "license": "Apache 2.0",
   "significance": "198B / 11B active sparse MoE, 256K context",
   "organization_evidence": [
    "The June 2026 Open-Weight Wave: 16 Models in One Week",
    "The standout releases",
    "| Model | Org | Modality | License | Why it matters |",
    "| Step-3.7-Flash | StepFun (CN) | Vision-LLM | Apache 2.0 | 198B / 11B active sparse MoE, 256K context |"
   ],
   "model_evidence": [
    "The June 2026 Open-Weight Wave: 16 Models in One Week",
    "The standout releases",
    "| Model | Org | Modality | License | Why it matters |",
    "| Step-3.7-Flash | StepFun (CN) | Vision-LLM | Apache 2.0 | 198B / 11B active sparse MoE, 256K context |"
   ],
   "release_date_evidence": [
    "The June 2026 Open-Weight Wave: 16 Models in One Week",
    "What actually happened",
    "Between roughly June 1 and June 8, a cluster of labs shipped at once."
   ],
   "access_type_evidence": [
    "The June 2026 Open-Weight Wave: 16 Models in One Week",
    "The first week of June 2026 was the densest stretch of open-weight releases anyone has tracked: frontier models across language, image, audio, video, and 3D, mostly under permissive licenses.",
    "The standout releases",
    "| Model | Org | Modality | License | Why it matters |",
    "| Step-3.7-Flash | StepFun (CN) | Vision-LLM | Apache 2.0 | 198B / 11B active sparse MoE, 256K context |"
   ],
   "license_evidence": [
    "The June 2026 Open-Weight Wave: 16 Models in One Week",
    "The standout releases",
    "| Model | Org | Modality | License | Why it matters |",
    "| Step-3.7-Flash | StepFun (CN) | Vision-LLM | Apache 2.0 | 198B / 11B active sparse MoE, 256K context |"
   ],
   "significance_evidence": [
    "The June 2026 Open-Weight Wave: 16 Models in One Week",
    "The standout releases",
    "| Model | Org | Modality | License | Why it matters |",
    "| Step-3.7-Flash | StepFun (CN) | Vision-LLM | Apache 2.0 | 198B / 11B active sparse MoE, 256K context |"
   ]
  },
  {
   "url": "https://singularitybyte.com/news/open-weight-wave-june-2026.html",
   "title": "The June 2026 Open-Weight Wave: 16 Models in One Week",
   "published_at": "2026-06-26T14:00:00",
   "organization": "JetBrains",
   "model": "Mellum2-12B",
   "release_date": "2026-06",
   "access_type": "open weight",
   "license": "Apache 2.0",
   "significance": "JetBrains' first open MoE, tuned for code completion",
   "organization_evidence": [
    "The June 2026 Open-Weight Wave: 16 Models in One Week",
    "The standout releases",
    "| Model | Org | Modality | License | Why it matters |",
    "| Mellum2-12B | JetBrains (EU) | Coding LLM | Apache 2.0 | JetBrains' first open MoE, tuned for code completion |"
   ],
   "model_evidence": [
    "The June 2026 Open-Weight Wave: 16 Models in One Week",
    "The standout releases",
    "| Model | Org | Modality | License | Why it matters |",
    "| Mellum2-12B | JetBrains (EU) | Coding LLM | Apache 2.0 | JetBrains' first open MoE, tuned for code completion |"
   ],
   "release_date_evidence": [
    "The June 2026 Open-Weight Wave: 16 Models in One Week",
    "What actually happened",
    "Between roughly June 1 and June 8, a cluster of labs shipped at once."
   ],
   "access_type_evidence": [
    "The June 2026 Open-Weight Wave: 16 Models in One Week",
    "The first week of June 2026 was the densest stretch of open-weight releases anyone has tracked: frontier models across language, image, audio, video, and 3D, mostly under permissive licenses.",
    "The standout releases",
    "| Model | Org | Modality | License | Why it matters |",
    "| Mellum2-12B | JetBrains (EU) | Coding LLM | Apache 2.0 | JetBrains' first open MoE, tuned for code completion |"
   ],
   "license_evidence": [
    "The June 2026 Open-Weight Wave: 16 Models in One Week",
    "The standout releases",
    "| Model | Org | Modality | License | Why it matters |",
    "| Mellum2-12B | JetBrains (EU) | Coding LLM | Apache 2.0 | JetBrains' first open MoE, tuned for code completion |"
   ],
   "significance_evidence": [
    "The June 2026 Open-Weight Wave: 16 Models in One Week",
    "The standout releases",
    "| Model | Org | Modality | License | Why it matters |",
    "| Mellum2-12B | JetBrains (EU) | Coding LLM | Apache 2.0 | JetBrains' first open MoE, tuned for code completion |"
   ]
  },
  {
   "url": "https://singularitybyte.com/news/open-weight-wave-june-2026.html",
   "title": "The June 2026 Open-Weight Wave: 16 Models in One Week",
   "published_at": "2026-06-26T14:00:00",
   "organization": "Baidu",
   "model": "PaddleOCR-VL-1.6",
   "release_date": "2026-06",
   "access_type": "open weight",
   "license": "Apache 2.0",
   "significance": "~1B model that beats document parsers 10x its size",
   "organization_evidence": [
    "The June 2026 Open-Weight Wave: 16 Models in One Week",
    "The standout releases",
    "| Model | Org | Modality | License | Why it matters |",
    "| PaddleOCR-VL-1.6 | Baidu (CN) | Document VLM | Apache 2.0 | ~1B model that beats document parsers 10x its size |"
   ],
   "model_evidence": [
    "The June 2026 Open-Weight Wave: 16 Models in One Week",
    "The standout releases",
    "| Model | Org | Modality | License | Why it matters |",
    "| PaddleOCR-VL-1.6 | Baidu (CN) | Document VLM | Apache 2.0 | ~1B model that beats document parsers 10x its size |"
   ],
   "release_date_evidence": [
    "The June 2026 Open-Weight Wave: 16 Models in One Week",
    "What actually happened",
    "Between roughly June 1 and June 8, a cluster of labs shipped at once."
   ],
   "access_type_evidence": [
    "The June 2026 Open-Weight Wave: 16 Models in One Week",
    "The first week of June 2026 was the densest stretch of open-weight releases anyone has tracked: frontier models across language, image, audio, video, and 3D, mostly under permissive licenses.",
    "The standout releases",
    "| Model | Org | Modality | License | Why it matters |",
    "| PaddleOCR-VL-1.6 | Baidu (CN) | Document VLM | Apache 2.0 | ~1B model that beats document parsers 10x its size |"
   ],
   "license_evidence": [
    "The June 2026 Open-Weight Wave: 16 Models in One Week",
    "The standout releases",
    "| Model | Org | Modality | License | Why it matters |",
    "| PaddleOCR-VL-1.6 | Baidu (CN) | Document VLM | Apache 2.0 | ~1B model that beats document parsers 10x its size |"
   ],
   "significance_evidence": [
    "The June 2026 Open-Weight Wave: 16 Models in One Week",
    "The standout releases",
    "| Model | Org | Modality | License | Why it matters |",
    "| PaddleOCR-VL-1.6 | Baidu (CN) | Document VLM | Apache 2.0 | ~1B model that beats document parsers 10x its size |"
   ]
  },
  {
   "url": "https://singularitybyte.com/news/open-weight-wave-june-2026.html",
   "title": "The June 2026 Open-Weight Wave: 16 Models in One Week",
   "published_at": "2026-06-26T14:00:00",
   "organization": "H Company",
   "model": "Holo-3.1-4B",
   "release_date": "2026-06",
   "access_type": "open weight",
   "license": "Apache 2.0",
   "significance": "4B agent for web, desktop, and mobile automation",
   "organization_evidence": [
    "The June 2026 Open-Weight Wave: 16 Models in One Week",
    "The standout releases",
    "| Model | Org | Modality | License | Why it matters |",
    "| Holo-3.1-4B | H Company (EU) | Computer-use VLM | Apache 2.0 | 4B agent for web, desktop, and mobile automation |"
   ],
   "model_evidence": [
    "The June 2026 Open-Weight Wave: 16 Models in One Week",
    "The standout releases",
    "| Model | Org | Modality | License | Why it matters |",
    "| Holo-3.1-4B | H Company (EU) | Computer-use VLM | Apache 2.0 | 4B agent for web, desktop, and mobile automation |"
   ],
   "release_date_evidence": [
    "The June 2026 Open-Weight Wave: 16 Models in One Week",
    "What actually happened",
    "Between roughly June 1 and June 8, a cluster of labs shipped at once."
   ],
   "access_type_evidence": [
    "The June 2026 Open-Weight Wave: 16 Models in One Week",
    "The first week of June 2026 was the densest stretch of open-weight releases anyone has tracked: frontier models across language, image, audio, video, and 3D, mostly under permissive licenses.",
    "The standout releases",
    "| Model | Org | Modality | License | Why it matters |",
    "| Holo-3.1-4B | H Company (EU) | Computer-use VLM | Apache 2.0 | 4B agent for web, desktop, and mobile automation |"
   ],
   "license_evidence": [
    "The June 2026 Open-Weight Wave: 16 Models in One Week",
    "The standout releases",
    "| Model | Org | Modality | License | Why it matters |",
    "| Holo-3.1-4B | H Company (EU) | Computer-use VLM | Apache 2.0 | 4B agent for web, desktop, and mobile automation |"
   ],
   "significance_evidence": [
    "The June 2026 Open-Weight Wave: 16 Models in One Week",
    "The standout releases",
    "| Model | Org | Modality | License | Why it matters |",
    "| Holo-3.1-4B | H Company (EU) | Computer-use VLM | Apache 2.0 | 4B agent for web, desktop, and mobile automation |"
   ]
  },
  {
   "url": "https://singularitybyte.com/news/open-weight-wave-june-2026.html",
   "title": "The June 2026 Open-Weight Wave: 16 Models in One Week",
   "published_at": "2026-06-26T14:00:00",
   "organization": "Z.ai",
   "model": "GLM-5.2",
   "release_date": "2026-06-17",
   "access_type": "open weight",
   "license": "MIT",
   "significance": "took the top of Artificial Analysis's open-weights Intelligence Index, above every US open model",
   "organization_evidence": [
    "The June 2026 Open-Weight Wave: 16 Models in One Week",
    "China leads intelligence and volume, the US counters on efficiency",
    "The mid-June coda made it stark: Z.ai's [GLM-5.2](/models/glm-5-2.html), released June 17 under MIT, took the top of Artificial Analysis's open-weights Intelligence Index, above every US open model."
   ],
   "model_evidence": [
    "The June 2026 Open-Weight Wave: 16 Models in One Week",
    "China leads intelligence and volume, the US counters on efficiency",
    "The mid-June coda made it stark: Z.ai's [GLM-5.2](/models/glm-5-2.html), released June 17 under MIT, took the top of Artificial Analysis's open-weights Intelligence Index, above every US open model."
   ],
   "release_date_evidence": [
    "The June 2026 Open-Weight Wave: 16 Models in One Week",
    "China leads intelligence and volume, the US counters on efficiency",
    "The mid-June coda made it stark: Z.ai's [GLM-5.2](/models/glm-5-2.html), released June 17 under MIT, took the top of Artificial Analysis's open-weights Intelligence Index, above every US open model."
   ],
   "access_type_evidence": [
    "The June 2026 Open-Weight Wave: 16 Models in One Week",
    "The first week of June 2026 was the densest stretch of open-weight releases anyone has tracked: frontier models across language, image, audio, video, and 3D, mostly under permissive licenses.",
    "China leads intelligence and volume, the US counters on efficiency",
    "The mid-June coda made it stark: Z.ai's [GLM-5.2](/models/glm-5-2.html), released June 17 under MIT, took the top of Artificial Analysis's open-weights Intelligence Index, above every US open model."
   ],
   "license_evidence": [
    "The June 2026 Open-Weight Wave: 16 Models in One Week",
    "China leads intelligence and volume, the US counters on efficiency",
    "The mid-June coda made it stark: Z.ai's [GLM-5.2](/models/glm-5-2.html), released June 17 under MIT, took the top of Artificial Analysis's open-weights Intelligence Index, above every US open model."
   ],
   "significance_evidence": [
    "The June 2026 Open-Weight Wave: 16 Models in One Week",
    "China leads intelligence and volume, the US counters on efficiency",
    "The mid-June coda made it stark: Z.ai's [GLM-5.2](/models/glm-5-2.html), released June 17 under MIT, took the top of Artificial Analysis's open-weights Intelligence Index, above every US open model."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/mimo-v2-5-pro-vs-kimi-k2-6",
   "title": "MiMo-V2.5-Pro vs Kimi K2.6: Model Comparison | Artificial Analysis",
   "published_at": null,
   "organization": "Xiaomi",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "1023B, 42B active at inference time",
   "organization_evidence": [
    "# MiMo-V2.5-Pro vs. Kimi K2.6",
    "| | ![Xiaomi logo](/img/logos/xiaomi_small.svg)Xiaomi MiMo-V2.5-Pro | ![Kimi logo](/_next/image?url=%2Fimg%2Flogos%2Fkimi.jpg&w=32&q=75)Kimi Kimi K2.6 | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# MiMo-V2.5-Pro vs. Kimi K2.6",
    "| Release Date | April 2026 | April 2026 | MiMo-V2.5-Pro has a more recent release date than Kimi K2.6 |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# MiMo-V2.5-Pro vs. Kimi K2.6",
    "| Parameters | 1023B, 42B active at inference time | 1000B, 32B active at inference time | MiMo-V2.5-Pro has more parameters than Kimi K2.6 |"
   ]
  },
  {
   "url": "https://edenai.co/post/thinking-machines-inkling-open-weights-ai-model-guide",
   "title": "Open-Weights AI Model Guide 2026",
   "published_at": "2026-07-28T07:52:12",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-31",
   "access_type": null,
   "license": null,
   "significance": "975 billion parameter MoE model that activates 41 billion parameters per request and scores 97.1% on AIME 2026",
   "organization_evidence": [
    "Summarize this article with:",
    "- **Inkling** is Thinking Machines Lab's first open-weights model: a 975 billion parameter MoE (Mixture-of-Experts, where only a fraction of parameters activate per request) that activates 41B parameters at inference time."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Summarize this article with:",
    "Released under Apache 2.0 in July 2026, it competes with frontier proprietary models on reasoning benchmarks while remaining fully open for self-hosting and fine-tuning."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Summarize this article with:",
    "- **Inkling** is Thinking Machines Lab's first open-weights model: a 975 billion parameter MoE (Mixture-of-Experts, where only a fraction of parameters activate per request) that activates 41B parameters at inference time.",
    "- It scores **97.1% on AIME 2026**, **87.2% on GPQA Diamond**, and **46.0% on HLE with tools**, placing it among the top open-weight models globally."
   ]
  },
  {
   "url": "https://computingforgeeks.com/open-source-llm-comparison",
   "title": "Open Source LLM Comparison Table (2026) | ComputingForGeeks",
   "published_at": "2026-03-28T07:53:41",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-26",
   "access_type": null,
   "license": null,
   "significance": "largest open model ever released: 2.8T total parameters, 104B active from 896 experts, native vision, video input, and a 1M token context",
   "organization_evidence": [
    "This reference covers every major open-source and open-weight large language model, with verified benchmark scores, license terms, hardware requirements, and hands-on performance data from real self-hosting tests.",
    "## Master Comparison Table",
    "| Model | Developer | Total Params | Active Params | Architecture | Context Window | Multimodal | License | Release |",
    "| **Kimi K3** | Moonshot AI | 2.8T | 104B | MoE (896 experts) | 1M tokens | Text + Image + Video | Kimi K3 (custom) | Jul 2026 |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "This reference covers every major open-source and open-weight large language model, with verified benchmark scores, license terms, hardware requirements, and hands-on performance data from real self-hosting tests.",
    "## Kimi K2 and K3 (Moonshot AI)",
    "Kimi K3, whose weights went public on July 26, 2026, is the largest open model ever released: 2.8T total parameters, 104B active from 896 experts, native vision, video input, and a 1M token context."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "This reference covers every major open-source and open-weight large language model, with verified benchmark scores, license terms, hardware requirements, and hands-on performance data from real self-hosting tests.",
    "## Kimi K2 and K3 (Moonshot AI)",
    "Kimi K3, whose weights went public on July 26, 2026, is the largest open model ever released: 2.8T total parameters, 104B active from 896 experts, native vision, video input, and a 1M token context."
   ]
  },
  {
   "url": "https://computingforgeeks.com/open-source-llm-comparison",
   "title": "Open Source LLM Comparison Table (2026) | ComputingForGeeks",
   "published_at": "2026-03-28T07:53:41",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "V4 Pro at 1.6T total parameters with 49B active, and V4 Flash at 284B total with 13B active",
   "organization_evidence": [
    "This reference covers every major open-source and open-weight large language model, with verified benchmark scores, license terms, hardware requirements, and hands-on performance data from real self-hosting tests.",
    "## Master Comparison Table",
    "| Model | Developer | Total Params | Active Params | Architecture | Context Window | Multimodal | License | Release |",
    "| **DeepSeek V4 Pro** | DeepSeek | 1.6T | 49B | MoE + sparse attention | 1M tokens | No | MIT | Apr 2026 |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "This reference covers every major open-source and open-weight large language model, with verified benchmark scores, license terms, hardware requirements, and hands-on performance data from real self-hosting tests.",
    "## DeepSeek V4, V3.2, and R1",
    "DeepSeek V4 arrived in April 2026 as two MIT-licensed models sharing a sparse-attention architecture (token-wise compression plus DeepSeek Sparse Attention) and a 1M token default context: V4 Pro at 1.6T total parameters with 49B active, and V4 Flash at 284B total with 13B active."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "This reference covers every major open-source and open-weight large language model, with verified benchmark scores, license terms, hardware requirements, and hands-on performance data from real self-hosting tests.",
    "## DeepSeek V4, V3.2, and R1",
    "DeepSeek V4 arrived in April 2026 as two MIT-licensed models sharing a sparse-attention architecture (token-wise compression plus DeepSeek Sparse Attention) and a 1M token default context: V4 Pro at 1.6T total parameters with 49B active, and V4 Flash at 284B total with 13B active."
   ]
  },
  {
   "url": "https://computingforgeeks.com/open-source-llm-comparison",
   "title": "Open Source LLM Comparison Table (2026) | ComputingForGeeks",
   "published_at": "2026-03-28T07:53:41",
   "organization": "Z.ai (Zhipu)",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "pushes to 753B total with roughly 40B active, a 1M token context, and the strongest open coding profile of any model on this page",
   "organization_evidence": [
    "This reference covers every major open-source and open-weight large language model, with verified benchmark scores, license terms, hardware requirements, and hands-on performance data from real self-hosting tests.",
    "## Master Comparison Table",
    "| Model | Developer | Total Params | Active Params | Architecture | Context Window | Multimodal | License | Release |",
    "| **GLM-5.2** | Z.ai (Zhipu) | 753B | ~40B | MoE | 1M tokens | No | MIT | Jun 2026 |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "This reference covers every major open-source and open-weight large language model, with verified benchmark scores, license terms, hardware requirements, and hands-on performance data from real self-hosting tests.",
    "## GLM-5 and GLM-5.2 (Z.ai)",
    "GLM-5.2, launched June 13, 2026 with weights on Hugging Face days later, pushes to 753B total with roughly 40B active, a 1M token context, and the strongest open coding profile of any model on this page: 62.1 on SWE-bench Pro, 81.0 on Terminal-Bench 2.1, 99.2 on AIME 2026, and 91.2 on GPQA Diamond."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "This reference covers every major open-source and open-weight large language model, with verified benchmark scores, license terms, hardware requirements, and hands-on performance data from real self-hosting tests.",
    "## GLM-5 and GLM-5.2 (Z.ai)",
    "GLM-5.2, launched June 13, 2026 with weights on Hugging Face days later, pushes to 753B total with roughly 40B active, a 1M token context, and the strongest open coding profile of any model on this page: 62.1 on SWE-bench Pro, 81.0 on Terminal-Bench 2.1, 99.2 on AIME 2026, and 91.2 on GPQA Diamond."
   ]
  },
  {
   "url": "https://computingforgeeks.com/open-source-llm-comparison",
   "title": "Open Source LLM Comparison Table (2026) | ComputingForGeeks",
   "published_at": "2026-03-28T07:53:41",
   "organization": "Z.ai (Zhipu)",
   "model": null,
   "release_date": "2026-02",
   "access_type": null,
   "license": null,
   "significance": "744B total parameters, 40B active, MIT license, trained entirely on 100,000 Huawei Ascend 910B chips with no US-manufactured hardware",
   "organization_evidence": [
    "This reference covers every major open-source and open-weight large language model, with verified benchmark scores, license terms, hardware requirements, and hands-on performance data from real self-hosting tests.",
    "## Master Comparison Table",
    "| Model | Developer | Total Params | Active Params | Architecture | Context Window | Multimodal | License | Release |",
    "| **GLM-5** | Z.ai (Zhipu) | 744B | 40B | MoE | 205K tokens | No | MIT | Feb 2026 |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "This reference covers every major open-source and open-weight large language model, with verified benchmark scores, license terms, hardware requirements, and hands-on performance data from real self-hosting tests.",
    "## GLM-5 and GLM-5.2 (Z.ai)",
    "Z.ai (formerly Zhipu AI) released GLM-5 in February 2026: 744B total parameters, 40B active, MIT license, trained entirely on 100,000 Huawei Ascend 910B chips with no US-manufactured hardware."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "This reference covers every major open-source and open-weight large language model, with verified benchmark scores, license terms, hardware requirements, and hands-on performance data from real self-hosting tests.",
    "## GLM-5 and GLM-5.2 (Z.ai)",
    "Z.ai (formerly Zhipu AI) released GLM-5 in February 2026: 744B total parameters, 40B active, MIT license, trained entirely on 100,000 Huawei Ascend 910B chips with no US-manufactured hardware."
   ]
  },
  {
   "url": "https://computingforgeeks.com/open-source-llm-comparison",
   "title": "Open Source LLM Comparison Table (2026) | ComputingForGeeks",
   "published_at": "2026-03-28T07:53:41",
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "hybrid attention design (Gated DeltaNet linear attention mixed with standard self-attention) with a native 262K context extensible to roughly 1M tokens via YaRN scaling, and both accept text, images, and video",
   "organization_evidence": [
    "This reference covers every major open-source and open-weight large language model, with verified benchmark scores, license terms, hardware requirements, and hands-on performance data from real self-hosting tests.",
    "## Master Comparison Table",
    "| Model | Developer | Total Params | Active Params | Architecture | Context Window | Multimodal | License | Release |",
    "| **Qwen3.6-27B** | Alibaba | 27B | 27B | Dense (hybrid attention) | 262K (1M via YaRN) | Text + Image + Video | Apache 2.0 | Apr 2026 |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "This reference covers every major open-source and open-weight large language model, with verified benchmark scores, license terms, hardware requirements, and hands-on performance data from real self-hosting tests.",
    "## Qwen 3, 3.5 and 3.6 (Alibaba)",
    "Qwen 3.6 (April 2026) is the generation that changed the economics."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "This reference covers every major open-source and open-weight large language model, with verified benchmark scores, license terms, hardware requirements, and hands-on performance data from real self-hosting tests.",
    "## Qwen 3, 3.5 and 3.6 (Alibaba)",
    "The two open releases pair a hybrid attention design (Gated DeltaNet linear attention mixed with standard self-attention) with a native 262K context extensible to roughly 1M tokens via YaRN scaling, and both accept text, images, and video."
   ]
  },
  {
   "url": "https://computingforgeeks.com/open-source-llm-comparison",
   "title": "Open Source LLM Comparison Table (2026) | ComputingForGeeks",
   "published_at": "2026-03-28T07:53:41",
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-02",
   "access_type": null,
   "license": null,
   "significance": "made the family natively multimodal, pushed context to 256K, and extended language coverage to 201 languages",
   "organization_evidence": [
    "This reference covers every major open-source and open-weight large language model, with verified benchmark scores, license terms, hardware requirements, and hands-on performance data from real self-hosting tests.",
    "## Master Comparison Table",
    "| Model | Developer | Total Params | Active Params | Architecture | Context Window | Multimodal | License | Release |",
    "| **Qwen 3.5 397B-A17B** | Alibaba | 397B | 17B | MoE | 256K tokens | Text + Image | Apache 2.0 | Feb 2026 |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "This reference covers every major open-source and open-weight large language model, with verified benchmark scores, license terms, hardware requirements, and hands-on performance data from real self-hosting tests.",
    "## Qwen 3, 3.5 and 3.6 (Alibaba)",
    "Qwen 3.5 (February 2026) made the family natively multimodal, pushed context to 256K, and extended language coverage to 201 languages."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "This reference covers every major open-source and open-weight large language model, with verified benchmark scores, license terms, hardware requirements, and hands-on performance data from real self-hosting tests.",
    "## Qwen 3, 3.5 and 3.6 (Alibaba)",
    "Qwen 3.5 (February 2026) made the family natively multimodal, pushed context to 256K, and extended language coverage to 201 languages."
   ]
  },
  {
   "url": "https://computingforgeeks.com/open-source-llm-comparison",
   "title": "Open Source LLM Comparison Table (2026) | ComputingForGeeks",
   "published_at": "2026-03-28T07:53:41",
   "organization": "Google",
   "model": null,
   "release_date": "2026-03-31",
   "access_type": null,
   "license": null,
   "significance": "launched March 31, 2026 with four sizes, and fixed the two biggest complaints about the family in one release",
   "organization_evidence": [
    "This reference covers every major open-source and open-weight large language model, with verified benchmark scores, license terms, hardware requirements, and hands-on performance data from real self-hosting tests.",
    "## Master Comparison Table",
    "| Model | Developer | Total Params | Active Params | Architecture | Context Window | Multimodal | License | Release |",
    "| **Gemma 4 31B** | Google | 30.7B | 30.7B | Dense | 256K tokens | Text + Image | Apache 2.0 | Mar 2026 |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "This reference covers every major open-source and open-weight large language model, with verified benchmark scores, license terms, hardware requirements, and hands-on performance data from real self-hosting tests.",
    "## Gemma 4 and Gemma 3 (Google)",
    "Gemma 4 launched March 31, 2026 with four sizes, and fixed the two biggest complaints about the family in one release."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "This reference covers every major open-source and open-weight large language model, with verified benchmark scores, license terms, hardware requirements, and hands-on performance data from real self-hosting tests.",
    "## Gemma 4 and Gemma 3 (Google)",
    "Gemma 4 launched March 31, 2026 with four sizes, and fixed the two biggest complaints about the family in one release."
   ]
  },
  {
   "url": "https://computingforgeeks.com/open-source-llm-comparison",
   "title": "Open Source LLM Comparison Table (2026) | ComputingForGeeks",
   "published_at": "2026-03-28T07:53:41",
   "organization": "Mistral AI",
   "model": null,
   "release_date": "2026-03",
   "access_type": null,
   "license": null,
   "significance": "119B total with only 6B active through 4-of-128 expert routing, unifying instruction following, configurable-depth reasoning, and multimodal input in one model",
   "organization_evidence": [
    "This reference covers every major open-source and open-weight large language model, with verified benchmark scores, license terms, hardware requirements, and hands-on performance data from real self-hosting tests.",
    "## Master Comparison Table",
    "| Model | Developer | Total Params | Active Params | Architecture | Context Window | Multimodal | License | Release |",
    "| **Mistral Small 4** | Mistral AI | 119B | 6B | MoE (128e, 4 active) | 256K tokens | Text + Image | Apache 2.0 | Mar 2026 |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "This reference covers every major open-source and open-weight large language model, with verified benchmark scores, license terms, hardware requirements, and hands-on performance data from real self-hosting tests.",
    "## Mistral Models",
    "Mistral Small 4 (March 2026) is the clever one: 119B total with only 6B active through 4-of-128 expert routing, unifying instruction following, configurable-depth reasoning, and multimodal input in one model."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "This reference covers every major open-source and open-weight large language model, with verified benchmark scores, license terms, hardware requirements, and hands-on performance data from real self-hosting tests.",
    "## Mistral Models",
    "Mistral Small 4 (March 2026) is the clever one: 119B total with only 6B active through 4-of-128 expert routing, unifying instruction following, configurable-depth reasoning, and multimodal input in one model."
   ]
  },
  {
   "url": "https://computingforgeeks.com/open-source-llm-comparison",
   "title": "Open Source LLM Comparison Table (2026) | ComputingForGeeks",
   "published_at": "2026-03-28T07:53:41",
   "organization": "Microsoft",
   "model": null,
   "release_date": "2026-03",
   "access_type": null,
   "license": null,
   "significance": "adds image understanding to the reasoning pipeline via a SigLIP-2 vision encoder, with dynamic resolution up to 3,600 visual tokens for GUI grounding and document analysis",
   "organization_evidence": [
    "This reference covers every major open-source and open-weight large language model, with verified benchmark scores, license terms, hardware requirements, and hands-on performance data from real self-hosting tests.",
    "## Master Comparison Table",
    "| Model | Developer | Total Params | Active Params | Architecture | Context Window | Multimodal | License | Release |",
    "| **Phi-4 Reasoning Vision** | Microsoft | 15B | 15B | Dense | 16K tokens | Text + Image | MIT | Mar 2026 |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "This reference covers every major open-source and open-weight large language model, with verified benchmark scores, license terms, hardware requirements, and hands-on performance data from real self-hosting tests.",
    "## Phi-4 (Microsoft)",
    "Phi-4 Reasoning Vision (15B, March 2026) adds image understanding to the reasoning pipeline via a SigLIP-2 vision encoder, with dynamic resolution up to 3,600 visual tokens for GUI grounding and document analysis."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "This reference covers every major open-source and open-weight large language model, with verified benchmark scores, license terms, hardware requirements, and hands-on performance data from real self-hosting tests.",
    "## Phi-4 (Microsoft)",
    "Phi-4 Reasoning Vision (15B, March 2026) adds image understanding to the reasoning pipeline via a SigLIP-2 vision encoder, with dynamic resolution up to 3,600 visual tokens for GUI grounding and document analysis."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/deepseek-v4-flash-0420",
   "title": "DeepSeek V4 Flash (max) - Intelligence, Performance & Price Analysis | Artificial Analysis",
   "published_at": null,
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "284B total parameters (13B active) and 1M context window",
   "organization_evidence": [
    "DeepSeek V4 Flash (Reasoning, Max Effort) Intelligence, Performance & Price Analysis",
    "DeepSeek V4 Flash (Reasoning, Max Effort) was created by DeepSeek."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "DeepSeek V4 Flash (Reasoning, Max Effort) Intelligence, Performance & Price Analysis",
    "DeepSeek V4 Flash (Reasoning, Max Effort) was released on April 24, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "DeepSeek V4 Flash (Reasoning, Max Effort) Intelligence, Performance & Price Analysis",
    "DeepSeek V4 Flash (Reasoning, Max Effort) scores 42 on the Artificial Analysis Intelligence Index, placing it well above average among comparable models (median: 27). When evaluating the Intelligence Index, it generated 230M tokens, which is very verbose in comparison to the median of 100M.",
    "| Context window | 1M ~1500 A4 pages of size 12 Arial font |",
    "| Total parameters | 284B |",
    "| Active parameters | 13B Number of parameters active per token during inference |"
   ]
  },
  {
   "url": "https://simplemondays.com/technology/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - Simple Mondays",
   "published_at": "2026-07-15T12:06:49",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor",
   "organization_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "China\u2019s Release Cadence Is the Story",
    "The Chinese open-weight landscape has expanded from a single lab two years ago to four distinct families: DeepSeek, Z.ai, Moonshot, and Alibaba, each with unique strategic focuses, such as affordability, long-horizon stability, or self-hosting capabilities."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "China\u2019s Release Cadence Is the Story",
    "Between April 24 and mid-June 2026, Chinese AI labs introduced four major open models: DeepSeek V4 on April 24, MiniMax M3 on June 1, and Kimi K2.7-Code and GLM-5.2 shortly thereafter."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "China\u2019s Release Cadence Is the Story",
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor"
   ]
  },
  {
   "url": "https://simplemondays.com/technology/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - Simple Mondays",
   "published_at": "2026-07-15T12:06:49",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "cheap 1M-token context, native multimodal, modified-MIT",
   "organization_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "China\u2019s Release Cadence Is the Story",
    "Between April 24 and mid-June 2026, Chinese AI labs introduced four major open models: DeepSeek V4 on April 24, MiniMax M3 on June 1, and Kimi K2.7-Code and GLM-5.2 shortly thereafter."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "China\u2019s Release Cadence Is the Story",
    "Between April 24 and mid-June 2026, Chinese AI labs introduced four major open models: DeepSeek V4 on April 24, MiniMax M3 on June 1, and Kimi K2.7-Code and GLM-5.2 shortly thereafter."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "China\u2019s Release Cadence Is the Story",
    "**MiniMax M3**cheap 1M-token context, native multimodal, modified-MIT"
   ]
  },
  {
   "url": "https://simplemondays.com/technology/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - Simple Mondays",
   "published_at": "2026-07-15T12:06:49",
   "organization": "Moonshot",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "agent-run specialist, ~30% fewer thinking tokens than K2.6",
   "organization_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "China\u2019s Release Cadence Is the Story",
    "The Chinese open-weight landscape has expanded from a single lab two years ago to four distinct families: DeepSeek, Z.ai, Moonshot, and Alibaba, each with unique strategic focuses, such as affordability, long-horizon stability, or self-hosting capabilities."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "China\u2019s Release Cadence Is the Story",
    "**Kimi K2.7-Code (Moonshot)**agent-run specialist, ~30% fewer thinking tokens than K2.6"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "China\u2019s Release Cadence Is the Story",
    "**Kimi K2.7-Code (Moonshot)**agent-run specialist, ~30% fewer thinking tokens than K2.6"
   ]
  },
  {
   "url": "https://simplemondays.com/technology/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - Simple Mondays",
   "published_at": "2026-07-15T12:06:49",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "753B MoE, MIT, top open-weight on Artificial Analysis index",
   "organization_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "China\u2019s Release Cadence Is the Story",
    "The Chinese open-weight landscape has expanded from a single lab two years ago to four distinct families: DeepSeek, Z.ai, Moonshot, and Alibaba, each with unique strategic focuses, such as affordability, long-horizon stability, or self-hosting capabilities."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "China\u2019s Release Cadence Is the Story",
    "**GLM-5.2 (Z.ai)**753B MoE, MIT, top open-weight on Artificial Analysis index"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "China\u2019s Release Cadence Is the Story",
    "**GLM-5.2 (Z.ai)**753B MoE, MIT, top open-weight on Artificial Analysis index"
   ]
  },
  {
   "url": "https://aidb.digital/blog/2026-07-17-moonshot-ai-releases-kimi-k3-the-world-s-largest-open-weight-model-at",
   "title": "Moonshot AI Releases Kimi K3, the World's Largest Open-Weight Model at 2.8 Trillion Parameters \u00b7 AIDB",
   "published_at": "2026-07-17T00:00:00",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-16",
   "access_type": null,
   "license": null,
   "significance": "a 2.8-trillion-parameter Mixture-of-Experts large language model",
   "organization_evidence": [
    "# Moonshot AI Releases Kimi K3, the World's Largest Open-Weight Model at 2.8 Trillion Parameters",
    "Moonshot AI, the Beijing-based artificial intelligence startup backed by Alibaba, released Kimi K3\u2014a 2.8-trillion-parameter model that the company says is now the largest open-source AI model in the world, and one that benchmarks show performs neck-and-neck with the most powerful proprietary systems from Anthropic and OpenAI."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Moonshot AI Releases Kimi K3, the World's Largest Open-Weight Model at 2.8 Trillion Parameters",
    "Kimi K3 went live on July 16, 2026, rolling out first on Kimi Code and inside the Kimi app, with two variants surfaced at launch\u2014K3 Max for chat and agent tasks, and K3 Swarm Max for large-scale parallel processing."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Moonshot AI Releases Kimi K3, the World's Largest Open-Weight Model at 2.8 Trillion Parameters",
    "The Beijing-based lab backed by Alibaba launched Kimi K3\u2014a 2.8-trillion-parameter Mixture-of-Experts large language model\u2014days before the 2026 World Artificial Intelligence Conference in Shanghai, claiming the largest open-weight release in AI history."
   ]
  },
  {
   "url": "https://pivotnews.ai/five/story/mira-murati-s-ai-lab-releases-open-weights-inkling-model",
   "title": "Mira Murati\u2019s AI Lab Releases Open-Weights Inkling Model",
   "published_at": "2026-07-17T14:41:46",
   "organization": "Thinking Machines",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "A 975B total parameter MoE transformer with broad multimodal capabilities, controllable thinking effort, and strong calibration.",
   "organization_evidence": [
    "Title: Inkling: Thinking Machines' Open-Weights Model URL Source: https://www. datacamp. com/blog/thinking-machines-inkling Markdown Content: Thinking Machines, the lab founded by former OpenAI CTO Mira Murati, has released its first model: Inkling.",
    "# Mira Murati\u2019s AI Lab Releases Open-Weights Inkling Model"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "It's the first release in a planned family, released July 15, 2026, alongside a preview of a smaller sibling, Inkling-Small.! Image 1",
    "# Mira Murati\u2019s AI Lab Releases Open-Weights Inkling Model",
    "Title: Inkling: Thinking Machines' Open-Weights Model URL Source: https://www. datacamp. com/blog/thinking-machines-inkling Markdown Content: Thinking Machines, the lab founded by former OpenAI CTO Mira Murati, has released its first model: Inkling."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Inkling is an open-weights MoE transformer with 975B total parameters and 41B active per token, positioned as a broad base model for fine-tuning rather than a top-of-leaderboard chatbot.",
    "# Mira Murati\u2019s AI Lab Releases Open-Weights Inkling Model",
    "Title: Inkling: Thinking Machines' Open-Weights Model URL Source: https://www. datacamp. com/blog/thinking-machines-inkling Markdown Content: Thinking Machines, the lab founded by former OpenAI CTO Mira Murati, has released its first model: Inkling."
   ]
  },
  {
   "url": "https://globalbyte.news/article/meituan-longcat-2-0-open-source-trillion-parameter-model-is-live-and-china-s-ai-week-was-packed",
   "title": "Meituan LongCat 2.0 Open Source Trillion Parameter Model Is Live - And China's AI Week Was Packed",
   "published_at": "2026-07-07T00:00:00",
   "organization": "Meituan",
   "model": null,
   "release_date": "2026-07-06",
   "access_type": null,
   "license": null,
   "significance": "A trillion-parameter sparse Mixture-of-Experts model featuring LongCat Sparse Attention and Zero Compute Experts for long-context processing with manageable inference costs, released with full weights and inference engine.",
   "organization_evidence": [
    "Meituan LongCat 2.0 Open Source Trillion Parameter Model Is Live - And China's AI Week Was Packed",
    "Meituan open-sources its LongCat-2.0 AI model with domestic chip adaptation; humanoid robot leader Star Era raises 1 billion yuan led by state-owned funds; Gaode improves its AI chauffeur service; and UU Run launches a zero-commission lifeline for eligible riders and their families.",
    "July 6, 2026, was a full news day."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Meituan LongCat 2.0 Open Source Trillion Parameter Model Is Live - And China's AI Week Was Packed",
    "July 6, 2026, was a full news day.",
    "Meituan officially released the **Meituan LongCat 2.0 open-source trillion-parameter model** - full weights, inference engine, and technical documentation all dropped at once."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Meituan LongCat 2.0 Open Source Trillion Parameter Model Is Live - And China's AI Week Was Packed",
    "Meituan officially released the **Meituan LongCat 2.0 open-source trillion-parameter model** - full weights, inference engine, and technical documentation all dropped at once.",
    "LongCat 2.0 is built on a sparse Mixture-of-Experts (MoE) architecture."
   ]
  },
  {
   "url": "https://moclaw.ai/blog/inkling-ai-thinking-machines",
   "title": "Inkling AI: Thinking Machines' First Open-Weight Model",
   "published_at": "2026-07-17T09:24:02",
   "organization": "Thinking Machines",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "975B-parameter open-weight multimodal mixture-of-experts model with a 1M context window",
   "organization_evidence": [
    "# Inkling AI: Thinking Machines' First Open-Weight Model",
    "Thinking Machines just released Inkling, a 975B open-weight multimodal model with a 1M context window.",
    "Thinking Machines' Inkling is the lab's first model, released July 15, 2026, and it's open-weight: you can download the full weights today."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Inkling AI: Thinking Machines' First Open-Weight Model",
    "Thinking Machines' Inkling is the lab's first model, released July 15, 2026, and it's open-weight: you can download the full weights today."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Inkling AI: Thinking Machines' First Open-Weight Model",
    "Thinking Machines just released Inkling, a 975B open-weight multimodal model with a 1M context window."
   ]
  },
  {
   "url": "https://felloai.com/cs/longcat-2-0",
   "title": "LongCat-2.0: Meituan\u2019s 1.6 Trillion-Parameter Open-Source Coding Model",
   "published_at": "2026-06-30T18:28:44",
   "organization": "Meituan",
   "model": null,
   "release_date": "2026-06-30",
   "access_type": null,
   "license": null,
   "significance": "1.6 trillion-parameter Mixture-of-Experts model with a native 1-million-token context window",
   "organization_evidence": [
    "**Meituan** has open-sourced **LongCat-2.0**, a **1.6 trillion-parameter** Mixture-of-Experts model with a native **1-million-token context window**, released on **June 30, 2026** under a permissive **Licence MIT**."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "**Meituan** has open-sourced **LongCat-2.0**, a **1.6 trillion-parameter** Mixture-of-Experts model with a native **1-million-token context window**, released on **June 30, 2026** under a permissive **Licence MIT**."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "**Meituan** has open-sourced **LongCat-2.0**, a **1.6 trillion-parameter** Mixture-of-Experts model with a native **1-million-token context window**, released on **June 30, 2026** under a permissive **Licence MIT**."
   ]
  },
  {
   "url": "https://digitalapplied.com/blog/open-weight-models-h1-2026-retrospective-deepseek-qwen-llama",
   "title": "Open-Weight Models H1 2026: DeepSeek, Qwen, Llama Recap",
   "published_at": "2026-05-10T23:00:00",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "A single architectural reset with efficiency at long context, featuring a hybrid attention stack.",
   "organization_evidence": [
    "Open-weight models in H1 2026 have stopped being a single story.",
    "DeepSeek shipped a single architectural reset, V4 Preview, on April 24."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Open-weight models in H1 2026 have stopped being a single story.",
    "DeepSeek shipped a single architectural reset, V4 Preview, on April 24."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Open-weight models in H1 2026 have stopped being a single story.",
    "DeepSeek shipped a single architectural reset, V4 Preview, on April 24.",
    "The signature contribution is efficiency at long context."
   ]
  },
  {
   "url": "https://digitalapplied.com/blog/open-weight-models-h1-2026-retrospective-deepseek-qwen-llama",
   "title": "Open-Weight Models H1 2026: DeepSeek, Qwen, Llama Recap",
   "published_at": "2026-05-10T23:00:00",
   "organization": "Alibaba/Qwen",
   "model": null,
   "release_date": "2026-02-16",
   "access_type": null,
   "license": null,
   "significance": "Launched the Qwen 3.5 family with an active release cadence and procurement-friendly licensing.",
   "organization_evidence": [
    "Open-weight models in H1 2026 have stopped being a single story.",
    "Alibaba/Qwen shipped Qwen 3.5 on [February 16](https://en.wikipedia.org/wiki/Qwen), a small-models drop on [March 2](https://www.marktechpost.com/2026/03/02/alibaba-just-released-qwen-3-5-small-models-a-family-of-0-8b-to-9b-parameters-built-for-on-device-applications/), and [Qwen 3.6 on April 16](https://github.com/QwenLM/Qwen3.6) with 35B-A3B and 27B open-weight checkpoints in the same April 16 drop."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Open-weight models in H1 2026 have stopped being a single story.",
    "Alibaba/Qwen shipped Qwen 3.5 on [February 16](https://en.wikipedia.org/wiki/Qwen), a small-models drop on [March 2](https://www.marktechpost.com/2026/03/02/alibaba-just-released-qwen-3-5-small-models-a-family-of-0-8b-to-9b-parameters-built-for-on-device-applications/), and [Qwen 3.6 on April 16](https://github.com/QwenLM/Qwen3.6) with 35B-A3B and 27B open-weight checkpoints in the same April 16 drop."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Open-weight models in H1 2026 have stopped being a single story.",
    "Qwen ran the most active release cadence with the Qwen 3.5 family in February and Qwen 3.6 in April."
   ]
  },
  {
   "url": "https://promptyze.com/chinese-open-models-already-own-download",
   "title": "Chinese Open Models Already Own the Download Race \u2014 Kimi K3 Is Just the Exclamation Point",
   "published_at": "2026-07-17T11:07:00",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-27",
   "access_type": null,
   "license": null,
   "significance": "the largest open-source model ever announced with 2.8 trillion total parameters, a 1-million-token context window, and native multimodal support",
   "organization_evidence": [
    "# Chinese Open Models Already Own the Download Race \u2014 Kimi K3 Is Just the Exclamation Point",
    "Then, on July 16, Moonshot AI released Kimi K3 and made that number feel almost understated.",
    "[VentureBeat](https://venturebeat.com/technology/chinas-moonshot-ai-releases-kimi-k3-the-largest-open-source-model-ever-rivaling-top-u-s-systems) called it the largest open-source model ever announced."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Chinese Open Models Already Own the Download Race \u2014 Kimi K3 Is Just the Exclamation Point",
    "Moonshot\u2019s Kimi K3 \u2014 2.8 trillion parameters, open weights as of July 27 \u2014 arrives as Chinese models hit 41% of Hugging Face downloads, surpassing the U.S. for the first time."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Chinese Open Models Already Own the Download Race \u2014 Kimi K3 Is Just the Exclamation Point",
    "[VentureBeat](https://venturebeat.com/technology/chinas-moonshot-ai-releases-kimi-k3-the-largest-open-source-model-ever-rivaling-top-u-s-systems) called it the largest open-source model ever announced.",
    "At 2.8 trillion total parameters with a 1-million-token context window, native multimodal support, and full weights dropping on July 27, it is hard to argue with that framing."
   ]
  },
  {
   "url": "https://news.geobrowser.io/story/1092aa0245e34ecaa2a1dc7a9baf14bd",
   "title": "Thinking Machines Lab Releases Inkling, 975B-Parameter Open-Weights Multimodal AI Model \u2014 Geo News",
   "published_at": "2026-07-15T00:00:00",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "975-billion-parameter open-weights Mixture-of-Experts model that natively processes text, images, and audio with a controllable thinking effort mechanism",
   "organization_evidence": [
    "Thinking Machines Lab Releases Inkling, 975B-Parameter Open-Weights Multimodal AI Model \u2014 Geo News",
    "Thinking Machines Lab, founded by former OpenAI CTO Mira Murati, has released Inkling, a 975-billion-parameter open-weights Mixture-of-Experts model under an Apache 2.0 license."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Thinking Machines Lab Releases Inkling, 975B-Parameter Open-Weights Multimodal AI Model \u2014 Geo News",
    "Thinking Machines Lab released Inkling, an open-weights Mixture-of-Experts transformer model with 975 billion total parameters and 41 billion active parameters, on July 15, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Thinking Machines Lab Releases Inkling, 975B-Parameter Open-Weights Multimodal AI Model \u2014 Geo News",
    "Thinking Machines Lab, founded by former OpenAI CTO Mira Murati, has released Inkling, a 975-billion-parameter open-weights Mixture-of-Experts model under an Apache 2.0 license."
   ]
  },
  {
   "url": "https://tech-insider.org/llama-4-vs-qwen-vs-mistral-2026",
   "title": "Llama 4 vs Qwen 3.5 vs Mistral: Open LLMs 2026",
   "published_at": "2026-07-06T12:18:50",
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-02",
   "access_type": null,
   "license": null,
   "significance": "Tops out at a 397B-total / 17B-active MoE model built from 256 experts, native 262,144-token context window, native vision-language support, and coverage of 201 languages.",
   "organization_evidence": [
    "The current flagship generation, **Qwen 3.5**, rolled out in February 2026 and tops out at a 397B-total / 17B-active MoE model built from 256 experts (8 routed plus 1 shared per token).",
    "## Meet the Contenders: Llama 4, Qwen 3.5, and Mistral Large 3",
    "### Alibaba Qwen 3.5 and 3.6: The Multilingual Speed Demon"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The current flagship generation, **Qwen 3.5**, rolled out in February 2026 and tops out at a 397B-total / 17B-active MoE model built from 256 experts (8 routed plus 1 shared per token).",
    "## Meet the Contenders: Llama 4, Qwen 3.5, and Mistral Large 3",
    "### Alibaba Qwen 3.5 and 3.6: The Multilingual Speed Demon"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The current flagship generation, **Qwen 3.5**, rolled out in February 2026 and tops out at a 397B-total / 17B-active MoE model built from 256 experts (8 routed plus 1 shared per token).",
    "It ships with a native 262,144-token context window that extends toward roughly one million tokens, native vision-language support, and coverage of an astonishing 201 languages.",
    "## Meet the Contenders: Llama 4, Qwen 3.5, and Mistral Large 3",
    "### Alibaba Qwen 3.5 and 3.6: The Multilingual Speed Demon"
   ]
  },
  {
   "url": "https://tech-insider.org/llama-4-vs-qwen-vs-mistral-2026",
   "title": "Llama 4 vs Qwen 3.5 vs Mistral: Open LLMs 2026",
   "published_at": "2026-07-06T12:18:50",
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "Includes a dense 27B build and a sparse 35B-A3B MoE explicitly tuned for agentic coding.",
   "organization_evidence": [
    "In April 2026, the Qwen team followed with **Qwen 3.6**, including a dense 27B build and a sparse 35B-A3B MoE (only 3B active parameters) explicitly tuned for agentic coding.",
    "## Meet the Contenders: Llama 4, Qwen 3.5, and Mistral Large 3",
    "### Alibaba Qwen 3.5 and 3.6: The Multilingual Speed Demon"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "In April 2026, the Qwen team followed with **Qwen 3.6**, including a dense 27B build and a sparse 35B-A3B MoE (only 3B active parameters) explicitly tuned for agentic coding.",
    "## Meet the Contenders: Llama 4, Qwen 3.5, and Mistral Large 3",
    "### Alibaba Qwen 3.5 and 3.6: The Multilingual Speed Demon"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "In April 2026, the Qwen team followed with **Qwen 3.6**, including a dense 27B build and a sparse 35B-A3B MoE (only 3B active parameters) explicitly tuned for agentic coding.",
    "## Meet the Contenders: Llama 4, Qwen 3.5, and Mistral Large 3",
    "### Alibaba Qwen 3.5 and 3.6: The Multilingual Speed Demon"
   ]
  },
  {
   "url": "https://tech-insider.org/llama-4-vs-qwen-vs-mistral-2026",
   "title": "Llama 4 vs Qwen 3.5 vs Mistral: Open LLMs 2026",
   "published_at": "2026-07-06T12:18:50",
   "organization": "Mistral AI",
   "model": null,
   "release_date": "2025-12",
   "access_type": null,
   "license": null,
   "significance": "A 675B-total / 41B-active MoE model with a 256K context window and text-and-image input across 80-plus languages.",
   "organization_evidence": [
    "Mistral AI answered in December 2025 with **Mistral Large 3**, a 675B-total / 41B-active MoE model \u2014 the largest total parameter count in this trio \u2014 under a clean Apache 2.0 license and a 256K context window with text-and-image input across 80-plus languages.",
    "## Meet the Contenders: Llama 4, Qwen 3.5, and Mistral Large 3",
    "### Mistral Large 3: The Apache-Licensed European Flagship"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Mistral AI answered in December 2025 with **Mistral Large 3**, a 675B-total / 41B-active MoE model \u2014 the largest total parameter count in this trio \u2014 under a clean Apache 2.0 license and a 256K context window with text-and-image input across 80-plus languages.",
    "## Meet the Contenders: Llama 4, Qwen 3.5, and Mistral Large 3",
    "### Mistral Large 3: The Apache-Licensed European Flagship"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Mistral AI answered in December 2025 with **Mistral Large 3**, a 675B-total / 41B-active MoE model \u2014 the largest total parameter count in this trio \u2014 under a clean Apache 2.0 license and a 256K context window with text-and-image input across 80-plus languages.",
    "## Meet the Contenders: Llama 4, Qwen 3.5, and Mistral Large 3",
    "### Mistral Large 3: The Apache-Licensed European Flagship"
   ]
  },
  {
   "url": "https://tech-insider.org/llama-4-vs-qwen-vs-mistral-2026",
   "title": "Llama 4 vs Qwen 3.5 vs Mistral: Open LLMs 2026",
   "published_at": "2026-07-06T12:18:50",
   "organization": "Mistral AI",
   "model": null,
   "release_date": "2026-03",
   "access_type": null,
   "license": null,
   "significance": "A remarkably lean 119B-total model that activates only 6B parameters per token, 256K context, and multimodal.",
   "organization_evidence": [
    "In March 2026 Mistral added **Mistral Small 4**, a remarkably lean 119B-total model that activates only 6B parameters per token (128 experts, 4 active), also 256K context and multimodal.",
    "## Meet the Contenders: Llama 4, Qwen 3.5, and Mistral Large 3",
    "### Mistral Large 3: The Apache-Licensed European Flagship"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "In March 2026 Mistral added **Mistral Small 4**, a remarkably lean 119B-total model that activates only 6B parameters per token (128 experts, 4 active), also 256K context and multimodal.",
    "## Meet the Contenders: Llama 4, Qwen 3.5, and Mistral Large 3",
    "### Mistral Large 3: The Apache-Licensed European Flagship"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "In March 2026 Mistral added **Mistral Small 4**, a remarkably lean 119B-total model that activates only 6B parameters per token (128 experts, 4 active), also 256K context and multimodal.",
    "## Meet the Contenders: Llama 4, Qwen 3.5, and Mistral Large 3",
    "### Mistral Large 3: The Apache-Licensed European Flagship"
   ]
  },
  {
   "url": "https://app.dealroom.co/news/note/poolside-releases-laguna-s-2-1-a-118b-open-weight-coding-model-pitched-as-the-west-s-answer-to-deepseek-and-qwen",
   "title": "Dealroom.co | Poolside releases Laguna S 2.1 \u2014 a 118B open-weight coding model pitched as the West's answer to DeepSeek and Qwen",
   "published_at": null,
   "organization": "Poolside",
   "model": null,
   "release_date": "2026-07-21",
   "access_type": null,
   "license": null,
   "significance": "A 118B total-parameter Mixture-of-Experts model with just 8B activated parameters per token, a context window up to 1M tokens, scoring 70.2% on Terminal-Bench 2.1 and 59.4% on SWE-Bench Pro.",
   "organization_evidence": [
    "# Poolside releases Laguna S 2.1 \u2014 a 118B open-weight coding model pitched as the West's answer to DeepSeek and Qwen",
    "On 21 July 2026, [Poolside](https://app.dealroom.co/companies/poolside_ai) released **Laguna S 2.1**, an open-weight foundation model built for agentic coding that the San Francisco lab says matches or beats models several times its size."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Poolside releases Laguna S 2.1 \u2014 a 118B open-weight coding model pitched as the West's answer to DeepSeek and Qwen",
    "On 21 July 2026, [Poolside](https://app.dealroom.co/companies/poolside_ai) released **Laguna S 2.1**, an open-weight foundation model built for agentic coding that the San Francisco lab says matches or beats models several times its size."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Poolside releases Laguna S 2.1 \u2014 a 118B open-weight coding model pitched as the West's answer to DeepSeek and Qwen",
    "**The model:** A 118B total-parameter Mixture-of-Experts model with just 8B activated parameters per token, a context window up to 1M tokens, and support for thinking / no-thinking modes.",
    "It scores 70.2% on Terminal-Bench 2.1 and 59.4% on SWE-Bench Pro (public), matching or beating models from DeepSeek, Nvidia and Tencent that carry two to eight times as many active parameters."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/articles/xai-launches-grok-4-3-with-improved-agentic-performance-and-lower-pricing?:~:text=%E2%9E%A4%20Large%20increase%20in%20real_world%20agentic%20task&text=The%20largest%20single%20benchmark%20improvement_0309%20v2%27s%20score%20of%201179",
   "title": "xAI launches Grok 4.3 with improved agentic performance and lower pricing",
   "published_at": "2026-04-30T00:00:00",
   "organization": "Thinking Machines",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "debuting at 41 on the Artificial Analysis Intelligence Index",
   "organization_evidence": [
    "Thinking Machines has released Inkling, the new leading U.S. open weights model",
    "Thinking Machines has released Inkling, the new leading U.S. open weights model, debuting at 41 on the Artificial Analysis Intelligence Index July 15, 2026"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Thinking Machines has released Inkling, the new leading U.S. open weights model, debuting at 41 on the Artificial Analysis Intelligence Index July 15, 2026"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Thinking Machines has released Inkling, the new leading U.S. open weights model, debuting at 41 on the Artificial Analysis Intelligence Index July 15, 2026"
   ]
  },
  {
   "url": "https://infoworld.com/article/4197743/thinking-machines-offers-enterprises-a-us-alternative-in-open-weight-ai.html",
   "title": "Thinking Machines Lab offers enterprises a US alternative in open-weight AI",
   "published_at": "2026-07-16T05:21:50",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-16",
   "access_type": null,
   "license": null,
   "significance": "975-billion-parameter model with a 1-million-token context window",
   "organization_evidence": [
    "# Thinking Machines Lab offers enterprises a US alternative in open-weight AI",
    "Thinking Machines Lab, the San Francisco startup founded by former OpenAI [CTO Mira Murati](https://www.computerworld.com/article/3829004/ex-openai-cto-mira-murati-launches-ai-startup-recruits-top-talent-from-rivals.html), has released Inkling, its first general-purpose AI model."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Jul 16, 20265 mins",
    "# Thinking Machines Lab offers enterprises a US alternative in open-weight AI"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## Inkling, a 975-billion-parameter model, can be customized through Tinker and supports a 1-million-token context window, but it enters a market where Chinese models lead several coding and reasoning benchmarks.",
    "# Thinking Machines Lab offers enterprises a US alternative in open-weight AI"
   ]
  },
  {
   "url": "https://miniloop.ai/blog/best-open-source-llms-2026",
   "title": "Best Open Source LLMs 2026",
   "published_at": "2026-01-21T22:51:21",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "scoring 80.6 SWE-Bench Verified and 90.1 GPQA Diamond with a 1M context window",
   "organization_evidence": [
    "Best Open Source LLMs: 25+ Models Compared for 2026",
    "DeepSeek shipped **V4 Pro** in April 2026, scoring **80.6 SWE-Bench Verified** and **90.1 GPQA Diamond** with a 1M context window."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Best Open Source LLMs: 25+ Models Compared for 2026",
    "DeepSeek shipped **V4 Pro** in April 2026, scoring **80.6 SWE-Bench Verified** and **90.1 GPQA Diamond** with a 1M context window."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Best Open Source LLMs: 25+ Models Compared for 2026",
    "DeepSeek shipped **V4 Pro** in April 2026, scoring **80.6 SWE-Bench Verified** and **90.1 GPQA Diamond** with a 1M context window."
   ]
  },
  {
   "url": "https://miniloop.ai/blog/best-open-source-llms-2026",
   "title": "Best Open Source LLMs 2026",
   "published_at": "2026-01-21T22:51:21",
   "organization": "Mistral AI",
   "model": null,
   "release_date": "2026-04-29",
   "access_type": null,
   "license": null,
   "significance": "128B dense, 77.6% SWE-Bench Verified",
   "organization_evidence": [
    "Best Open Source LLMs: 25+ Models Compared for 2026",
    "## Best for EU Deployment: [Mistral Medium 3.5](https://mistral.ai?utm_source=miniloop)\n\n> **Update (May 2026):** Mistral shipped **Mistral Medium 3.5**",
    "Mistral shipped **Mistral Medium 3.5** on April 29, 2026 (128B dense, **77.6% SWE-Bench Verified**), now the EU-friendly coding pick."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Best Open Source LLMs: 25+ Models Compared for 2026",
    "## Best for EU Deployment: [Mistral Medium 3.5](https://mistral.ai?utm_source=miniloop)\n\n> **Update (May 2026):** Mistral shipped **Mistral Medium 3.5**",
    "Mistral shipped **Mistral Medium 3.5** on April 29, 2026 (128B dense, **77.6% SWE-Bench Verified**), now the EU-friendly coding pick."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Best Open Source LLMs: 25+ Models Compared for 2026",
    "## Best for EU Deployment: [Mistral Medium 3.5](https://mistral.ai?utm_source=miniloop)\n\n> **Update (May 2026):** Mistral shipped **Mistral Medium 3.5**",
    "Mistral shipped **Mistral Medium 3.5** on April 29, 2026 (128B dense, **77.6% SWE-Bench Verified**), now the EU-friendly coding pick."
   ]
  },
  {
   "url": "https://aidailyshot.com/blog/open-weight-language-models-2026-open-source-ai",
   "title": "Open Weight Language Models: Why 2026 Is the Year of Open-Source AI",
   "published_at": "2026-02-27T04:30:00",
   "organization": "Hugging Face",
   "model": null,
   "release_date": "2026-02",
   "access_type": null,
   "license": null,
   "significance": "near-GPT-4 performance",
   "organization_evidence": [
    "## The Great Unshackling: Major Releases That Changed the Game",
    "In February, [Hugging Face\u2019s OpenWeight-12B](https://huggingface.co/blog/open-weights-policy) dropped with near-GPT-4 performance, and within days, it was cloned, fine-tuned, and supercharged by hundreds of researchers and startups worldwide."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## The Great Unshackling: Major Releases That Changed the Game",
    "In February, [Hugging Face\u2019s OpenWeight-12B](https://huggingface.co/blog/open-weights-policy) dropped with near-GPT-4 performance, and within days, it was cloned, fine-tuned, and supercharged by hundreds of researchers and startups worldwide."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## The Great Unshackling: Major Releases That Changed the Game",
    "In February, [Hugging Face\u2019s OpenWeight-12B](https://huggingface.co/blog/open-weights-policy) dropped with near-GPT-4 performance, and within days, it was cloned, fine-tuned, and supercharged by hundreds of researchers and startups worldwide."
   ]
  },
  {
   "url": "https://aidailyshot.com/blog/open-weight-language-models-2026-open-source-ai",
   "title": "Open Weight Language Models: Why 2026 Is the Year of Open-Source AI",
   "published_at": "2026-02-27T04:30:00",
   "organization": "Google",
   "model": null,
   "release_date": "2026-03",
   "access_type": null,
   "license": null,
   "significance": "formal open-weight releases",
   "organization_evidence": [
    "## The Great Unshackling: Major Releases That Changed the Game",
    "By March, Google had pivoted, announcing formal open-weight releases for Gemini-Next under a permissive license, citing \u201cthe inevitability of open ecosystems.\u201d"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## The Great Unshackling: Major Releases That Changed the Game",
    "By March, Google had pivoted, announcing formal open-weight releases for Gemini-Next under a permissive license, citing \u201cthe inevitability of open ecosystems.\u201d"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## The Great Unshackling: Major Releases That Changed the Game",
    "By March, Google had pivoted, announcing formal open-weight releases for Gemini-Next under a permissive license, citing \u201cthe inevitability of open ecosystems.\u201d"
   ]
  },
  {
   "url": "https://aidailyshot.com/blog/open-weight-language-models-2026-open-source-ai",
   "title": "Open Weight Language Models: Why 2026 Is the Year of Open-Source AI",
   "published_at": "2026-02-27T04:30:00",
   "organization": "Meta",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "raw power with full weights and pretraining data",
   "organization_evidence": [
    "## The Great Unshackling: Major Releases That Changed the Game",
    "Meta, not to be outdone, pushed out Llama 4-70B with full weights and pretraining data\u2014no filters, no gating, just raw power."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## The Great Unshackling: Major Releases That Changed the Game",
    "Meta, not to be outdone, pushed out Llama 4-70B with full weights and pretraining data\u2014no filters, no gating, just raw power."
   ]
  },
  {
   "url": "https://aidailyshot.com/blog/open-weight-language-models-2026-open-source-ai",
   "title": "Open Weight Language Models: Why 2026 Is the Year of Open-Source AI",
   "published_at": "2026-02-27T04:30:00",
   "organization": "OpenAI",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "60B model",
   "organization_evidence": [
    "## The Great Unshackling: Major Releases That Changed the Game",
    "Even OpenAI, fiercely proprietary for years, quietly partnered with academic consortia to co-develop and open-source the \u201cMosaic\u201d 60B model, tacitly admitting the tide had turned."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## The Great Unshackling: Major Releases That Changed the Game",
    "Even OpenAI, fiercely proprietary for years, quietly partnered with academic consortia to co-develop and open-source the \u201cMosaic\u201d 60B model, tacitly admitting the tide had turned."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/hy3-non-reasoning",
   "title": "Hy3-preview - Intelligence, Performance & Price Analysis | Artificial Analysis",
   "published_at": null,
   "organization": "Tencent",
   "model": null,
   "release_date": "2026-04-23",
   "access_type": null,
   "license": null,
   "significance": "a Mixture of Experts (MoE) model with 295 billion total parameters, but only 21 billion active parameters are used during inference",
   "organization_evidence": [
    "Hy3-preview (Non-reasoning) was created by Tencent.",
    "Tencent has launched a newer model, [Hy3](/models/hy3)."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Hy3-preview (Non-reasoning) was released on April 23, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Hy3-preview (Non-reasoning) is a Mixture of Experts (MoE) model with 295 billion total parameters, but only 21 billion active parameters are used during inference."
   ]
  },
  {
   "url": "https://siliconangle.com/2026/06/30/chinas-meituan-open-sources-massive-longcat-2-0-ai-model-saying-trained-domestic-chips",
   "title": "China\u2019s Meituan open-sources massive LongCat-2.0 AI model, saying it was trained on domestic chips",
   "published_at": "2026-06-30T16:39:10",
   "organization": "Meituan",
   "model": "LongCat-2.0",
   "release_date": "2026-06-30",
   "access_type": "open-source",
   "license": null,
   "significance": "1.6-trillion-parameter sparse Mixture of Experts model with a 1-million-token context window",
   "organization_evidence": [
    "China\u2019s Meituan open-sources massive LongCat-2.0 AI model, saying it was trained on domestic chips",
    "Beijing, China-based Meituan Inc. [today debuted](https://longcat.chat/blog/longcat-2.0/) its next-gen LongCat-2.0 open-source large language model, stating that the company trained the 1.6-trillion-parameter model on domestic Chinese chips and compute clusters."
   ],
   "model_evidence": [
    "China\u2019s Meituan open-sources massive LongCat-2.0 AI model, saying it was trained on domestic chips",
    "Beijing, China-based Meituan Inc. [today debuted](https://longcat.chat/blog/longcat-2.0/) its next-gen LongCat-2.0 open-source large language model, stating that the company trained the 1.6-trillion-parameter model on domestic Chinese chips and compute clusters."
   ],
   "release_date_evidence": [
    "UPDATED 12:00 EDT / JUNE 30 2026",
    "Beijing, China-based Meituan Inc. [today debuted](https://longcat.chat/blog/longcat-2.0/) its next-gen LongCat-2.0 open-source large language model, stating that the company trained the 1.6-trillion-parameter model on domestic Chinese chips and compute clusters."
   ],
   "access_type_evidence": [
    "China\u2019s Meituan open-sources massive LongCat-2.0 AI model, saying it was trained on domestic chips",
    "Beijing, China-based Meituan Inc. [today debuted](https://longcat.chat/blog/longcat-2.0/) its next-gen LongCat-2.0 open-source large language model, stating that the company trained the 1.6-trillion-parameter model on domestic Chinese chips and compute clusters."
   ],
   "license_evidence": null,
   "significance_evidence": [
    "Beijing, China-based Meituan Inc. [today debuted](https://longcat.chat/blog/longcat-2.0/) its next-gen LongCat-2.0 open-source large language model, stating that the company trained the 1.6-trillion-parameter model on domestic Chinese chips and compute clusters.",
    "Weighing in at 1.6 trillion parameters, the model is no lightweight and it delivers with a 1-million-token context window."
   ]
  },
  {
   "url": "https://iotdigitaltwinplm.com/kimi-k3-explained-reasoning-model-architecture-benchmarks-2026",
   "title": "Kimi K3 Explained: Moonshot\u2019s 2.8T Open-Weight Reasoning Model (2026)",
   "published_at": "2026-07-28T07:01:00",
   "organization": "Moonshot AI",
   "model": "Kimi K3",
   "release_date": "2026-07-26",
   "access_type": "open-weight",
   "license": "Modified MIT",
   "significance": "2.8-trillion-parameter mixture-of-experts reasoning and agentic model with 16 active experts per token and a 1-million-token context window",
   "organization_evidence": [
    "Kimi K3 Explained: Moonshot\u2019s 2.8T Open-Weight Reasoning Model (2026)",
    "Kimi K3 is a 2.8-trillion-parameter mixture-of-experts language model from Moonshot AI, released with fully open weights on July 26, 2026."
   ],
   "model_evidence": [
    "Kimi K3 Explained: Moonshot\u2019s 2.8T Open-Weight Reasoning Model (2026)",
    "**Kimi K3** is Moonshot AI\u2019s newest flagship large language model, and it is the largest open-weight model anyone has shipped with a straight face: 2.8 trillion total parameters, spread across 896 experts, of which only 16 fire on any given token."
   ],
   "release_date_evidence": [
    "Kimi K3 Explained: Moonshot\u2019s 2.8T Open-Weight Reasoning Model (2026)",
    "Kimi K3 is a 2.8-trillion-parameter mixture-of-experts language model from Moonshot AI, released with fully open weights on July 26, 2026."
   ],
   "access_type_evidence": [
    "Kimi K3 Explained: Moonshot\u2019s 2.8T Open-Weight Reasoning Model (2026)",
    "Kimi K3 is a 2.8-trillion-parameter mixture-of-experts language model from Moonshot AI, released with fully open weights on July 26, 2026."
   ],
   "license_evidence": [
    "K2.6, released April 20, 2026, under a Modified MIT license, pushed further on reasoning post-training and set the stage for what Moonshot was quietly calling its \u201cThinking\u201d line \u2014 a separate mode, distinct from the base Instant mode, tuned specifically for multi-step problems where the model reasons before it answers.",
    "The reported license lineage continues from K2\u2019s Modified MIT terms, though as with any fast-moving open-weight release, engineers evaluating K3 for production use should verify the exact license text in the model repository rather than assuming continuity.",
    "Kimi K3 Explained: Moonshot\u2019s 2.8T Open-Weight Reasoning Model (2026)"
   ],
   "significance_evidence": [
    "Kimi K3 Explained: Moonshot\u2019s 2.8T Open-Weight Reasoning Model (2026)",
    "Kimi K3 is a 2.8-trillion-parameter mixture-of-experts language model from Moonshot AI, released with fully open weights on July 26, 2026.",
    "It activates only 16 of 896 experts per token, supports a 1-million-token context window, and is positioned as a reasoning-and-agentic model competing with GPT-5.6, Claude Fable 5, and DeepSeek-V4."
   ]
  },
  {
   "url": "https://fazm.ai/blog/llm-model-release-april-2026",
   "title": "Fazm - AI Computer Agent for macOS | Voice-Controlled Desktop Automation",
   "published_at": "2026-04-12T13:23:22",
   "organization": "Meta",
   "model": "Llama 4 Scout",
   "release_date": "2026-04-05",
   "access_type": "open weights",
   "license": null,
   "significance": "MoE (17B active / 109B total), 10M context window",
   "organization_evidence": [
    "# Every LLM Model Release in April 2026: Specs, Benchmarks, and Selection Guide",
    "Every major LLM model release that shipped a production API or downloadable weights during April 2026:",
    "| Llama 4 Scout | Meta | Apr 5 | MoE (17B active / 109B total) | 17B | 10M | Free (open weights) | Free (open weights) |"
   ],
   "model_evidence": [
    "# Every LLM Model Release in April 2026: Specs, Benchmarks, and Selection Guide",
    "Every major LLM model release that shipped a production API or downloadable weights during April 2026:",
    "| Llama 4 Scout | Meta | Apr 5 | MoE (17B active / 109B total) | 17B | 10M | Free (open weights) | Free (open weights) |"
   ],
   "release_date_evidence": [
    "# Every LLM Model Release in April 2026: Specs, Benchmarks, and Selection Guide",
    "Every major LLM model release that shipped a production API or downloadable weights during April 2026:",
    "| Llama 4 Scout | Meta | Apr 5 | MoE (17B active / 109B total) | 17B | 10M | Free (open weights) | Free (open weights) |"
   ],
   "access_type_evidence": [
    "# Every LLM Model Release in April 2026: Specs, Benchmarks, and Selection Guide",
    "Every major LLM model release that shipped a production API or downloadable weights during April 2026:",
    "| Llama 4 Scout | Meta | Apr 5 | MoE (17B active / 109B total) | 17B | 10M | Free (open weights) | Free (open weights) |"
   ],
   "license_evidence": null,
   "significance_evidence": [
    "# Every LLM Model Release in April 2026: Specs, Benchmarks, and Selection Guide",
    "Every major LLM model release that shipped a production API or downloadable weights during April 2026:",
    "| Llama 4 Scout | Meta | Apr 5 | MoE (17B active / 109B total) | 17B | 10M | Free (open weights) | Free (open weights) |"
   ]
  },
  {
   "url": "https://fazm.ai/blog/llm-model-release-april-2026",
   "title": "Fazm - AI Computer Agent for macOS | Voice-Controlled Desktop Automation",
   "published_at": "2026-04-12T13:23:22",
   "organization": "Meta",
   "model": "Llama 4 Maverick",
   "release_date": "2026-04-05",
   "access_type": "open weights",
   "license": null,
   "significance": "MoE (17B active / 400B total), 1M context window",
   "organization_evidence": [
    "# Every LLM Model Release in April 2026: Specs, Benchmarks, and Selection Guide",
    "Every major LLM model release that shipped a production API or downloadable weights during April 2026:",
    "| Llama 4 Maverick | Meta | Apr 5 | MoE (17B active / 400B total) | 17B | 1M | Free (open weights) | Free (open weights) |"
   ],
   "model_evidence": [
    "# Every LLM Model Release in April 2026: Specs, Benchmarks, and Selection Guide",
    "Every major LLM model release that shipped a production API or downloadable weights during April 2026:",
    "| Llama 4 Maverick | Meta | Apr 5 | MoE (17B active / 400B total) | 17B | 1M | Free (open weights) | Free (open weights) |"
   ],
   "release_date_evidence": [
    "# Every LLM Model Release in April 2026: Specs, Benchmarks, and Selection Guide",
    "Every major LLM model release that shipped a production API or downloadable weights during April 2026:",
    "| Llama 4 Maverick | Meta | Apr 5 | MoE (17B active / 400B total) | 17B | 1M | Free (open weights) | Free (open weights) |"
   ],
   "access_type_evidence": [
    "# Every LLM Model Release in April 2026: Specs, Benchmarks, and Selection Guide",
    "Every major LLM model release that shipped a production API or downloadable weights during April 2026:",
    "| Llama 4 Maverick | Meta | Apr 5 | MoE (17B active / 400B total) | 17B | 1M | Free (open weights) | Free (open weights) |"
   ],
   "license_evidence": null,
   "significance_evidence": [
    "# Every LLM Model Release in April 2026: Specs, Benchmarks, and Selection Guide",
    "Every major LLM model release that shipped a production API or downloadable weights during April 2026:",
    "| Llama 4 Maverick | Meta | Apr 5 | MoE (17B active / 400B total) | 17B | 1M | Free (open weights) | Free (open weights) |"
   ]
  },
  {
   "url": "https://fazm.ai/blog/llm-model-release-april-2026",
   "title": "Fazm - AI Computer Agent for macOS | Voice-Controlled Desktop Automation",
   "published_at": "2026-04-12T13:23:22",
   "organization": "Alibaba",
   "model": "Qwen 3 235B",
   "release_date": "2026-04-07",
   "access_type": "open weights",
   "license": null,
   "significance": "MoE (22B active / 235B total), 128K context window",
   "organization_evidence": [
    "# Every LLM Model Release in April 2026: Specs, Benchmarks, and Selection Guide",
    "Every major LLM model release that shipped a production API or downloadable weights during April 2026:",
    "| Qwen 3 235B | Alibaba | Apr 7 | MoE (22B active / 235B total) | 22B | 128K | $1.50 | $6.00 |"
   ],
   "model_evidence": [
    "# Every LLM Model Release in April 2026: Specs, Benchmarks, and Selection Guide",
    "Every major LLM model release that shipped a production API or downloadable weights during April 2026:",
    "| Qwen 3 235B | Alibaba | Apr 7 | MoE (22B active / 235B total) | 22B | 128K | $1.50 | $6.00 |"
   ],
   "release_date_evidence": [
    "# Every LLM Model Release in April 2026: Specs, Benchmarks, and Selection Guide",
    "Every major LLM model release that shipped a production API or downloadable weights during April 2026:",
    "| Qwen 3 235B | Alibaba | Apr 7 | MoE (22B active / 235B total) | 22B | 128K | $1.50 | $6.00 |"
   ],
   "access_type_evidence": [
    "# Every LLM Model Release in April 2026: Specs, Benchmarks, and Selection Guide",
    "Pricing note",
    "Open-weight models like Llama 4 and Qwen 3 cost nothing to download, but self-hosting requires GPU infrastructure."
   ],
   "license_evidence": null,
   "significance_evidence": [
    "# Every LLM Model Release in April 2026: Specs, Benchmarks, and Selection Guide",
    "Every major LLM model release that shipped a production API or downloadable weights during April 2026:",
    "| Qwen 3 235B | Alibaba | Apr 7 | MoE (22B active / 235B total) | 22B | 128K | $1.50 | $6.00 |"
   ]
  },
  {
   "url": "https://theaugustdispatch.com/2026/07/16/thinking-machines-launches-inkling-a-new-open-weight-model-and-a-game-changer-in-ai",
   "title": "Thinking Machines Launches Inkling: A New Open Weight Model and A Game Changer in AI - August Wheel",
   "published_at": "2026-07-16T13:54:57",
   "organization": "Thinking Machines",
   "model": "Inkling",
   "release_date": "2026-07-15",
   "access_type": "open weight",
   "license": "Apache 2.0",
   "significance": "multimodal open weight model scoring highest among US built open weight models on Artificial Analysis' Intelligence Index with a 1 million token context window",
   "organization_evidence": [
    "Thinking Machines Launches Inkling: A New Open Weight Model and A Game Changer in AI",
    "Thinking Machines just released Inkling, a free open weight AI model."
   ],
   "model_evidence": [
    "Thinking Machines Launches Inkling: A New Open Weight Model and A Game Changer in AI",
    "Thinking Machines just released Inkling, a free open weight AI model."
   ],
   "release_date_evidence": [
    "Thinking Machines Launches Inkling: A New Open Weight Model and A Game Changer in AI",
    "- Released July 15, 2026, under a fully open Apache 2.0 license"
   ],
   "access_type_evidence": [
    "Thinking Machines Launches Inkling: A New Open Weight Model and A Game Changer in AI",
    "Thinking Machines just released Inkling, a free open weight AI model.",
    "It\u2019s open weight, meaning the full model is free to download and modify, and it comes less than a year after Meta reportedly offered individual researchers at this same company over a billion dollars to leave."
   ],
   "license_evidence": [
    "Thinking Machines Launches Inkling: A New Open Weight Model and A Game Changer in AI",
    "- Released July 15, 2026, under a fully open Apache 2.0 license"
   ],
   "significance_evidence": [
    "Thinking Machines Launches Inkling: A New Open Weight Model and A Game Changer in AI",
    "- Multimodal: reads text, images, and audio natively in the same request, with a 1 million token context window",
    "- Scores 41 on Artificial Analysis\u2019 Intelligence Index, the highest of any US built open weight model, but still behind Claude Opus 4.8 (56) and GPT-5.6 (59)"
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/qwen3-6-plus-vs-qwen3-5-122b-a10b",
   "title": "Qwen3.6 Plus vs Qwen3.5 122B A10B (Reasoning): Model Comparison",
   "published_at": null,
   "organization": "Alibaba",
   "model": null,
   "release_date": "February, 2026",
   "access_type": null,
   "license": null,
   "significance": "122B open reasoning model",
   "organization_evidence": [
    "# Qwen3.6 Plus vs. Qwen3.5 122B A10B (Reasoning)",
    "| Creator | ![Alibaba](/img/logos/alibaba_small.svg)[](https://www.alibabacloud.com/) | ![Alibaba](/img/logos/alibaba_small.svg)[](https://www.alibabacloud.com/) | |",
    "| Open Source (Weights) | No | Yes[](https://huggingface.co/Qwen/Qwen3.5-122B-A10B) | Qwen3.5 122B A10B (Reasoning) is open source while Qwen3.6 Plus is proprietary |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Qwen3.6 Plus vs. Qwen3.5 122B A10B (Reasoning)",
    "| Release Date | April, 2026 | February, 2026 | Qwen3.6 Plus has a more recent release date than Qwen3.5 122B A10B (Reasoning) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Qwen3.6 Plus vs. Qwen3.5 122B A10B (Reasoning)",
    "Comparison between Qwen3.6 Plus and Qwen3.5 122B A10B (Reasoning) across intelligence, price, speed, context window and more.",
    "| Metric | ![Alibaba logo](/img/logos/alibaba_small.svg)Qwen3.6 Plus | ![Alibaba logo](/img/logos/alibaba_small.svg)Qwen3.5 122B A10B (Reasoning) | Analysis |"
   ]
  },
  {
   "url": "https://teqvolt.com/open-source/trinity-large-thinking-arcee-400b-open-reasoning-model",
   "title": "Trinity-Large-Thinking: 400B U.S.-Made Open Reasoning Model",
   "published_at": "2026-04-30T06:46:44",
   "organization": "Arcee AI",
   "model": null,
   "release_date": "2026-04-01",
   "access_type": null,
   "license": null,
   "significance": "400-billion-parameter sparse Mixture-of-Experts model trained on 2,048 NVIDIA B300 GPUs, optimized for long-horizon agentic reasoning and tool use.",
   "organization_evidence": [
    "Trinity-Large-Thinking: The 400B U.S.-Made Open Model That Bet the Company",
    "Then, on April 1, 2026, a 100-person startup called **Arcee AI** released **Trinity-Large-Thinking** under an Apache 2.0 license."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Trinity-Large-Thinking: The 400B U.S.-Made Open Model That Bet the Company",
    "Then, on April 1, 2026, a 100-person startup called **Arcee AI** released **Trinity-Large-Thinking** under an Apache 2.0 license."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Trinity-Large-Thinking: The 400B U.S.-Made Open Model That Bet the Company",
    "It is a 400-billion-parameter sparse Mixture-of-Experts model trained on 2,048 NVIDIA B300 GPUs over roughly 33 days, optimized for long-horizon agentic reasoning."
   ]
  },
  {
   "url": "https://felloai.com/kimi-k3",
   "title": "Kimi K3: Moonshot\u2019s 2.8T Open-Weight Model Explained, Specs, Pricing and Benchmarks",
   "published_at": "2026-07-16T23:59:00",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-27",
   "access_type": null,
   "license": null,
   "significance": "At 2.8 trillion total parameters, with 104 billion active per token, K3 is the largest open model anyone has released.",
   "organization_evidence": [
    "Moonshot AI published the full **Kimi K3** weights on **July 27, 2026**, eleven days after the model went live in the Kimi app.",
    "| Developer | Moonshot AI |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Moonshot AI published the full **Kimi K3** weights on **July 27, 2026**, eleven days after the model went live in the Kimi app.",
    "| Released | July 16, 2026 (weights July 27, 2026) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "At **2.8 trillion** total parameters, with **104 billion** active per token, K3 is the largest open model anyone has released.",
    "Moonshot AI published the full **Kimi K3** weights on **July 27, 2026**, eleven days after the model went live in the Kimi app."
   ]
  },
  {
   "url": "https://venturebeat.com/technology/liquid-ais-smallest-model-yet-lfm2-5-230m-beats-models-4x-its-size-at-data-extraction-can-run-anywhere",
   "title": "Liquid AI's smallest model yet LFM2.5-230M beats models 4X its size at data extraction, can run 'anywhere'",
   "published_at": "2026-06-25T23:31:23",
   "organization": "Liquid AI",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "A 230-million-parameter foundation model explicitly designed for on-device agentic workflows, outperforming models more than 4X its size on selected benchmarks.",
   "organization_evidence": [
    "Liquid AI, founded by former MIT computer scientists, today released its smallest AI language model yet, [LFM2.5-230M](https://www.liquid.ai/blog/lfm2-5-230m), and enterprises would do well to consider it for their uses in data extraction and local deployment on smartphones, laptops and robotics."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "This is a 230-million-parameter foundation model explicitly designed for on-device agentic workflows, and as Liquid states in its release blog post, that small size makes it possible to run nearly \"anywhere.\"",
    "According to Liquid, it also outperforms models more than 4X its size on selected benchmarks, specifically doing better at data extraction than the 800 million parameter count Alibaba Qwen3.5-0.8B (Instruct) and 1-billion parameter Google Gemma 3 1B."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/glm-5-1-non-reasoning-vs-gpt-5-3-codex",
   "title": "GLM-5.1 (Non-reasoning) vs GPT-5.3 Codex (xhigh): Model Comparison",
   "published_at": null,
   "organization": "Z AI",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "A major large language model with a 200k token context window, image input support, and open source weights released in April 2026.",
   "organization_evidence": [
    "# GLM-5.1 (Non-reasoning) vs. GPT-5.3 Codex (xhigh)",
    "| Creator | ![Z AI](/img/logos/zai_small.svg)[](https://z.ai/) | ![OpenAI](/img/logos/openai_small.svg)[](https://openai.com/) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# GLM-5.1 (Non-reasoning) vs. GPT-5.3 Codex (xhigh)",
    "| Release Date | April, 2026 | February, 2026 | GLM-5.1 (Non-reasoning) has a more recent release date than GPT-5.3 Codex (xhigh) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# GLM-5.1 (Non-reasoning) vs. GPT-5.3 Codex (xhigh)",
    "| Context Window | 200k tokens (~300 A4 pages of size 12 Arial font) | 400k tokens (~600 A4 pages of size 12 Arial font) | GLM-5.1 (Non-reasoning) is smaller than GPT-5.3 Codex (xhigh) |",
    "| Release Date | April, 2026 | February, 2026 | GLM-5.1 (Non-reasoning) has a more recent release date than GPT-5.3 Codex (xhigh) |",
    "| Image Input Support | No | Yes | GPT-5.3 Codex (xhigh) has image input support while GLM-5.1 (Non-reasoning) does not |",
    "| Open Source (Weights) | Yes[](https://huggingface.co/zai-org/GLM-5.1) | No | GLM-5.1 (Non-reasoning) is open source while GPT-5.3 Codex (xhigh) is proprietary |"
   ]
  },
  {
   "url": "https://unite.ai/thinking-machines-lab-unveils-inkling-its-first-open-weights-multimodal-ai-model",
   "title": "Thinking Machines Lab Unveils Inkling, Its First Open-Weights Multimodal AI Model",
   "published_at": "2026-07-16T17:31:43",
   "organization": "Thinking Machines Lab",
   "model": "Inkling",
   "release_date": "2026-07-15",
   "access_type": "open-weights",
   "license": "Apache 2.0",
   "significance": "nearly one-trillion-parameter system that can process text, images, and audio",
   "organization_evidence": [
    "Thinking Machines Lab Unveils Inkling, Its First Open-Weights Multimodal AI Model",
    "[Thinking Machines Lab](https://thinkingmachines.ai/) has released [Inkling](https://thinkingmachines.ai/news/introducing-inkling/), its first general-purpose artificial intelligence model, giving developers access to a nearly one-trillion-parameter system that can process text, images, and audio."
   ],
   "model_evidence": [
    "Thinking Machines Lab Unveils Inkling, Its First Open-Weights Multimodal AI Model",
    "[Thinking Machines Lab](https://thinkingmachines.ai/) has released [Inkling](https://thinkingmachines.ai/news/introducing-inkling/), its first general-purpose artificial intelligence model, giving developers access to a nearly one-trillion-parameter system that can process text, images, and audio."
   ],
   "release_date_evidence": [
    "Thinking Machines Lab Unveils Inkling, Its First Open-Weights Multimodal AI Model",
    "The July 15, 2026 launch represents the most significant product milestone yet for the heavily funded AI company founded by former OpenAI Chief Technology Officer [Mira Murati](https://en.wikipedia.org/wiki/Mira_Murati)."
   ],
   "access_type_evidence": [
    "Thinking Machines Lab Unveils Inkling, Its First Open-Weights Multimodal AI Model",
    "[Thinking Machines Lab](https://thinkingmachines.ai/) has released [Inkling](https://thinkingmachines.ai/news/introducing-inkling/), its first general-purpose artificial intelligence model, giving developers access to a nearly one-trillion-parameter system that can process text, images, and audio."
   ],
   "license_evidence": [
    "Thinking Machines Lab Unveils Inkling, Its First Open-Weights Multimodal AI Model",
    "Its [weights can be downloaded through Hugging Face](https://huggingface.co/blog/thinkingmachines-inkling) under an Apache 2.0 license, while the model can also be fine-tuned through Thinking Machines\u2019 Tinker training platform."
   ],
   "significance_evidence": [
    "Thinking Machines Lab Unveils Inkling, Its First Open-Weights Multimodal AI Model",
    "[Thinking Machines Lab](https://thinkingmachines.ai/) has released [Inkling](https://thinkingmachines.ai/news/introducing-inkling/), its first general-purpose artificial intelligence model, giving developers access to a nearly one-trillion-parameter system that can process text, images, and audio."
   ]
  },
  {
   "url": "https://geekrealmhub.com/moonshot-ai-kimi-k3-open-weight",
   "title": "Moonshot AI unveils Kimi K3: Analysis of China\u2019s 2.8T open-weight model",
   "published_at": "2026-07-23T00:06:40",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-16",
   "access_type": null,
   "license": null,
   "significance": "A 2.8 trillion-parameter model that is now the largest open-weight AI system in the world and outperforms leading closed U.S. models.",
   "organization_evidence": [
    "# Moonshot AI unveils Kimi K3: Analysis of China\u2019s 2.8T open-weight model",
    "Chinese startup Moonshot AI released Kimi K3 on July 16, a 2.8 trillion-parameter model that is now the largest open-weight [AI](https://geekrealmhub.com/tag/ai/) system in the world."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Moonshot AI unveils Kimi K3: Analysis of China\u2019s 2.8T open-weight model",
    "Chinese startup Moonshot AI released Kimi K3 on July 16, a 2.8 trillion-parameter model that is now the largest open-weight [AI](https://geekrealmhub.com/tag/ai/) system in the world."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Moonshot AI unveils Kimi K3: Analysis of China\u2019s 2.8T open-weight model",
    "Chinese startup Moonshot AI released Kimi K3 on July 16, a 2.8 trillion-parameter model that is now the largest open-weight [AI](https://geekrealmhub.com/tag/ai/) system in the world."
   ]
  },
  {
   "url": "https://geekrealmhub.com/moonshot-ai-kimi-k3-open-weight",
   "title": "Moonshot AI unveils Kimi K3: Analysis of China\u2019s 2.8T open-weight model",
   "published_at": "2026-07-23T00:06:40",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "Successor to K2 known for strong performances on coding benchmarks and predecessor to K3.",
   "organization_evidence": [
    "# Moonshot AI unveils Kimi K3: Analysis of China\u2019s 2.8T open-weight model",
    "K2, released in July 2025, was already known for strong performances on coding benchmarks; its successor K2.6 followed in April 2026."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Moonshot AI unveils Kimi K3: Analysis of China\u2019s 2.8T open-weight model",
    "K2, released in July 2025, was already known for strong performances on coding benchmarks; its successor K2.6 followed in April 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Moonshot AI unveils Kimi K3: Analysis of China\u2019s 2.8T open-weight model",
    "K2, released in July 2025, was already known for strong performances on coding benchmarks; its successor K2.6 followed in April 2026."
   ]
  },
  {
   "url": "https://techstrong.ai/articles/thinking-machines-launches-inkling-to-challenge-chinese-open-source-dominance",
   "title": "Thinking Machines Launches Inkling to Challenge Chinese Open-Source Dominance",
   "published_at": "2026-07-15T20:50:30",
   "organization": "Thinking Machines",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "A mixture-of-experts model containing 975 billion parameters with 41 billion active parameters, trained on 45 trillion tokens of text, image, audio, and video.",
   "organization_evidence": [
    "Thinking Machines Launches Inkling to Challenge Chinese Open-Source Dominance",
    "Thinking Machines, the artificial intelligence (AI) startup founded by former OpenAI Chief Technology Officer Mira Murati, announced Wednesday the release of Inkling, a massive open-weight AI model."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Thinking Machines Launches Inkling to Challenge Chinese Open-Source Dominance",
    "2.3 min readPublished On: July 15, 2026",
    "Thinking Machines, the artificial intelligence (AI) startup founded by former OpenAI Chief Technology Officer Mira Murati, announced Wednesday the release of Inkling, a massive open-weight AI model."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Thinking Machines Launches Inkling to Challenge Chinese Open-Source Dominance",
    "To bridge this divide, Inkling is designed as a mixture-of-experts (MoE) system containing 975 billion parameters, though it routes only 41 billion active parameters per task to optimize speed and cost."
   ]
  },
  {
   "url": "https://datacamp.com/blog/thinking-machines-inkling",
   "title": "Inkling: Thinking Machines' First Open-Weights Model",
   "published_at": "2026-07-16T11:51:11",
   "organization": "Thinking Machines",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "A Mixture-of-Experts transformer with 975B total parameters, 41B active per token, and a 1M-token context window.",
   "organization_evidence": [
    "Thinking Machines, the lab founded by former OpenAI CTO Mira Murati, has released its first model: Inkling."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "It's the first release in a planned family, released July 15, 2026, alongside a preview of a smaller sibling, Inkling-Small.",
    "What Is Inkling?",
    "Inkling is an open-weights MoE transformer with 975B total parameters and 41B active per token, positioned as a broad base model for [fine-tuning](https://www.datacamp.com/tutorial/fine-tuning-large-language-models) rather than a top-of-leaderboard chatbot."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Inkling is a Mixture-of-Experts (MoE) transformer with 975B total parameters and 41B active, a 1M-token context window, and pretraining on 45 trillion tokens of text, images, audio, and video.",
    "Thinking Machines, the lab founded by former OpenAI CTO Mira Murati, has released its first model: Inkling."
   ]
  },
  {
   "url": "https://github.com/JeevaVenkidu/awesome-open-ai-alternatives",
   "title": "GitHub - JeevaVenkidu/awesome-open-ai-alternatives: A curated list of open-source alternatives to proprietary AI tools \u2014 LLMs, agents, RAG, coding assistants, image/video/voice generation, and local deployment. Updated July 2026.",
   "published_at": null,
   "organization": null,
   "model": null,
   "release_date": "2026-07-27",
   "access_type": null,
   "license": null,
   "significance": "Frontier reasoning",
   "organization_evidence": null,
   "model_evidence": null,
   "release_date_evidence": [
    "# Awesome Open Source AI Alternatives \ud83d\ude80",
    "## \ud83d\udcda Table of Contents",
    "## \ud83e\udde0 Large Language Models (LLMs)",
    "| Model | Architecture | Best For | License | VRAM (Approx) | Context | Release Date |",
    "| **Kimi K3** \ud83d\udd25 *(coming July 27)* | 2.8T MoE / 16B active | Frontier reasoning | [![MIT](https://camo.githubusercontent.com/7013272bd27ece47364536a221edb554cd69683b68a46fc0ee96881174c4214c/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f6c6963656e73652d4d49542d626c75652e737667)](https://opensource.org/licenses/MIT) | Multi-GPU | TBD | 2026-07-27 planned |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Awesome Open Source AI Alternatives \ud83d\ude80",
    "## \ud83d\udcda Table of Contents",
    "## \ud83e\udde0 Large Language Models (LLMs)",
    "| Model | Architecture | Best For | License | VRAM (Approx) | Context | Release Date |",
    "| **Kimi K3** \ud83d\udd25 *(coming July 27)* | 2.8T MoE / 16B active | Frontier reasoning | [![MIT](https://camo.githubusercontent.com/7013272bd27ece47364536a221edb554cd69683b68a46fc0ee96881174c4214c/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f6c6963656e73652d4d49542d626c75652e737667)](https://opensource.org/licenses/MIT) | Multi-GPU | TBD | 2026-07-27 planned |"
   ]
  },
  {
   "url": "https://blockgeni.com/thinking-machines-inkling-open-weight-ai-model-mira",
   "title": "Mira Murati's Thinking Machines Releases Inkling, an Open-Weight AI Model Built for Customization - Blockgeni",
   "published_at": "2026-07-28T08:01:30",
   "organization": "Thinking Machines",
   "model": null,
   "release_date": "2026-07-16",
   "access_type": null,
   "license": null,
   "significance": "trained from scratch with full open weights, paired with a companion customization tool Tinker, and designed for a wide scope across agentic tasks, reasoning, coding, instruction-following, factuality, vision, and audio",
   "organization_evidence": [
    "Mira Murati\u2019s Thinking Machines Releases Inkling, an Open-Weight AI Model Built for Customization",
    "Mira Murati, the former chief technology officer of OpenAI, has delivered the first public product from her AI startup Thinking Machines Lab: a fully open-weight model named Inkling, designed to be customized by businesses and developers rather than used as a fixed, proprietary service."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Mira Murati\u2019s Thinking Machines Releases Inkling, an Open-Weight AI Model Built for Customization",
    "The release of Inkling on July 16, 2026, is the first concrete signal of what that work produced."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Mira Murati\u2019s Thinking Machines Releases Inkling, an Open-Weight AI Model Built for Customization",
    "According to the company, Inkling is \u201ca model we trained from scratch with the full weights available, so that people can make it their own.\u201d",
    "To make that customization practical, Thinking Machines has paired Inkling with a companion tool called Tinker, which allows users to adjust the model\u2019s weights directly \u2014 effectively reshaping how the AI reasons and responds without requiring full retraining from scratch.",
    "The company describes Inkling as \u201cdesigned to be broad,\u201d trained across what it lists as agentic tasks, reasoning, coding, instruction-following, factuality, vision, and audio \u2014 a deliberately wide scope."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/gemma-4-e4b",
   "title": "Gemma 4 E4B - Intelligence, Performance & Price Analysis",
   "published_at": null,
   "organization": "Google",
   "model": null,
   "release_date": "2026-04-03",
   "access_type": null,
   "license": null,
   "significance": "An 8B parameters (4.5B active) reasoning model with text, image, speech, and video input, and 130k context window.",
   "organization_evidence": [
    "Gemma 4 E4B (Reasoning) Intelligence, Performance & Price Analysis",
    "Gemma 4 E4B (Reasoning) was created by Google."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Gemma 4 E4B (Reasoning) Intelligence, Performance & Price Analysis",
    "Released April 2026",
    "Gemma 4 E4B (Reasoning) was released on April 3, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Gemma 4 E4B (Reasoning) Intelligence, Performance & Price Analysis",
    "Gemma 4 E4B (Reasoning) is above average in intelligence and reasonably priced when comparing to other open weight models of similar size. The model supports text, image, speech, and video input, outputs text, and has a 128k tokens context window.",
    "Gemma 4 E4B (Reasoning) has 8 billion parameters (4.5 billion active)."
   ]
  },
  {
   "url": "https://strongmocha.com/ai-infrastructure-data-centers/fast-and-furious-china-s-signal-releases-four-frontier-ai-models-quickly",
   "title": "Fast And Furious: China\u2019s Signal Releases Four Frontier AI Models Quickly - StrongMocha",
   "published_at": "2026-07-15T06:14:42",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor",
   "organization_evidence": [
    "From April 24 to June 15, 2026, Chinese labs launched four major open-weight AI models: **DeepSeek V4**, **MiniMax M3**, **Kimi K2.7-Code**, and **GLM-5.2**.",
    "The Chinese open AI ecosystem now comprises four distinct labs\u2014DeepSeek, Z.ai, Moonshot, and Alibaba\u2014each with unique strategic focuses, from cost-efficiency to long-horizon stability and self-hosting capabilities."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "From April 24 to June 15, 2026, Chinese labs launched four major open-weight AI models: **DeepSeek V4**, **MiniMax M3**, **Kimi K2.7-Code**, and **GLM-5.2**.",
    "### The production line \u2014 spring 2026",
    "APR 24",
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "### The production line \u2014 spring 2026",
    "APR 24",
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor"
   ]
  },
  {
   "url": "https://strongmocha.com/ai-infrastructure-data-centers/fast-and-furious-china-s-signal-releases-four-frontier-ai-models-quickly",
   "title": "Fast And Furious: China\u2019s Signal Releases Four Frontier AI Models Quickly - StrongMocha",
   "published_at": "2026-07-15T06:14:42",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "cheap 1M-token context, native multimodal, modified-MIT",
   "organization_evidence": [
    "From April 24 to June 15, 2026, Chinese labs launched four major open-weight AI models: **DeepSeek V4**, **MiniMax M3**, **Kimi K2.7-Code**, and **GLM-5.2**."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "From April 24 to June 15, 2026, Chinese labs launched four major open-weight AI models: **DeepSeek V4**, **MiniMax M3**, **Kimi K2.7-Code**, and **GLM-5.2**.",
    "### The production line \u2014 spring 2026",
    "JUN 01"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "### The production line \u2014 spring 2026",
    "JUN 01"
   ]
  },
  {
   "url": "https://strongmocha.com/ai-infrastructure-data-centers/fast-and-furious-china-s-signal-releases-four-frontier-ai-models-quickly",
   "title": "Fast And Furious: China\u2019s Signal Releases Four Frontier AI Models Quickly - StrongMocha",
   "published_at": "2026-07-15T06:14:42",
   "organization": "Moonshot",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "agent-run specialist, ~30% fewer thinking tokens than K2.6",
   "organization_evidence": [
    "From April 24 to June 15, 2026, Chinese labs launched four major open-weight AI models: **DeepSeek V4**, **MiniMax M3**, **Kimi K2.7-Code**, and **GLM-5.2**.",
    "The Chinese open AI ecosystem now comprises four distinct labs\u2014DeepSeek, Z.ai, Moonshot, and Alibaba\u2014each with unique strategic focuses, from cost-efficiency to long-horizon stability and self-hosting capabilities."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "From April 24 to June 15, 2026, Chinese labs launched four major open-weight AI models: **DeepSeek V4**, **MiniMax M3**, **Kimi K2.7-Code**, and **GLM-5.2**.",
    "### The production line \u2014 spring 2026",
    "JUN 13"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "### The production line \u2014 spring 2026",
    "JUN 13"
   ]
  },
  {
   "url": "https://strongmocha.com/ai-infrastructure-data-centers/fast-and-furious-china-s-signal-releases-four-frontier-ai-models-quickly",
   "title": "Fast And Furious: China\u2019s Signal Releases Four Frontier AI Models Quickly - StrongMocha",
   "published_at": "2026-07-15T06:14:42",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "753B MoE, MIT, top open-weight on Artificial Analysis index",
   "organization_evidence": [
    "From April 24 to June 15, 2026, Chinese labs launched four major open-weight AI models: **DeepSeek V4**, **MiniMax M3**, **Kimi K2.7-Code**, and **GLM-5.2**.",
    "The Chinese open AI ecosystem now comprises four distinct labs\u2014DeepSeek, Z.ai, Moonshot, and Alibaba\u2014each with unique strategic focuses, from cost-efficiency to long-horizon stability and self-hosting capabilities."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "From April 24 to June 15, 2026, Chinese labs launched four major open-weight AI models: **DeepSeek V4**, **MiniMax M3**, **Kimi K2.7-Code**, and **GLM-5.2**.",
    "### The production line \u2014 spring 2026",
    "JUN 13\u201316"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "### The production line \u2014 spring 2026",
    "JUN 13\u201316"
   ]
  },
  {
   "url": "https://digitalapplied.com/blog/nvidia-nemotron-3-ultra-550b-open-reasoning-model-2026",
   "title": "NVIDIA Nemotron 3 Ultra: 550B Open Reasoning Model Live",
   "published_at": "2026-06-04T23:00:00",
   "organization": "NVIDIA",
   "model": null,
   "release_date": "2026-06-04",
   "access_type": null,
   "license": null,
   "significance": "a 550-billion-parameter open Mixture-of-Experts reasoning model that ships not just weights but training data and recipes under a permissive Linux Foundation license",
   "organization_evidence": [
    "NVIDIA released Nemotron 3 Ultra on June 4, 2026 \u2014 a 550-billion-parameter open Mixture-of-Experts reasoning model that ships not just weights but training data and recipes under a permissive Linux Foundation license."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "NVIDIA released Nemotron 3 Ultra on June 4, 2026 \u2014 a 550-billion-parameter open Mixture-of-Experts reasoning model that ships not just weights but training data and recipes under a permissive Linux Foundation license."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "NVIDIA released Nemotron 3 Ultra on June 4, 2026 \u2014 a 550-billion-parameter open Mixture-of-Experts reasoning model that ships not just weights but training data and recipes under a permissive Linux Foundation license."
   ]
  },
  {
   "url": "https://huggingface.co/blog/nyxia/hybrid-recurrent-attention-weight-sharing",
   "title": "Auron: Depth-Efficient Language Models via Hybrid Recurrent-Attention Weight Sharing",
   "published_at": "2026-04-16T19:35:14",
   "organization": "Cube Digital Media Ltd / Soulkyn",
   "model": null,
   "release_date": "2026-03",
   "access_type": null,
   "license": null,
   "significance": "Depth-efficient hybrid language models featuring Ouroboros weight sharing and the Chimera Topology at multiple scales (279M, 510M, and 1.1B parameters).",
   "organization_evidence": [
    "# Auron: Depth-Efficient Language Models via Hybrid Recurrent-Attention Weight Sharing",
    "**Author:** Florian Gasquez ([florian@soulkyn.com](mailto:florian@soulkyn.com)) **Affiliation:** Cube Digital Media Ltd / Soulkyn **Links:**[fyx.jp](https://fyx.jp) | [soulkyn.com](https://soulkyn.com) | [GitHub](https://github.com/Fy-/Auron) | [HuggingFace](https://huggingface.co/nyxia)**Date:** March 2026 **License:** CC BY 4.0"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Auron: Depth-Efficient Language Models via Hybrid Recurrent-Attention Weight Sharing",
    "**Author:** Florian Gasquez ([florian@soulkyn.com](mailto:florian@soulkyn.com)) **Affiliation:** Cube Digital Media Ltd / Soulkyn **Links:**[fyx.jp](https://fyx.jp) | [soulkyn.com](https://soulkyn.com) | [GitHub](https://github.com/Fy-/Auron) | [HuggingFace](https://huggingface.co/nyxia)**Date:** March 2026 **License:** CC BY 4.0"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Auron: Depth-Efficient Language Models via Hybrid Recurrent-Attention Weight Sharing",
    "We train three model scales -- 279M, 510M, and 1.1B total parameters -- on a mixed pretraining dataset of 5B tokens across four sources, all to completion (250K steps with WSD warmdown)."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/muse-glimmer",
   "title": "Muse Glimmer (high) - Intelligence, Performance & Price Analysis",
   "published_at": null,
   "organization": "Meta",
   "model": null,
   "release_date": "2026-08-10",
   "access_type": null,
   "license": null,
   "significance": "A 30B reasoning model with text and image input capabilities, scoring 35 on the Artificial Analysis Intelligence Index.",
   "organization_evidence": [
    "Muse Glimmer (high) Intelligence, Performance & Price Analysis",
    "Muse Glimmer (high) was created by Meta."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Muse Glimmer (high) Intelligence, Performance & Price Analysis",
    "Muse Glimmer (high) was released on August 10, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Muse Glimmer (high) Intelligence, Performance & Price Analysis",
    "| Total parameters | 30B |",
    "Muse Glimmer (high) scores 35 on the Artificial Analysis Intelligence Index, placing it well above average among comparable models (median: 9).",
    "Yes, Muse Glimmer (high) is a reasoning model. It uses extended thinking or chain-of-thought reasoning to work through complex problems before providing an answer."
   ]
  },
  {
   "url": "https://bestcadpapers.com/art-and-technology/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - Best CAD papers",
   "published_at": "2026-07-15T13:24:05",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor",
   "organization_evidence": [
    "From late April to mid-June 2026, Chinese labs released four major open-weight models: **DeepSeek V4** on April 24, **MiniMax M3** on June 1, and **Kimi K2.7-Code** and **GLM-5.2** within days of each other in mid-June.",
    "## Four Frontier-Class Open Models in Eight Weeks"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "From late April to mid-June 2026, Chinese labs released four major open-weight models: **DeepSeek V4** on April 24, **MiniMax M3** on June 1, and **Kimi K2.7-Code** and **GLM-5.2** within days of each other in mid-June.",
    "## Four Frontier-Class Open Models in Eight Weeks"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor",
    "## Four Frontier-Class Open Models in Eight Weeks",
    "### The production line \u2014 spring 2026"
   ]
  },
  {
   "url": "https://bestcadpapers.com/art-and-technology/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - Best CAD papers",
   "published_at": "2026-07-15T13:24:05",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "cheap 1M-token context, native multimodal, modified-MIT",
   "organization_evidence": [
    "From late April to mid-June 2026, Chinese labs released four major open-weight models: **DeepSeek V4** on April 24, **MiniMax M3** on June 1, and **Kimi K2.7-Code** and **GLM-5.2** within days of each other in mid-June.",
    "## Four Frontier-Class Open Models in Eight Weeks"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "From late April to mid-June 2026, Chinese labs released four major open-weight models: **DeepSeek V4** on April 24, **MiniMax M3** on June 1, and **Kimi K2.7-Code** and **GLM-5.2** within days of each other in mid-June.",
    "## Four Frontier-Class Open Models in Eight Weeks"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "**MiniMax M3**cheap 1M-token context, native multimodal, modified-MIT",
    "## Four Frontier-Class Open Models in Eight Weeks",
    "### The production line \u2014 spring 2026"
   ]
  },
  {
   "url": "https://bestcadpapers.com/art-and-technology/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - Best CAD papers",
   "published_at": "2026-07-15T13:24:05",
   "organization": "Moonshot",
   "model": null,
   "release_date": "2026-06",
   "access_type": null,
   "license": null,
   "significance": "agent-run specialist, ~30% fewer thinking tokens than K2.6",
   "organization_evidence": [
    "**Kimi K2.7-Code (Moonshot)**agent-run specialist, ~30% fewer thinking tokens than K2.6",
    "## Four Frontier-Class Open Models in Eight Weeks",
    "### The production line \u2014 spring 2026"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "From late April to mid-June 2026, Chinese labs released four major open-weight models: **DeepSeek V4** on April 24, **MiniMax M3** on June 1, and **Kimi K2.7-Code** and **GLM-5.2** within days of each other in mid-June.",
    "## Four Frontier-Class Open Models in Eight Weeks"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "**Kimi K2.7-Code (Moonshot)**agent-run specialist, ~30% fewer thinking tokens than K2.6",
    "## Four Frontier-Class Open Models in Eight Weeks",
    "### The production line \u2014 spring 2026"
   ]
  },
  {
   "url": "https://bestcadpapers.com/art-and-technology/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - Best CAD papers",
   "published_at": "2026-07-15T13:24:05",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-06",
   "access_type": null,
   "license": null,
   "significance": "753B MoE, MIT, top open-weight on Artificial Analysis index",
   "organization_evidence": [
    "**GLM-5.2 (Z.ai)**753B MoE, MIT, top open-weight on Artificial Analysis index",
    "## Four Frontier-Class Open Models in Eight Weeks",
    "### The production line \u2014 spring 2026"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "From late April to mid-June 2026, Chinese labs released four major open-weight models: **DeepSeek V4** on April 24, **MiniMax M3** on June 1, and **Kimi K2.7-Code** and **GLM-5.2** within days of each other in mid-June.",
    "## Four Frontier-Class Open Models in Eight Weeks"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "**GLM-5.2 (Z.ai)**753B MoE, MIT, top open-weight on Artificial Analysis index",
    "## Four Frontier-Class Open Models in Eight Weeks",
    "### The production line \u2014 spring 2026"
   ]
  },
  {
   "url": "http://wan27.org/blog/inkling-thinking-machines-explained",
   "title": "What Is Inkling? Thinking Machines Lab's First Open Model Explained (2026) | Wan 2.7",
   "published_at": "2026-04-06T00:00:00",
   "organization": "Thinking Machines Lab",
   "model": "Inkling",
   "release_date": "2026-07-15",
   "access_type": "open-weight",
   "license": null,
   "significance": "a 975B open-weight MoE model (41B active) from Thinking Machines Lab",
   "organization_evidence": [
    "What Is Inkling? Thinking Machines Lab's First Open Model Explained (2026)",
    "On July 15, 2026, Thinking Machines Lab \u2014 the roughly 200-person startup that ex-OpenAI CTO Mira Murati has kept in near-total stealth for over a year \u2014 dropped **Inkling**, a 975-billion-parameter open-weight model, with the full weights on Hugging Face the same afternoon."
   ],
   "model_evidence": [
    "What Is Inkling? Thinking Machines Lab's First Open Model Explained (2026)",
    "On July 15, 2026, Thinking Machines Lab \u2014 the roughly 200-person startup that ex-OpenAI CTO Mira Murati has kept in near-total stealth for over a year \u2014 dropped **Inkling**, a 975-billion-parameter open-weight model, with the full weights on Hugging Face the same afternoon."
   ],
   "release_date_evidence": [
    "What Is Inkling? Thinking Machines Lab's First Open Model Explained (2026)",
    "On July 15, 2026, Thinking Machines Lab \u2014 the roughly 200-person startup that ex-OpenAI CTO Mira Murati has kept in near-total stealth for over a year \u2014 dropped **Inkling**, a 975-billion-parameter open-weight model, with the full weights on Hugging Face the same afternoon."
   ],
   "access_type_evidence": [
    "What Is Inkling? Thinking Machines Lab's First Open Model Explained (2026)",
    "On July 15, 2026, Thinking Machines Lab \u2014 the roughly 200-person startup that ex-OpenAI CTO Mira Murati has kept in near-total stealth for over a year \u2014 dropped **Inkling**, a 975-billion-parameter open-weight model, with the full weights on Hugging Face the same afternoon."
   ],
   "license_evidence": null,
   "significance_evidence": [
    "What Is Inkling? Thinking Machines Lab's First Open Model Explained (2026)",
    "Inkling is a 975B open-weight MoE model (41B active) from Thinking Machines Lab."
   ]
  },
  {
   "url": "https://tensorfeed.ai/timeline",
   "title": "AI Timeline | Major Milestones & Model Releases",
   "published_at": "2026-05-13T00:00:00",
   "organization": "Meituan",
   "model": null,
   "release_date": "2026-06-30",
   "access_type": null,
   "license": null,
   "significance": "1.6 trillion parameter mixture-of-experts coding model",
   "organization_evidence": [
    "# AI Timeline",
    "A chronological record of major AI model releases, industry milestones, and pivotal events from 2024 to present.",
    "### Meituan Open Sources LongCat-2.0, a 1.6T Agentic Coding Model Trained on Chinese Chips",
    "Meituan",
    "Meituan open sourced LongCat-2.0 on June 30, 2026, a 1.6 trillion parameter mixture-of-experts coding model published to GitHub and Hugging Face under an MIT license."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# AI Timeline",
    "A chronological record of major AI model releases, industry milestones, and pivotal events from 2024 to present.",
    "### Meituan Open Sources LongCat-2.0, a 1.6T Agentic Coding Model Trained on Chinese Chips",
    "Meituan",
    "Meituan open sourced LongCat-2.0 on June 30, 2026, a 1.6 trillion parameter mixture-of-experts coding model published to GitHub and Hugging Face under an MIT license."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# AI Timeline",
    "A chronological record of major AI model releases, industry milestones, and pivotal events from 2024 to present.",
    "### Meituan Open Sources LongCat-2.0, a 1.6T Agentic Coding Model Trained on Chinese Chips",
    "Meituan",
    "Meituan open sourced LongCat-2.0 on June 30, 2026, a 1.6 trillion parameter mixture-of-experts coding model published to GitHub and Hugging Face under an MIT license."
   ]
  },
  {
   "url": "https://tensorfeed.ai/timeline",
   "title": "AI Timeline | Major Milestones & Model Releases",
   "published_at": "2026-05-13T00:00:00",
   "organization": "MiniMax",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "open-weight coding and agentic model built on MiniMax Sparse Attention",
   "organization_evidence": [
    "# AI Timeline",
    "A chronological record of major AI model releases, industry milestones, and pivotal events from 2024 to present.",
    "### MiniMax Releases M3 Open-Weight Coding Model",
    "MiniMax",
    "MiniMax launched M3, an open-weight coding and agentic model built on MiniMax Sparse Attention, which replaces full attention with KV-block selection and cuts per-token compute at 1M context to roughly one twentieth of the previous generation."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# AI Timeline",
    "A chronological record of major AI model releases, industry milestones, and pivotal events from 2024 to present.",
    "### MiniMax Releases M3 Open-Weight Coding Model",
    "MiniMax",
    "MiniMax launched M3, an open-weight coding and agentic model built on MiniMax Sparse Attention, which replaces full attention with KV-block selection and cuts per-token compute at 1M context to roughly one twentieth of the previous generation."
   ]
  },
  {
   "url": "https://tensorfeed.ai/timeline",
   "title": "AI Timeline | Major Milestones & Model Releases",
   "published_at": "2026-05-13T00:00:00",
   "organization": "Cohere",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "218B-total / 25B-active mixture-of-experts model with a 128K input context",
   "organization_evidence": [
    "# AI Timeline",
    "A chronological record of major AI model releases, industry milestones, and pivotal events from 2024 to present.",
    "### Cohere Releases Command A+ Open Source Under Apache 2.0",
    "Cohere",
    "Cohere released Command A+, a 218B-total / 25B-active mixture-of-experts model with a 128K input context and 64K max generation, under an Apache 2.0 license."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# AI Timeline",
    "A chronological record of major AI model releases, industry milestones, and pivotal events from 2024 to present.",
    "### Cohere Releases Command A+ Open Source Under Apache 2.0",
    "Cohere",
    "Cohere released Command A+, a 218B-total / 25B-active mixture-of-experts model with a 128K input context and 64K max generation, under an Apache 2.0 license."
   ]
  },
  {
   "url": "https://tensorfeed.ai/timeline",
   "title": "AI Timeline | Major Milestones & Model Releases",
   "published_at": "2026-05-13T00:00:00",
   "organization": "NVIDIA",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "open-weight multimodal model that processes text, image, video, and audio in a unified sequence",
   "organization_evidence": [
    "# AI Timeline",
    "A chronological record of major AI model releases, industry milestones, and pivotal events from 2024 to present.",
    "### NVIDIA Releases Nemotron 3 Nano Omni",
    "NVIDIA",
    "NVIDIA ships Nemotron 3 Nano Omni 30B-A3B-Reasoning, an open-weight multimodal model that processes text, image, video, and audio in a unified sequence."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# AI Timeline",
    "A chronological record of major AI model releases, industry milestones, and pivotal events from 2024 to present.",
    "### NVIDIA Releases Nemotron 3 Nano Omni",
    "NVIDIA",
    "NVIDIA ships Nemotron 3 Nano Omni 30B-A3B-Reasoning, an open-weight multimodal model that processes text, image, video, and audio in a unified sequence."
   ]
  },
  {
   "url": "https://tensorfeed.ai/timeline",
   "title": "AI Timeline | Major Milestones & Model Releases",
   "published_at": "2026-05-13T00:00:00",
   "organization": "DeepSeek",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "V4 Pro (1.6T parameters, 49B active) and V4 Flash (284B total, 13B active)",
   "organization_evidence": [
    "# AI Timeline",
    "A chronological record of major AI model releases, industry milestones, and pivotal events from 2024 to present.",
    "### DeepSeek V4 Pro & Flash Released",
    "DeepSeek",
    "DeepSeek releases V4 Pro (1.6T parameters, 49B active) and V4 Flash (284B total, 13B active) under the MIT license with native 1M context windows."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# AI Timeline",
    "A chronological record of major AI model releases, industry milestones, and pivotal events from 2024 to present.",
    "### DeepSeek V4 Pro & Flash Released",
    "DeepSeek",
    "DeepSeek releases V4 Pro (1.6T parameters, 49B active) and V4 Flash (284B total, 13B active) under the MIT license with native 1M context windows."
   ]
  },
  {
   "url": "https://n8nlab.io/news/huawei-pangu-ascend-frontier",
   "title": "No Nvidia Required: Huawei Ships 505B Open-Weight AI Model",
   "published_at": "2026-08-02T18:11:11",
   "organization": "Huawei",
   "model": null,
   "release_date": "2026-07-31",
   "access_type": null,
   "license": null,
   "significance": "a 505-billion-parameter Mixture-of-Experts language model that completed its full pretraining run on Huawei's own Ascend 910B NPUs",
   "organization_evidence": [
    "Analysis \u00b7 AI Models2 August 20266 min read",
    "# No Nvidia Required: Huawei Ships 505B Open-Weight AI Model",
    "On July 31, Huawei released the weights, inference code, and technical report for openPangu-2.0-Pro \u2014 a 505-billion-parameter Mixture-of-Experts language model that completed its full pretraining run on Huawei's own Ascend 910B NPUs."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Analysis \u00b7 AI Models2 August 20266 min read",
    "# No Nvidia Required: Huawei Ships 505B Open-Weight AI Model",
    "On July 31, Huawei released the weights, inference code, and technical report for openPangu-2.0-Pro \u2014 a 505-billion-parameter Mixture-of-Experts language model that completed its full pretraining run on Huawei's own Ascend 910B NPUs."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Analysis \u00b7 AI Models2 August 20266 min read",
    "# No Nvidia Required: Huawei Ships 505B Open-Weight AI Model",
    "On July 31, Huawei released the weights, inference code, and technical report for openPangu-2.0-Pro \u2014 a 505-billion-parameter Mixture-of-Experts language model that completed its full pretraining run on Huawei's own Ascend 910B NPUs."
   ]
  },
  {
   "url": "https://n8nlab.io/news/huawei-pangu-ascend-frontier",
   "title": "No Nvidia Required: Huawei Ships 505B Open-Weight AI Model",
   "published_at": "2026-08-02T18:11:11",
   "organization": "Huawei",
   "model": "openPangu-2.0-Flash",
   "release_date": "2026-06-30",
   "access_type": "open-source",
   "license": "OpenPangu Model License Agreement 2.0",
   "significance": "a 92-billion-parameter model that is part of the openPangu 2.0 open-source release",
   "organization_evidence": [
    "Analysis \u00b7 AI Models2 August 20266 min read",
    "# No Nvidia Required: Huawei Ships 505B Open-Weight AI Model",
    "At HDC 2026 in June, Huawei executive Richard Yu announced openPangu 2.0 would open-source seven major components \u2014 pretraining code, post-training code, training operators, weights, and inference code \u2014 starting June 30."
   ],
   "model_evidence": [
    "Analysis \u00b7 AI Models2 August 20266 min read",
    "# No Nvidia Required: Huawei Ships 505B Open-Weight AI Model",
    "The 92-billion-parameter openPangu-2.0-Flash shipped first, followed by the full Pro variant on July 31."
   ],
   "release_date_evidence": [
    "Analysis \u00b7 AI Models2 August 20266 min read",
    "# No Nvidia Required: Huawei Ships 505B Open-Weight AI Model",
    "At HDC 2026 in June, Huawei executive Richard Yu announced openPangu 2.0 would open-source seven major components \u2014 pretraining code, post-training code, training operators, weights, and inference code \u2014 starting June 30.",
    "The 92-billion-parameter openPangu-2.0-Flash shipped first, followed by the full Pro variant on July 31."
   ],
   "access_type_evidence": [
    "Analysis \u00b7 AI Models2 August 20266 min read",
    "# No Nvidia Required: Huawei Ships 505B Open-Weight AI Model",
    "At HDC 2026 in June, Huawei executive Richard Yu announced openPangu 2.0 would open-source seven major components \u2014 pretraining code, post-training code, training operators, weights, and inference code \u2014 starting June 30."
   ],
   "license_evidence": [
    "Analysis \u00b7 AI Models2 August 20266 min read",
    "## Builder Impact",
    "Third, the license contains a landmine.",
    "Huawei distributes the model under its custom OpenPangu Model License Agreement 2.0, which explicitly bars access, deployment, or use within the European Union."
   ],
   "significance_evidence": [
    "Analysis \u00b7 AI Models2 August 20266 min read",
    "# No Nvidia Required: Huawei Ships 505B Open-Weight AI Model",
    "The 92-billion-parameter openPangu-2.0-Flash shipped first, followed by the full Pro variant on July 31."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/gemma-4-26b-a4b-vs-gpt-5-4-nano",
   "title": "Gemma 4 26B A4B (Reasoning) vs GPT-5.4 nano (xhigh): Model Comparison",
   "published_at": null,
   "organization": "Google",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "26B parameters, 256k context window, reasoning model with open source weights",
   "organization_evidence": [
    "Comparison between Gemma 4 26B A4B (Reasoning) and GPT-5.4 nano (xhigh) across intelligence, price, speed, context window and more.",
    "| Creator | ![Google](/img/logos/google_small.svg)[](https://deepmind.google/technologies/gemini/) | ![OpenAI](/img/logos/openai_small.svg)[](https://openai.com/) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Comparison between Gemma 4 26B A4B (Reasoning) and GPT-5.4 nano (xhigh) across intelligence, price, speed, context window and more.",
    "| Release Date | April, 2026 | March, 2026 | Gemma 4 26B A4B (Reasoning) has a more recent release date than GPT-5.4 nano (xhigh) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Comparison between Gemma 4 26B A4B (Reasoning) and GPT-5.4 nano (xhigh) across intelligence, price, speed, context window and more.",
    "| Context Window | 256k tokens (~384 A4 pages of size 12 Arial font) | 400k tokens (~600 A4 pages of size 12 Arial font) | Gemma 4 26B A4B (Reasoning) is smaller than GPT-5.4 nano (xhigh) |",
    "| Open Source (Weights) | Yes[](https://huggingface.co/google/gemma-4-26B-A4B-it) | No | Gemma 4 26B A4B (Reasoning) is open source while GPT-5.4 nano (xhigh) is proprietary |"
   ]
  },
  {
   "url": "https://press-report.net/ai-news/signal-s-ai-milestone-four-frontier-models-launched-in-just-two-months",
   "title": "Signal\u2019s AI Milestone: Four Frontier Models Launched In Just Two Months - press-report.net",
   "published_at": "2026-07-15T06:14:40",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor",
   "organization_evidence": [
    "These releases, from late April to mid-June 2026, include DeepSeek V4, MiniMax M3, Kimi K2.7-Code, and GLM-5.2, all downloadable and most under permissive licenses.",
    "Signal\u2019s AI Milestone: Four Frontier Models Launched In Just Two Months",
    "### The production line \u2014 spring 2026",
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The models include DeepSeek V4 on April 24, MiniMax M3 on June 1, and Kimi K2.7-Code and GLM-5.2 within days of each other in mid-June.",
    "Signal\u2019s AI Milestone: Four Frontier Models Launched In Just Two Months",
    "### The production line \u2014 spring 2026",
    "APR 24"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor",
    "Signal\u2019s AI Milestone: Four Frontier Models Launched In Just Two Months",
    "### The production line \u2014 spring 2026"
   ]
  },
  {
   "url": "https://press-report.net/ai-news/signal-s-ai-milestone-four-frontier-models-launched-in-just-two-months",
   "title": "Signal\u2019s AI Milestone: Four Frontier Models Launched In Just Two Months - press-report.net",
   "published_at": "2026-07-15T06:14:40",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "cheap 1M-token context, native multimodal, modified-MIT",
   "organization_evidence": [
    "These releases, from late April to mid-June 2026, include DeepSeek V4, MiniMax M3, Kimi K2.7-Code, and GLM-5.2, all downloadable and most under permissive licenses.",
    "Signal\u2019s AI Milestone: Four Frontier Models Launched In Just Two Months",
    "### The production line \u2014 spring 2026",
    "**MiniMax M3**cheap 1M-token context, native multimodal, modified-MIT"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The models include DeepSeek V4 on April 24, MiniMax M3 on June 1, and Kimi K2.7-Code and GLM-5.2 within days of each other in mid-June.",
    "Signal\u2019s AI Milestone: Four Frontier Models Launched In Just Two Months",
    "### The production line \u2014 spring 2026",
    "JUN 01"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "**MiniMax M3**cheap 1M-token context, native multimodal, modified-MIT",
    "Signal\u2019s AI Milestone: Four Frontier Models Launched In Just Two Months",
    "### The production line \u2014 spring 2026"
   ]
  },
  {
   "url": "https://press-report.net/ai-news/signal-s-ai-milestone-four-frontier-models-launched-in-just-two-months",
   "title": "Signal\u2019s AI Milestone: Four Frontier Models Launched In Just Two Months - press-report.net",
   "published_at": "2026-07-15T06:14:40",
   "organization": "Moonshot",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "agent-run specialist, ~30% fewer thinking tokens than K2.6",
   "organization_evidence": [
    "These releases, from late April to mid-June 2026, include DeepSeek V4, MiniMax M3, Kimi K2.7-Code, and GLM-5.2, all downloadable and most under permissive licenses.",
    "Signal\u2019s AI Milestone: Four Frontier Models Launched In Just Two Months",
    "### The production line \u2014 spring 2026",
    "**Kimi K2.7-Code (Moonshot)**agent-run specialist, ~30% fewer thinking tokens than K2.6"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The models include DeepSeek V4 on April 24, MiniMax M3 on June 1, and Kimi K2.7-Code and GLM-5.2 within days of each other in mid-June.",
    "Signal\u2019s AI Milestone: Four Frontier Models Launched In Just Two Months",
    "### The production line \u2014 spring 2026",
    "JUN 13"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "**Kimi K2.7-Code (Moonshot)**agent-run specialist, ~30% fewer thinking tokens than K2.6",
    "Signal\u2019s AI Milestone: Four Frontier Models Launched In Just Two Months",
    "### The production line \u2014 spring 2026"
   ]
  },
  {
   "url": "https://press-report.net/ai-news/signal-s-ai-milestone-four-frontier-models-launched-in-just-two-months",
   "title": "Signal\u2019s AI Milestone: Four Frontier Models Launched In Just Two Months - press-report.net",
   "published_at": "2026-07-15T06:14:40",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-06-16",
   "access_type": null,
   "license": null,
   "significance": "753B MoE, MIT, top open-weight on Artificial Analysis index",
   "organization_evidence": [
    "These releases, from late April to mid-June 2026, include DeepSeek V4, MiniMax M3, Kimi K2.7-Code, and GLM-5.2, all downloadable and most under permissive licenses.",
    "Signal\u2019s AI Milestone: Four Frontier Models Launched In Just Two Months",
    "### The production line \u2014 spring 2026",
    "**GLM-5.2 (Z.ai)**753B MoE, MIT, top open-weight on Artificial Analysis index"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The models include DeepSeek V4 on April 24, MiniMax M3 on June 1, and Kimi K2.7-Code and GLM-5.2 within days of each other in mid-June.",
    "Signal\u2019s AI Milestone: Four Frontier Models Launched In Just Two Months",
    "### The production line \u2014 spring 2026",
    "JUN 13\u201316"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "**GLM-5.2 (Z.ai)**753B MoE, MIT, top open-weight on Artificial Analysis index",
    "Signal\u2019s AI Milestone: Four Frontier Models Launched In Just Two Months",
    "### The production line \u2014 spring 2026"
   ]
  },
  {
   "url": "https://chatforest.com/builders-log/inkling-thinking-machines-975b-moe-apache-open-weight-tinker-fine-tuning-builder-guide",
   "title": "Inkling: Thinking Machines' 975B Open-Weight MoE \u2014 Self-Host, API Pricing, and Fine-Tuning with Tinker",
   "published_at": "2026-07-16T15:00:00",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "A 975B parameter MoE model with 41B active parameters, 1M context window, and native multimodality.",
   "organization_evidence": [
    "Builder's Log Published Jul 17, 2026",
    "# Inkling: Thinking Machines' 975B Open-Weight MoE \u2014 Self-Host, API Pricing, and Fine-Tuning with Tinker",
    "On July 15, 2026, [Thinking Machines Lab](https://thinkingmachines.ai/news/introducing-inkling/) \u2014 the AI startup [founded by ex-OpenAI CTO Mira Murati](https://techcrunch.com/2025/02/18/thinking-machines-lab-is-ex-openai-cto-mira-muratis-new-startup) alongside chief scientist and OpenAI co-founder John Schulman \u2014 released [Inkling](https://thinkingmachines.ai/inkling/), its first open-weight model."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Builder's Log Published Jul 17, 2026",
    "# Inkling: Thinking Machines' 975B Open-Weight MoE \u2014 Self-Host, API Pricing, and Fine-Tuning with Tinker",
    "On July 15, 2026, [Thinking Machines Lab](https://thinkingmachines.ai/news/introducing-inkling/) \u2014 the AI startup [founded by ex-OpenAI CTO Mira Murati](https://techcrunch.com/2025/02/18/thinking-machines-lab-is-ex-openai-cto-mira-muratis-new-startup) alongside chief scientist and OpenAI co-founder John Schulman \u2014 released [Inkling](https://thinkingmachines.ai/inkling/), its first open-weight model."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Builder's Log Published Jul 17, 2026",
    "# Inkling: Thinking Machines' 975B Open-Weight MoE \u2014 Self-Host, API Pricing, and Fine-Tuning with Tinker",
    "Total parameters: 975 billion.",
    "Active parameters per forward pass: approximately 41 billion \u2014 a 4.2% activation ratio, which keeps inference costs roughly in line with a 41B dense model despite the massive knowledge base encoded in the full weight set.",
    "Context window: 1 million tokens."
   ]
  },
  {
   "url": "https://bearplex.com/best/open-source-llm",
   "title": "Best Open-Source LLM in 2026",
   "published_at": "2026-07-06T00:00:00",
   "organization": "Zhipu AI",
   "model": null,
   "release_date": "2026-06-17",
   "access_type": null,
   "license": null,
   "significance": "753B-parameter Mixture-of-Experts model with a 1M-token context window",
   "organization_evidence": [
    "GLM-5.2",
    "Z.ai (Zhipu AI)",
    "Announced June 17, 2026 with weights on Hugging Face under plain MIT, it is a 753B-parameter Mixture-of-Experts model (roughly 40B active per token) with a 1M-token context window, up from GLM-5.1's 200K."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "GLM-5.2",
    "Z.ai (Zhipu AI)",
    "Announced June 17, 2026 with weights on Hugging Face under plain MIT, it is a 753B-parameter Mixture-of-Experts model (roughly 40B active per token) with a 1M-token context window, up from GLM-5.1's 200K."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "GLM-5.2",
    "Z.ai (Zhipu AI)",
    "Announced June 17, 2026 with weights on Hugging Face under plain MIT, it is a 753B-parameter Mixture-of-Experts model (roughly 40B active per token) with a 1M-token context window, up from GLM-5.1's 200K."
   ]
  },
  {
   "url": "https://bearplex.com/best/open-source-llm",
   "title": "Best Open-Source LLM in 2026",
   "published_at": "2026-07-06T00:00:00",
   "organization": "Google",
   "model": null,
   "release_date": "2026-03-31",
   "access_type": null,
   "license": null,
   "significance": "First Gemma generation under plain Apache 2.0 spanning edge models, a 26B MoE, and a 31B dense flagship",
   "organization_evidence": [
    "Gemma 4",
    "Google",
    "Launched March 31, 2026 as the first Gemma generation under plain Apache 2.0, the family spans E2B and E4B edge models (effective 2B and 4B footprints, 128K context, native audio input, offline on phones, Raspberry Pi, and Jetson-class hardware per Google), a 26B MoE activating only 3.8B parameters per token, and a 31B dense flagship with 256K context whose unquantized bfloat16 weights Google states fit a single 80GB H100."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Gemma 4",
    "Google",
    "Launched March 31, 2026 as the first Gemma generation under plain Apache 2.0, the family spans E2B and E4B edge models (effective 2B and 4B footprints, 128K context, native audio input, offline on phones, Raspberry Pi, and Jetson-class hardware per Google), a 26B MoE activating only 3.8B parameters per token, and a 31B dense flagship with 256K context whose unquantized bfloat16 weights Google states fit a single 80GB H100."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Gemma 4",
    "Google",
    "Launched March 31, 2026 as the first Gemma generation under plain Apache 2.0, the family spans E2B and E4B edge models (effective 2B and 4B footprints, 128K context, native audio input, offline on phones, Raspberry Pi, and Jetson-class hardware per Google), a 26B MoE activating only 3.8B parameters per token, and a 31B dense flagship with 256K context whose unquantized bfloat16 weights Google states fit a single 80GB H100."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/qwen3-6-35b-a3b?intelligence=coding-index",
   "title": "Qwen3.6 35B A3B - Intelligence, Performance & Price Analysis",
   "published_at": null,
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "Intelligence index of 43 with 199.1 output tokens per second",
   "organization_evidence": [
    "![Qwen3.6 35B A3B (Reasoning) logo](/img/logos/alibaba_small.svg)",
    "# Qwen3.6 35B A3B (Reasoning) Intelligence, Performance & Price Analysis"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Released April 2026",
    "# Qwen3.6 35B A3B (Reasoning) Intelligence, Performance & Price Analysis"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "43",
    "Artificial Analysis Intelligence Index",
    "199.1",
    "Output tokens per second",
    "# Qwen3.6 35B A3B (Reasoning) Intelligence, Performance & Price Analysis"
   ]
  },
  {
   "url": "https://digitalapplied.com/blog/meituan-longcat-2-0-chinese-giants-open-weight-wave",
   "title": "LongCat 2.0 and the Chinese Giants Shipping Frontier AI",
   "published_at": "2026-07-21T23:00:00",
   "organization": "Meituan",
   "model": "LongCat 2.0",
   "release_date": "2026-06-30",
   "access_type": "open-weight",
   "license": "MIT",
   "significance": "1.6-trillion-parameter sparse Mixture-of-Experts model with 1M context, edging past GPT-5.5 on SWE-bench Pro.",
   "organization_evidence": [
    "Meituan LongCat 2.0 is the strangest sentence in AI this summer: a food-delivery super-app now ships a 1.6-trillion-parameter open-weight model that, by its maker's own testing, edges past GPT-5.5 on a serious software-engineering benchmark."
   ],
   "model_evidence": [
    "Meituan LongCat 2.0 is the strangest sentence in AI this summer: a food-delivery super-app now ships a 1.6-trillion-parameter open-weight model that, by its maker's own testing, edges past GPT-5.5 on a serious software-engineering benchmark."
   ],
   "release_date_evidence": [
    "Unveiled June 30, 2026 on GitHub, Hugging Face, and Meituan's own platform, it turned out to be the engine behind \"Owl Alpha\" \u2014 the anonymous stealth model that had been topping OpenRouter's developer charts.",
    "Meituan LongCat 2.0 is the strangest sentence in AI this summer: a food-delivery super-app now ships a 1.6-trillion-parameter open-weight model that, by its maker's own testing, edges past GPT-5.5 on a serious software-engineering benchmark."
   ],
   "access_type_evidence": [
    "Meituan LongCat 2.0 is the strangest sentence in AI this summer: a food-delivery super-app now ships a 1.6-trillion-parameter open-weight model that, by its maker's own testing, edges past GPT-5.5 on a serious software-engineering benchmark."
   ],
   "license_evidence": [
    "Safetensors weights published under MIT \u2014 among the most permissive licenses in the frontier-adjacent class.",
    "Meituan LongCat 2.0 is the strangest sentence in AI this summer: a food-delivery super-app now ships a 1.6-trillion-parameter open-weight model that, by its maker's own testing, edges past GPT-5.5 on a serious software-engineering benchmark."
   ],
   "significance_evidence": [
    "Meituan LongCat 2.0 is the strangest sentence in AI this summer: a food-delivery super-app now ships a 1.6-trillion-parameter open-weight model that, by its maker's own testing, edges past GPT-5.5 on a serious software-engineering benchmark."
   ]
  },
  {
   "url": "https://digitalapplied.com/blog/meituan-longcat-2-0-chinese-giants-open-weight-wave",
   "title": "LongCat 2.0 and the Chinese Giants Shipping Frontier AI",
   "published_at": "2026-07-21T23:00:00",
   "organization": "Tencent",
   "model": "Hunyuan Hy3",
   "release_date": "2026-07-06",
   "access_type": "open-weight",
   "license": "Apache 2.0",
   "significance": "295B total parameters, 21B active, 256K context reasoning model.",
   "organization_evidence": [
    "The third giant is Tencent, whose [Hunyuan Hy3 open-weight reasoning model](/blog/tencent-hunyuan-hy3-open-weight-reasoning-model-2026) fully released on July 6, 2026 \u2014 295B total parameters, 21B active, 256K context, Apache 2.0, listed on OpenRouter at roughly $0.20 / $0.80 per million tokens."
   ],
   "model_evidence": [
    "The third giant is Tencent, whose [Hunyuan Hy3 open-weight reasoning model](/blog/tencent-hunyuan-hy3-open-weight-reasoning-model-2026) fully released on July 6, 2026 \u2014 295B total parameters, 21B active, 256K context, Apache 2.0, listed on OpenRouter at roughly $0.20 / $0.80 per million tokens."
   ],
   "release_date_evidence": [
    "The third giant is Tencent, whose [Hunyuan Hy3 open-weight reasoning model](/blog/tencent-hunyuan-hy3-open-weight-reasoning-model-2026) fully released on July 6, 2026 \u2014 295B total parameters, 21B active, 256K context, Apache 2.0, listed on OpenRouter at roughly $0.20 / $0.80 per million tokens."
   ],
   "access_type_evidence": [
    "The third giant is Tencent, whose [Hunyuan Hy3 open-weight reasoning model](/blog/tencent-hunyuan-hy3-open-weight-reasoning-model-2026) fully released on July 6, 2026 \u2014 295B total parameters, 21B active, 256K context, Apache 2.0, listed on OpenRouter at roughly $0.20 / $0.80 per million tokens."
   ],
   "license_evidence": [
    "The third giant is Tencent, whose [Hunyuan Hy3 open-weight reasoning model](/blog/tencent-hunyuan-hy3-open-weight-reasoning-model-2026) fully released on July 6, 2026 \u2014 295B total parameters, 21B active, 256K context, Apache 2.0, listed on OpenRouter at roughly $0.20 / $0.80 per million tokens."
   ],
   "significance_evidence": [
    "The third giant is Tencent, whose [Hunyuan Hy3 open-weight reasoning model](/blog/tencent-hunyuan-hy3-open-weight-reasoning-model-2026) fully released on July 6, 2026 \u2014 295B total parameters, 21B active, 256K context, Apache 2.0, listed on OpenRouter at roughly $0.20 / $0.80 per million tokens."
   ]
  },
  {
   "url": "https://singularitybyte.com/models/thinking-machines-inkling.html",
   "title": "Inkling",
   "published_at": "2026-07-22T15:20:00",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "a 975B-parameter Mixture-of-Experts model with 41B active per token, multimodal input, 1M-token context, trained on 45T tokens",
   "organization_evidence": [
    "# Inkling",
    "Mira Murati's Thinking Machines Lab shipped Inkling: 975B parameters, Apache 2.0, weights on Hugging Face on day one."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Inkling",
    "On July 15, 2026, Thinking Machines Lab shipped Inkling: 975 billion parameters, Apache 2.0, weights on Hugging Face on day one."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Inkling",
    "- **What it is:** a 975B-parameter Mixture-of-Experts model with 41B active per token, multimodal input (text, image, audio), 1M-token context, trained on 45T tokens, released under a clean Apache 2.0 license with weights on Hugging Face at launch."
   ]
  },
  {
   "url": "https://benchmarklist.com/timeline",
   "title": "AI Model Release Timeline | BenchmarkList",
   "published_at": null,
   "organization": "Meta",
   "model": null,
   "release_date": "2026-08-05",
   "access_type": null,
   "license": null,
   "significance": "Coding-focused reasoning model",
   "organization_evidence": [
    "Model release timeline",
    "## Release Index",
    "| Date | Provider | Model | Type | Focus | Benchmark coverage |",
    "| Aug 5, 2026 | Meta | [Muse Spark 1.2](/models/meta-muse-spark-1.2/) | Coding-focused reasoning model | Coding, Agentic, Tool use, Long-horizon work | 3 benchmarks / 3 scores |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Model release timeline",
    "## Release Index",
    "| Date | Provider | Model | Type | Focus | Benchmark coverage |",
    "| Aug 5, 2026 | Meta | [Muse Spark 1.2](/models/meta-muse-spark-1.2/) | Coding-focused reasoning model | Coding, Agentic, Tool use, Long-horizon work | 3 benchmarks / 3 scores |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Model release timeline",
    "## Release Index",
    "| Date | Provider | Model | Type | Focus | Benchmark coverage |",
    "| Aug 5, 2026 | Meta | [Muse Spark 1.2](/models/meta-muse-spark-1.2/) | Coding-focused reasoning model | Coding, Agentic, Tool use, Long-horizon work | 3 benchmarks / 3 scores |"
   ]
  },
  {
   "url": "https://benchmarklist.com/timeline",
   "title": "AI Model Release Timeline | BenchmarkList",
   "published_at": null,
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-06-16",
   "access_type": null,
   "license": null,
   "significance": "Open-weight flagship coding and agentic model",
   "organization_evidence": [
    "Model release timeline",
    "## Release Index",
    "| Date | Provider | Model | Type | Focus | Benchmark coverage |",
    "| Jun 16, 2026 | Z.ai | [GLM-5.2](/models/z-ai-glm-5.2/) | Open-weight flagship coding and agentic model | Coding, Agents, Long context, Open weight | 99 benchmarks / 118 scores |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Model release timeline",
    "## Release Index",
    "| Date | Provider | Model | Type | Focus | Benchmark coverage |",
    "| Jun 16, 2026 | Z.ai | [GLM-5.2](/models/z-ai-glm-5.2/) | Open-weight flagship coding and agentic model | Coding, Agents, Long context, Open weight | 99 benchmarks / 118 scores |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Model release timeline",
    "## Release Index",
    "| Date | Provider | Model | Type | Focus | Benchmark coverage |",
    "| Jun 16, 2026 | Z.ai | [GLM-5.2](/models/z-ai-glm-5.2/) | Open-weight flagship coding and agentic model | Coding, Agents, Long context, Open weight | 99 benchmarks / 118 scores |"
   ]
  },
  {
   "url": "https://benchmarklist.com/timeline",
   "title": "AI Model Release Timeline | BenchmarkList",
   "published_at": null,
   "organization": "Google",
   "model": null,
   "release_date": "2026-06-04",
   "access_type": null,
   "license": null,
   "significance": "Open-weight live music model",
   "organization_evidence": [
    "Model release timeline",
    "## Release Index",
    "| Date | Provider | Model | Type | Focus | Benchmark coverage |",
    "| Jun 4, 2026 | Google | [Magenta RealTime 2](/models/google-magenta-realtime-2/) | Open-weight live music model | Open weight, Music generation, Local inference | No matched results yet |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Model release timeline",
    "## Release Index",
    "| Date | Provider | Model | Type | Focus | Benchmark coverage |",
    "| Jun 4, 2026 | Google | [Magenta RealTime 2](/models/google-magenta-realtime-2/) | Open-weight live music model | Open weight, Music generation, Local inference | No matched results yet |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Model release timeline",
    "## Release Index",
    "| Date | Provider | Model | Type | Focus | Benchmark coverage |",
    "| Jun 4, 2026 | Google | [Magenta RealTime 2](/models/google-magenta-realtime-2/) | Open-weight live music model | Open weight, Music generation, Local inference | No matched results yet |"
   ]
  },
  {
   "url": "https://benchmarklist.com/timeline",
   "title": "AI Model Release Timeline | BenchmarkList",
   "published_at": null,
   "organization": "Google",
   "model": null,
   "release_date": "2026-06-03",
   "access_type": null,
   "license": null,
   "significance": "Open-weight multimodal model",
   "organization_evidence": [
    "Model release timeline",
    "## Release Index",
    "| Date | Provider | Model | Type | Focus | Benchmark coverage |",
    "| Jun 3, 2026 | Google | [Gemma 4 12B](/models/google-gemma-4-12b-it/) | Open-weight multimodal model | Open weight, Multimodal, Local inference | 28 benchmarks / 28 scores |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Model release timeline",
    "## Release Index",
    "| Date | Provider | Model | Type | Focus | Benchmark coverage |",
    "| Jun 3, 2026 | Google | [Gemma 4 12B](/models/google-gemma-4-12b-it/) | Open-weight multimodal model | Open weight, Multimodal, Local inference | 28 benchmarks / 28 scores |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Model release timeline",
    "## Release Index",
    "| Date | Provider | Model | Type | Focus | Benchmark coverage |",
    "| Jun 3, 2026 | Google | [Gemma 4 12B](/models/google-gemma-4-12b-it/) | Open-weight multimodal model | Open weight, Multimodal, Local inference | 28 benchmarks / 28 scores |"
   ]
  },
  {
   "url": "https://benchmarklist.com/timeline",
   "title": "AI Model Release Timeline | BenchmarkList",
   "published_at": null,
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "Frontier open-weight model",
   "organization_evidence": [
    "Model release timeline",
    "## Release Index",
    "| Date | Provider | Model | Type | Focus | Benchmark coverage |",
    "| Jun 1, 2026 | MiniMax | [MiniMax M3](/models/minimax-minimax-m3/) | Frontier open-weight model | Coding, Agents, Long context, Multimodal | 80 benchmarks / 87 scores |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Model release timeline",
    "## Release Index",
    "| Date | Provider | Model | Type | Focus | Benchmark coverage |",
    "| Jun 1, 2026 | MiniMax | [MiniMax M3](/models/minimax-minimax-m3/) | Frontier open-weight model | Coding, Agents, Long context, Multimodal | 80 benchmarks / 87 scores |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Model release timeline",
    "## Release Index",
    "| Date | Provider | Model | Type | Focus | Benchmark coverage |",
    "| Jun 1, 2026 | MiniMax | [MiniMax M3](/models/minimax-minimax-m3/) | Frontier open-weight model | Coding, Agents, Long context, Multimodal | 80 benchmarks / 87 scores |"
   ]
  },
  {
   "url": "https://benchmarklist.com/timeline",
   "title": "AI Model Release Timeline | BenchmarkList",
   "published_at": null,
   "organization": "Cohere",
   "model": null,
   "release_date": "2026-05-20",
   "access_type": null,
   "license": null,
   "significance": "Flagship open-weight MoE model",
   "organization_evidence": [
    "Model release timeline",
    "## Release Index",
    "| Date | Provider | Model | Type | Focus | Benchmark coverage |",
    "| May 20, 2026 | Cohere | [Command A+](/models/cohere-command-a-plus-05-2026/) | Flagship open-weight MoE model | Agentic, Multimodal, Open weight | 4 benchmarks / 4 scores |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Model release timeline",
    "## Release Index",
    "| Date | Provider | Model | Type | Focus | Benchmark coverage |",
    "| May 20, 2026 | Cohere | [Command A+](/models/cohere-command-a-plus-05-2026/) | Flagship open-weight MoE model | Agentic, Multimodal, Open weight | 4 benchmarks / 4 scores |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Model release timeline",
    "## Release Index",
    "| Date | Provider | Model | Type | Focus | Benchmark coverage |",
    "| May 20, 2026 | Cohere | [Command A+](/models/cohere-command-a-plus-05-2026/) | Flagship open-weight MoE model | Agentic, Multimodal, Open weight | 4 benchmarks / 4 scores |"
   ]
  },
  {
   "url": "https://benchmarklist.com/timeline",
   "title": "AI Model Release Timeline | BenchmarkList",
   "published_at": null,
   "organization": "Mistral AI",
   "model": null,
   "release_date": "2026-04-28",
   "access_type": null,
   "license": null,
   "significance": "Flagship agentic model",
   "organization_evidence": [
    "Model release timeline",
    "## Release Index",
    "| Date | Provider | Model | Type | Focus | Benchmark coverage |",
    "| Apr 28, 2026 | Mistral AI | [Mistral Medium 3.5](/models/mistralai-mistral-medium-3-5/) | Flagship agentic model | Coding, Agentic, Multimodal | 46 benchmarks / 50 scores |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Model release timeline",
    "## Release Index",
    "| Date | Provider | Model | Type | Focus | Benchmark coverage |",
    "| Apr 28, 2026 | Mistral AI | [Mistral Medium 3.5](/models/mistralai-mistral-medium-3-5/) | Flagship agentic model | Coding, Agentic, Multimodal | 46 benchmarks / 50 scores |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Model release timeline",
    "## Release Index",
    "| Date | Provider | Model | Type | Focus | Benchmark coverage |",
    "| Apr 28, 2026 | Mistral AI | [Mistral Medium 3.5](/models/mistralai-mistral-medium-3-5/) | Flagship agentic model | Coding, Agentic, Multimodal | 46 benchmarks / 50 scores |"
   ]
  },
  {
   "url": "https://benchmarklist.com/timeline",
   "title": "AI Model Release Timeline | BenchmarkList",
   "published_at": null,
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "Open MoE model (preview)",
   "organization_evidence": [
    "Model release timeline",
    "## Release Index",
    "| Date | Provider | Model | Type | Focus | Benchmark coverage |",
    "| Apr 24, 2026 | DeepSeek | DeepSeek-V4 | Open MoE model (preview) | Agentic, Reasoning, Long context | Unlinked |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Model release timeline",
    "## Release Index",
    "| Date | Provider | Model | Type | Focus | Benchmark coverage |",
    "| Apr 24, 2026 | DeepSeek | DeepSeek-V4 | Open MoE model (preview) | Agentic, Reasoning, Long context | Unlinked |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Model release timeline",
    "## Release Index",
    "| Date | Provider | Model | Type | Focus | Benchmark coverage |",
    "| Apr 24, 2026 | DeepSeek | DeepSeek-V4 | Open MoE model (preview) | Agentic, Reasoning, Long context | Unlinked |"
   ]
  },
  {
   "url": "https://benchmarklist.com/timeline",
   "title": "AI Model Release Timeline | BenchmarkList",
   "published_at": null,
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-04-20",
   "access_type": null,
   "license": null,
   "significance": "Open agentic coding model",
   "organization_evidence": [
    "Model release timeline",
    "## Release Index",
    "| Date | Provider | Model | Type | Focus | Benchmark coverage |",
    "| Apr 20, 2026 | Moonshot AI | [Kimi K2.6](/models/moonshotai-kimi-k2.6/) | Open agentic coding model | Coding, Agentic, Multimodal | 230 benchmarks / 258 scores |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Model release timeline",
    "## Release Index",
    "| Date | Provider | Model | Type | Focus | Benchmark coverage |",
    "| Apr 20, 2026 | Moonshot AI | [Kimi K2.6](/models/moonshotai-kimi-k2.6/) | Open agentic coding model | Coding, Agentic, Multimodal | 230 benchmarks / 258 scores |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Model release timeline",
    "## Release Index",
    "| Date | Provider | Model | Type | Focus | Benchmark coverage |",
    "| Apr 20, 2026 | Moonshot AI | [Kimi K2.6](/models/moonshotai-kimi-k2.6/) | Open agentic coding model | Coding, Agentic, Multimodal | 230 benchmarks / 258 scores |"
   ]
  },
  {
   "url": "https://benchmarklist.com/timeline",
   "title": "AI Model Release Timeline | BenchmarkList",
   "published_at": null,
   "organization": "Google",
   "model": null,
   "release_date": "2026-04-02",
   "access_type": null,
   "license": null,
   "significance": "Open-weight multimodal model family",
   "organization_evidence": [
    "Model release timeline",
    "## Release Index",
    "| Date | Provider | Model | Type | Focus | Benchmark coverage |",
    "| Apr 2, 2026 | Google | Gemma 4 | Open-weight multimodal model family | Open weight, Multimodal, Agentic | Unlinked |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Model release timeline",
    "## Release Index",
    "| Date | Provider | Model | Type | Focus | Benchmark coverage |",
    "| Apr 2, 2026 | Google | Gemma 4 | Open-weight multimodal model family | Open weight, Multimodal, Agentic | Unlinked |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Model release timeline",
    "## Release Index",
    "| Date | Provider | Model | Type | Focus | Benchmark coverage |",
    "| Apr 2, 2026 | Google | Gemma 4 | Open-weight multimodal model family | Open weight, Multimodal, Agentic | Unlinked |"
   ]
  },
  {
   "url": "https://benchmarklist.com/timeline",
   "title": "AI Model Release Timeline | BenchmarkList",
   "published_at": null,
   "organization": "Mistral AI",
   "model": null,
   "release_date": "2026-03-23",
   "access_type": null,
   "license": null,
   "significance": "Text-to-speech model",
   "organization_evidence": [
    "Model release timeline",
    "## Release Index",
    "| Date | Provider | Model | Type | Focus | Benchmark coverage |",
    "| Mar 23, 2026 | Mistral AI | Voxtral TTS | Text-to-speech model | Multimodal, Open weight | Unlinked |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Model release timeline",
    "## Release Index",
    "| Date | Provider | Model | Type | Focus | Benchmark coverage |",
    "| Mar 23, 2026 | Mistral AI | Voxtral TTS | Text-to-speech model | Multimodal, Open weight | Unlinked |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Model release timeline",
    "## Release Index",
    "| Date | Provider | Model | Type | Focus | Benchmark coverage |",
    "| Mar 23, 2026 | Mistral AI | Voxtral TTS | Text-to-speech model | Multimodal, Open weight | Unlinked |"
   ]
  },
  {
   "url": "https://benchmarklist.com/timeline",
   "title": "AI Model Release Timeline | BenchmarkList",
   "published_at": null,
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-03-18",
   "access_type": null,
   "license": null,
   "significance": "Open-weight self-evolving agentic model",
   "organization_evidence": [
    "Model release timeline",
    "## Release Index",
    "| Date | Provider | Model | Type | Focus | Benchmark coverage |",
    "| Mar 18, 2026 | MiniMax | [MiniMax-M2.7](/models/minimax-minimax-m2.7/) | Open-weight self-evolving agentic model | Agentic, Coding, Open weight | 146 benchmarks / 164 scores |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Model release timeline",
    "## Release Index",
    "| Date | Provider | Model | Type | Focus | Benchmark coverage |",
    "| Mar 18, 2026 | MiniMax | [MiniMax-M2.7](/models/minimax-minimax-m2.7/) | Open-weight self-evolving agentic model | Agentic, Coding, Open weight | 146 benchmarks / 164 scores |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Model release timeline",
    "## Release Index",
    "| Date | Provider | Model | Type | Focus | Benchmark coverage |",
    "| Mar 18, 2026 | MiniMax | [MiniMax-M2.7](/models/minimax-minimax-m2.7/) | Open-weight self-evolving agentic model | Agentic, Coding, Open weight | 146 benchmarks / 164 scores |"
   ]
  },
  {
   "url": "https://benchmarklist.com/timeline",
   "title": "AI Model Release Timeline | BenchmarkList",
   "published_at": null,
   "organization": "Mistral AI",
   "model": null,
   "release_date": "2026-03-16",
   "access_type": null,
   "license": null,
   "significance": "Open MoE unified model",
   "organization_evidence": [
    "Model release timeline",
    "## Release Index",
    "| Date | Provider | Model | Type | Focus | Benchmark coverage |",
    "| Mar 16, 2026 | Mistral AI | [Mistral Small 4](/models/mistralai-mistral-small-2603/) | Open MoE unified model | Open weight, Multimodal, Reasoning | 28 benchmarks / 37 scores |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Model release timeline",
    "## Release Index",
    "| Date | Provider | Model | Type | Focus | Benchmark coverage |",
    "| Mar 16, 2026 | Mistral AI | [Mistral Small 4](/models/mistralai-mistral-small-2603/) | Open MoE unified model | Open weight, Multimodal, Reasoning | 28 benchmarks / 37 scores |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Model release timeline",
    "## Release Index",
    "| Date | Provider | Model | Type | Focus | Benchmark coverage |",
    "| Mar 16, 2026 | Mistral AI | [Mistral Small 4](/models/mistralai-mistral-small-2603/) | Open MoE unified model | Open weight, Multimodal, Reasoning | 28 benchmarks / 37 scores |"
   ]
  },
  {
   "url": "https://benchmarklist.com/timeline",
   "title": "AI Model Release Timeline | BenchmarkList",
   "published_at": null,
   "organization": "Cohere",
   "model": null,
   "release_date": "2026-02-17",
   "access_type": null,
   "license": null,
   "significance": "Open multilingual edge model",
   "organization_evidence": [
    "Model release timeline",
    "## Release Index",
    "| Date | Provider | Model | Type | Focus | Benchmark coverage |",
    "| Feb 17, 2026 | Cohere | Tiny Aya | Open multilingual edge model | Multilingual, Open weight | Unlinked |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Model release timeline",
    "## Release Index",
    "| Date | Provider | Model | Type | Focus | Benchmark coverage |",
    "| Feb 17, 2026 | Cohere | Tiny Aya | Open multilingual edge model | Multilingual, Open weight | Unlinked |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Model release timeline",
    "## Release Index",
    "| Date | Provider | Model | Type | Focus | Benchmark coverage |",
    "| Feb 17, 2026 | Cohere | Tiny Aya | Open multilingual edge model | Multilingual, Open weight | Unlinked |"
   ]
  },
  {
   "url": "https://benchmarklist.com/timeline",
   "title": "AI Model Release Timeline | BenchmarkList",
   "published_at": null,
   "organization": "Qwen",
   "model": null,
   "release_date": "2026-02-16",
   "access_type": null,
   "license": null,
   "significance": "Open model family",
   "organization_evidence": [
    "Model release timeline",
    "## Release Index",
    "| Date | Provider | Model | Type | Focus | Benchmark coverage |",
    "| Feb 16, 2026 | Qwen | Qwen3.5 | Open model family | Open weight, Agentic, Long context | Unlinked |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Model release timeline",
    "## Release Index",
    "| Date | Provider | Model | Type | Focus | Benchmark coverage |",
    "| Feb 16, 2026 | Qwen | Qwen3.5 | Open model family | Open weight, Agentic, Long context | Unlinked |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Model release timeline",
    "## Release Index",
    "| Date | Provider | Model | Type | Focus | Benchmark coverage |",
    "| Feb 16, 2026 | Qwen | Qwen3.5 | Open model family | Open weight, Agentic, Long context | Unlinked |"
   ]
  },
  {
   "url": "https://benchmarklist.com/timeline",
   "title": "AI Model Release Timeline | BenchmarkList",
   "published_at": null,
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-02-12",
   "access_type": null,
   "license": null,
   "significance": "Open-weight coding/agentic SOTA model",
   "organization_evidence": [
    "Model release timeline",
    "## Release Index",
    "| Date | Provider | Model | Type | Focus | Benchmark coverage |",
    "| Feb 12, 2026 | MiniMax | [MiniMax-M2.5](/models/minimax-minimax-m2.5/) | Open-weight coding/agentic SOTA model | Coding, Agentic, Open weight | 95 benchmarks / 99 scores |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Model release timeline",
    "## Release Index",
    "| Date | Provider | Model | Type | Focus | Benchmark coverage |",
    "| Feb 12, 2026 | MiniMax | [MiniMax-M2.5](/models/minimax-minimax-m2.5/) | Open-weight coding/agentic SOTA model | Coding, Agentic, Open weight | 95 benchmarks / 99 scores |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Model release timeline",
    "## Release Index",
    "| Date | Provider | Model | Type | Focus | Benchmark coverage |",
    "| Feb 12, 2026 | MiniMax | [MiniMax-M2.5](/models/minimax-minimax-m2.5/) | Open-weight coding/agentic SOTA model | Coding, Agentic, Open weight | 95 benchmarks / 99 scores |"
   ]
  },
  {
   "url": "https://benchmarklist.com/timeline",
   "title": "AI Model Release Timeline | BenchmarkList",
   "published_at": null,
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-01-27",
   "access_type": null,
   "license": null,
   "significance": "Open multimodal agentic model",
   "organization_evidence": [
    "Model release timeline",
    "## Release Index",
    "| Date | Provider | Model | Type | Focus | Benchmark coverage |",
    "| Jan 27, 2026 | Moonshot AI | [Kimi K2.5](/models/moonshotai-kimi-k2.5/) | Open multimodal agentic model | Multimodal, Agentic, Open weight | 239 benchmarks / 257 scores |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Model release timeline",
    "## Release Index",
    "| Date | Provider | Model | Type | Focus | Benchmark coverage |",
    "| Jan 27, 2026 | Moonshot AI | [Kimi K2.5](/models/moonshotai-kimi-k2.5/) | Open multimodal agentic model | Multimodal, Agentic, Open weight | 239 benchmarks / 257 scores |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Model release timeline",
    "## Release Index",
    "| Date | Provider | Model | Type | Focus | Benchmark coverage |",
    "| Jan 27, 2026 | Moonshot AI | [Kimi K2.5](/models/moonshotai-kimi-k2.5/) | Open multimodal agentic model | Multimodal, Agentic, Open weight | 239 benchmarks / 257 scores |"
   ]
  },
  {
   "url": "https://earlyterms.com/term/inkling",
   "title": "Inkling is Thinking Machines Lab's first open-weights large language model: a 975-billion-parameter Mixture-of",
   "published_at": "2026-07-15T00:00:00",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "a 975-billion-parameter Mixture-of-Experts system with 41B active parameters that natively reasons over text, images, and audio",
   "organization_evidence": [
    "# Inkling",
    "Inkling is Thinking Machines Lab's first open-weights large language model: a 975-billion-parameter Mixture-of-Experts system with 41B active parameters that natively reasons over text, images, and audio."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Inkling",
    "Released [July 15, 2026](https://thinkingmachines.ai/news/introducing-inkling/) by Mira Murati's Thinking Machines Lab under Apache 2.0, Inkling trained on 45 trillion tokens and ships pre-wired for fine-tuning on the lab's Tinker platform."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Inkling",
    "Inkling is Thinking Machines Lab's first open-weights large language model: a 975-billion-parameter Mixture-of-Experts system with 41B active parameters that natively reasons over text, images, and audio."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/qwen3-5-9b-non-reasoning?model-size=intelligence-vs-total-parameters",
   "title": "Qwen3.5 9B - Intelligence, Performance & Price Analysis",
   "published_at": null,
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-03",
   "access_type": null,
   "license": null,
   "significance": null,
   "organization_evidence": [
    "![Qwen3.5 9B (Non-reasoning) logo](/img/logos/alibaba_small.svg)",
    "# Qwen3.5 9B (Non-reasoning) Intelligence, Performance & Price Analysis"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "\u2022\nReleased March 2026",
    "# Qwen3.5 9B (Non-reasoning) Intelligence, Performance & Price Analysis"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": null
  },
  {
   "url": "https://public.wmo.int/media/news-from-members/fenghe-goes-open-source-global-initiative-kicks",
   "title": "Fenghe goes open-source as Global Initiative kicks off",
   "published_at": "2026-07-17T14:05:17",
   "organization": "China Meteorological Administration",
   "model": null,
   "release_date": "2026-07-17",
   "access_type": null,
   "license": null,
   "significance": "the world's first open-source meteorological large language model with a hundred-billion-parameter scale",
   "organization_evidence": [
    "# Fenghe goes open-source as Global Initiative kicks off",
    "On July 17, at the 2026 World Artificial Intelligence Conference (WAIC) Meteorological Forum, the China Meteorological Administration (CMA) unveiled Fenghe, a Large Language Model-Based Al MeteorologicalService System, marking the commencement of its Global Open-Source Initiative."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Fenghe goes open-source as Global Initiative kicks off",
    "17 July 2026",
    "On July 17, at the 2026 World Artificial Intelligence Conference (WAIC) Meteorological Forum, the China Meteorological Administration (CMA) unveiled Fenghe, a Large Language Model-Based Al MeteorologicalService System, marking the commencement of its Global Open-Source Initiative."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Fenghe goes open-source as Global Initiative kicks off",
    "Co-developed by the CMA Public Meteorological Service Centre, the Xiong'an Artificial Intelligence Research Institute, Z.AI Co., Ltd. and other entities, Fenghe is the world's first open-source meteorological large language model with a hundred-billion-parameter scale."
   ]
  },
  {
   "url": "https://noqta.tn/en/news/ant-group-ling-2-6-1t-trillion-parameter-open-source-2026",
   "title": "Ant Group Open-Sources Ling-2.6-1T, a Trillion-Parameter Model Built for Agents",
   "published_at": "2026-05-01T00:00:00",
   "organization": "Ant Group",
   "model": null,
   "release_date": "2026-04-30",
   "access_type": null,
   "license": null,
   "significance": "trillion-parameter Mixture-of-Experts language model purpose-built for agentic workflows",
   "organization_evidence": [
    "Ant Group's InclusionAI lab on April 30, 2026 released the open weights for **Ling-2.6-1T**, a trillion-parameter Mixture-of-Experts (MoE) language model purpose-built for agentic workflows, tool use, and long-running software tasks."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Ant Group's InclusionAI lab on April 30, 2026 released the open weights for **Ling-2.6-1T**, a trillion-parameter Mixture-of-Experts (MoE) language model purpose-built for agentic workflows, tool use, and long-running software tasks."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Ant Group's InclusionAI lab on April 30, 2026 released the open weights for **Ling-2.6-1T**, a trillion-parameter Mixture-of-Experts (MoE) language model purpose-built for agentic workflows, tool use, and long-running software tasks."
   ]
  },
  {
   "url": "https://gen-ai.news/stories/thinking-machines-lab-releases-inkling-a-975b-parameter-open-weights-multimodal--3217ba",
   "title": "Thinking Machines Lab Releases Inkling: A 975B-Parameter Open-Weights Multimodal MoE With 41B Active Parameters And Controllable Thinking Effort",
   "published_at": "2026-07-15T23:48:58",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "A 975B-parameter open-weights multimodal MoE with 41B active parameters and controllable thinking effort.",
   "organization_evidence": [
    "Multimodal",
    "# Thinking Machines Lab Releases Inkling: A 975B-Parameter Open-Weights Multimodal MoE With 41B Active Parameters And Controllable Thinking Effort",
    "Thinking Machines Lab released Inkling on July 15, 2026 - its first model trained entirely from scratch."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Multimodal",
    "# Thinking Machines Lab Releases Inkling: A 975B-Parameter Open-Weights Multimodal MoE With 41B Active Parameters And Controllable Thinking Effort",
    "Thinking Machines Lab released Inkling on July 15, 2026 - its first model trained entirely from scratch."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Multimodal",
    "# Thinking Machines Lab Releases Inkling: A 975B-Parameter Open-Weights Multimodal MoE With 41B Active Parameters And Controllable Thinking Effort"
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/claude-opus-4-8-vs-glm-5-1",
   "title": "Claude Opus 4.8 (Adaptive Reasoning, Max Effort) vs GLM-5.1 (Reasoning): Model Comparison | Artificial Analysis",
   "published_at": null,
   "organization": "Z AI",
   "model": "GLM-5.1 (Reasoning)",
   "release_date": "2026-04",
   "access_type": "open source",
   "license": null,
   "significance": "Reasoning model featuring open weights",
   "organization_evidence": [
    "# Claude Opus 4.8 (Adaptive Reasoning, Max Effort) vs. GLM-5.1 (Reasoning)",
    "Comparison between Claude Opus 4.8 (Adaptive Reasoning, Max Effort) and GLM-5.1 (Reasoning) across intelligence, price, speed, context window and more.",
    "## Model Comparison",
    "| | ![Anthropic logo](/img/logos/anthropic_small.svg)Anthropic Claude Opus 4.8 (Adaptive Reasoning, Max Effort) | ![Z AI logo](/img/logos/zai_small.svg)Z AI GLM-5.1 (Reasoning) | |"
   ],
   "model_evidence": [
    "# Claude Opus 4.8 (Adaptive Reasoning, Max Effort) vs. GLM-5.1 (Reasoning)",
    "Comparison between Claude Opus 4.8 (Adaptive Reasoning, Max Effort) and GLM-5.1 (Reasoning) across intelligence, price, speed, context window and more.",
    "## Model Comparison",
    "| | ![Anthropic logo](/img/logos/anthropic_small.svg)Anthropic Claude Opus 4.8 (Adaptive Reasoning, Max Effort) | ![Z AI logo](/img/logos/zai_small.svg)Z AI GLM-5.1 (Reasoning) | |",
    "| Intelligence Index | 57 | 41 | Claude Opus 4.8 (Adaptive Reasoning, Max Effort) is more intelligent than GLM-5.1 (Reasoning) |"
   ],
   "release_date_evidence": [
    "# Claude Opus 4.8 (Adaptive Reasoning, Max Effort) vs. GLM-5.1 (Reasoning)",
    "Comparison between Claude Opus 4.8 (Adaptive Reasoning, Max Effort) and GLM-5.1 (Reasoning) across intelligence, price, speed, context window and more.",
    "## Model Comparison",
    "| | ![Anthropic logo](/img/logos/anthropic_small.svg)Anthropic Claude Opus 4.8 (Adaptive Reasoning, Max Effort) | ![Z AI logo](/img/logos/zai_small.svg)Z AI GLM-5.1 (Reasoning) | |",
    "| Release Date | May 2026 | April 2026 | Claude Opus 4.8 (Adaptive Reasoning, Max Effort) has a more recent release date than GLM-5.1 (Reasoning) |"
   ],
   "access_type_evidence": [
    "# Claude Opus 4.8 (Adaptive Reasoning, Max Effort) vs. GLM-5.1 (Reasoning)",
    "Comparison between Claude Opus 4.8 (Adaptive Reasoning, Max Effort) and GLM-5.1 (Reasoning) across intelligence, price, speed, context window and more.",
    "## Model Comparison",
    "| | ![Anthropic logo](/img/logos/anthropic_small.svg)Anthropic Claude Opus 4.8 (Adaptive Reasoning, Max Effort) | ![Z AI logo](/img/logos/zai_small.svg)Z AI GLM-5.1 (Reasoning) | |",
    "| Open Source (Weights) | No | Yes | GLM-5.1 (Reasoning) is open source while Claude Opus 4.8 (Adaptive Reasoning, Max Effort) is proprietary |"
   ],
   "license_evidence": null,
   "significance_evidence": [
    "# Claude Opus 4.8 (Adaptive Reasoning, Max Effort) vs. GLM-5.1 (Reasoning)",
    "Comparison between Claude Opus 4.8 (Adaptive Reasoning, Max Effort) and GLM-5.1 (Reasoning) across intelligence, price, speed, context window and more.",
    "## Model Comparison",
    "| | ![Anthropic logo](/img/logos/anthropic_small.svg)Anthropic Claude Opus 4.8 (Adaptive Reasoning, Max Effort) | ![Z AI logo](/img/logos/zai_small.svg)Z AI GLM-5.1 (Reasoning) | |",
    "| Intelligence Index | 57 | 41 | Claude Opus 4.8 (Adaptive Reasoning, Max Effort) is more intelligent than GLM-5.1 (Reasoning) |",
    "| Reasoning | Yes | Yes | Both Claude Opus 4.8 (Adaptive Reasoning, Max Effort) and GLM-5.1 (Reasoning) have reasoning |"
   ]
  },
  {
   "url": "https://testingcatalog.com/thinking-machines-debuts-open-weight-inkling-ai-model",
   "title": "Thinking Machines debuts open-weight Inkling AI model",
   "published_at": "2026-07-15T22:09:17",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07",
   "access_type": null,
   "license": null,
   "significance": "Mixture-of-Experts transformer with 975 billion total parameters and 41 billion active during inference, supporting up to one million token context windows",
   "organization_evidence": [
    "# Thinking Machines debuts open-weight Inkling AI model",
    "Thinking Machines Lab launches Inkling, its open-weight Mixture-of-Experts AI model, offering broad customization and multi-modal support.",
    "Thinking Machines Lab has released Inkling, its first open-weights foundation model, giving developers and companies full access to customize and deploy it."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Thinking Machines debuts open-weight Inkling AI model",
    "![Thinking Machines](https://storage.ghost.io/c/2a/1b/2a1b1782-8506-4d7d-bf53-ad3fb52e2a0f/content/images/size/w2000/2026/07/cover-social-inkling-post.png)"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Thinking Machines debuts open-weight Inkling AI model",
    "Inkling is a Mixture-of-Experts transformer with 975 billion total parameters and 41 billion active during inference."
   ]
  },
  {
   "url": "https://lmtimeline.com",
   "title": "AI Timeline \u2014 Model Releases, AI News & History",
   "published_at": "2026-08-19T00:00:00",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-08-13",
   "access_type": null,
   "license": null,
   "significance": "reports 87.9 on Terminal-Bench 2.1 and 62.7 on DeepSWE",
   "organization_evidence": [
    "## Latest in AI",
    "6. \u2014 **DeepSeek ships V4-Pro to general availability** DeepSeek moved V4-Pro to general availability as DeepSeek-V4-Pro-0813 on the app, website, and API under the existing name deepseek-v4-pro."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## Latest in AI",
    "6. \u2014 **DeepSeek ships V4-Pro to general availability** DeepSeek moved V4-Pro to general availability as DeepSeek-V4-Pro-0813 on the app, website, and API under the existing name deepseek-v4-pro."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## Latest in AI",
    "6. \u2014 **DeepSeek ships V4-Pro to general availability** DeepSeek moved V4-Pro to general availability as DeepSeek-V4-Pro-0813 on the app, website, and API under the existing name deepseek-v4-pro. The company reports 87.9 on Terminal-Bench 2.1, which tests whether a model can operate a real command line, and 62.7 on DeepSWE; MIT-licensed weights for the 0813 build are on Hugging Face."
   ]
  },
  {
   "url": "https://lmtimeline.com",
   "title": "AI Timeline \u2014 Model Releases, AI News & History",
   "published_at": "2026-08-19T00:00:00",
   "organization": "Meta",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "a 30-billion-parameter local-agent model built for always-on local agent work",
   "organization_evidence": [
    "## Latest in AI",
    "9. \u2014 **Meta open-sources Muse Glimmer** Meta released Muse Glimmer, a 30-billion-parameter local-agent model published on Hugging Face under Apache 2.0."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## Latest in AI",
    "9. \u2014 **Meta open-sources Muse Glimmer** Meta released Muse Glimmer, a 30-billion-parameter local-agent model published on Hugging Face under Apache 2.0. The company says it is built for always-on local agent work and is small enough to run on a Mac or PC with a single consumer GPU."
   ]
  },
  {
   "url": "https://lmtimeline.com",
   "title": "AI Timeline \u2014 Model Releases, AI News & History",
   "published_at": "2026-08-19T00:00:00",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "largest open model to date at 2.8 trillion parameters with a 1-million-token context window",
   "organization_evidence": [
    "## Latest in AI",
    "17. \u2014 **Moonshot AI launches Kimi K3, a 2.8-trillion-parameter open flagship** Moonshot AI began rolling out Kimi K3, which it bills as the largest open model to date at 2.8 trillion parameters, through its platform API as model ID kimi-k3."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## Latest in AI",
    "17. \u2014 **Moonshot AI launches Kimi K3, a 2.8-trillion-parameter open flagship** Moonshot AI began rolling out Kimi K3, which it bills as the largest open model to date at 2.8 trillion parameters, through its platform API as model ID kimi-k3. K3 introduces a new architecture the company calls Kimi Delta Attention \u2014 a hybrid linear-attention mechanism paired with Attention Residuals \u2014 alongside a 1-million-token context window, native image and video understanding, and always-on reasoning aimed at software engineering, knowledge work and deep reasoning."
   ]
  },
  {
   "url": "https://lmtimeline.com",
   "title": "AI Timeline \u2014 Model Releases, AI News & History",
   "published_at": "2026-08-19T00:00:00",
   "organization": "xAI",
   "model": null,
   "release_date": "2026-07-14",
   "access_type": null,
   "license": null,
   "significance": "coding CLI open-sourced on GitHub under Apache 2.0",
   "organization_evidence": [
    "## Latest in AI",
    "18. \u2014 **xAI open-sources Grok Build after its CLI was found exfiltrating full code repos** After independent researcher 'cereblab' found that xAI's Grok Build coding CLI (v0.2.93) was uploading entire Git repositories \u2014 full commit history and unredacted secrets included \u2014 to an xAI Google Cloud Storage bucket by default, even with a privacy toggle enabled, xAI disabled the upload server-side on 14 July, added an opt-out, and published the full Grok Build source on GitHub under Apache 2.0."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## Latest in AI",
    "18. \u2014 **xAI open-sources Grok Build after its CLI was found exfiltrating full code repos** After independent researcher 'cereblab' found that xAI's Grok Build coding CLI (v0.2.93) was uploading entire Git repositories \u2014 full commit history and unredacted secrets included \u2014 to an xAI Google Cloud Storage bucket by default, even with a privacy toggle enabled, xAI disabled the upload server-side on 14 July, added an opt-out, and published the full Grok Build source on GitHub under Apache 2.0."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## Latest in AI",
    "18. \u2014 **xAI open-sources Grok Build after its CLI was found exfiltrating full code repos** After independent researcher 'cereblab' found that xAI's Grok Build coding CLI (v0.2.93) was uploading entire Git repositories \u2014 full commit history and unredacted secrets included \u2014 to an xAI Google Cloud Storage bucket by default, even with a privacy toggle enabled, xAI disabled the upload server-side on 14 July, added an opt-out, and published the full Grok Build source on GitHub under Apache 2.0."
   ]
  },
  {
   "url": "https://techpillow.co/blog/kimi-k3-moonshot-ai-open-weight-frontier-model-2026",
   "title": "Kimi K3: Moonshot AI Releases Its 2.8T Open-Weight Frontier Model",
   "published_at": "2026-07-18T00:00:00",
   "organization": "Moonshot AI",
   "model": "Kimi K3",
   "release_date": "2026-07-16",
   "access_type": "open-weight",
   "license": null,
   "significance": "the largest open-weight language model ever built with 2.8 trillion total parameters, a sparse Mixture-of-Experts architecture, and a 1-million-token context window",
   "organization_evidence": [
    "![Kimi K3: Moonshot AI Releases Its 2.8T Open-Weight Frontier Model](/_next/image?url=%2Fimages%2Fblog%2Fai.png&w=3840&q=75)",
    "On 16 July 2026, Chinese AI company Moonshot AI released Kimi K3, the largest open-weight language model ever built."
   ],
   "model_evidence": [
    "![Kimi K3: Moonshot AI Releases Its 2.8T Open-Weight Frontier Model](/_next/image?url=%2Fimages%2Fblog%2Fai.png&w=3840&q=75)",
    "On 16 July 2026, Chinese AI company Moonshot AI released Kimi K3, the largest open-weight language model ever built."
   ],
   "release_date_evidence": [
    "![Kimi K3: Moonshot AI Releases Its 2.8T Open-Weight Frontier Model](/_next/image?url=%2Fimages%2Fblog%2Fai.png&w=3840&q=75)",
    "On 16 July 2026, Chinese AI company Moonshot AI released Kimi K3, the largest open-weight language model ever built."
   ],
   "access_type_evidence": [
    "![Kimi K3: Moonshot AI Releases Its 2.8T Open-Weight Frontier Model](/_next/image?url=%2Fimages%2Fblog%2Fai.png&w=3840&q=75)",
    "On 16 July 2026, Chinese AI company Moonshot AI released Kimi K3, the largest open-weight language model ever built."
   ],
   "license_evidence": null,
   "significance_evidence": [
    "![Kimi K3: Moonshot AI Releases Its 2.8T Open-Weight Frontier Model](/_next/image?url=%2Fimages%2Fblog%2Fai.png&w=3840&q=75)",
    "On 16 July 2026, Chinese AI company Moonshot AI released Kimi K3, the largest open-weight language model ever built."
   ]
  },
  {
   "url": "https://this-info.com/electronics-technology/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - This Info",
   "published_at": "2026-07-15T12:07:20",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "1.6 trillion parameters but activates only 49 billion per pass, with a 1 million token context",
   "organization_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "DeepSeek V4, released on April 24, features 1.6 trillion parameters but activates only 49 billion per pass, with a 1 million token context, and is priced at the low end of the market.",
    "APR 24",
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "DeepSeek V4, released on April 24, features 1.6 trillion parameters but activates only 49 billion per pass, with a 1 million token context, and is priced at the low end of the market.",
    "APR 24",
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "DeepSeek V4, released on April 24, features 1.6 trillion parameters but activates only 49 billion per pass, with a 1 million token context, and is priced at the low end of the market."
   ]
  },
  {
   "url": "https://this-info.com/electronics-technology/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - This Info",
   "published_at": "2026-07-15T12:07:20",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "cheap 1M-token context, native multimodal",
   "organization_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "These models include DeepSeek V4, MiniMax M3, Kimi K2.7-Code, and GLM-5.2, all available for download and most under permissive licenses.",
    "JUN 01",
    "**MiniMax M3**cheap 1M-token context, native multimodal, modified-MIT"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "Z.ai\u2019s GLM-5.2 and Kimi K2.7-Code followed in early June, with GLM-5.2 holding the top spot on the Artificial Analysis index for open-weight models, scoring 83 out of 100.",
    "JUN 01",
    "**MiniMax M3**cheap 1M-token context, native multimodal, modified-MIT"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "**MiniMax M3**cheap 1M-token context, native multimodal, modified-MIT"
   ]
  },
  {
   "url": "https://this-info.com/electronics-technology/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - This Info",
   "published_at": "2026-07-15T12:07:20",
   "organization": "Moonshot",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "agent-run specialist, ~30% fewer thinking tokens than K2.6",
   "organization_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "Kimi K2.7-Code emphasizes long-horizon agent stability, reducing thinking tokens by approximately 30% compared to its predecessor.",
    "JUN 13",
    "**Kimi K2.7-Code (Moonshot)**agent-run specialist, ~30% fewer thinking tokens than K2.6"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "Z.ai\u2019s GLM-5.2 and Kimi K2.7-Code followed in early June, with GLM-5.2 holding the top spot on the Artificial Analysis index for open-weight models, scoring 83 out of 100.",
    "JUN 13",
    "**Kimi K2.7-Code (Moonshot)**agent-run specialist, ~30% fewer thinking tokens than K2.6"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "**Kimi K2.7-Code (Moonshot)**agent-run specialist, ~30% fewer thinking tokens than K2.6"
   ]
  },
  {
   "url": "https://this-info.com/electronics-technology/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - This Info",
   "published_at": "2026-07-15T12:07:20",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "753B MoE, top open-weight on Artificial Analysis index",
   "organization_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "Z.ai\u2019s GLM-5.2 and Kimi K2.7-Code followed in early June, with GLM-5.2 holding the top spot on the Artificial Analysis index for open-weight models, scoring 83 out of 100.",
    "JUN 13\u201316",
    "**GLM-5.2 (Z.ai)**753B MoE, MIT, top open-weight on Artificial Analysis index"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "Z.ai\u2019s GLM-5.2 and Kimi K2.7-Code followed in early June, with GLM-5.2 holding the top spot on the Artificial Analysis index for open-weight models, scoring 83 out of 100.",
    "JUN 13\u201316",
    "**GLM-5.2 (Z.ai)**753B MoE, MIT, top open-weight on Artificial Analysis index"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "Z.ai\u2019s GLM-5.2 and Kimi K2.7-Code followed in early June, with GLM-5.2 holding the top spot on the Artificial Analysis index for open-weight models, scoring 83 out of 100.",
    "**GLM-5.2 (Z.ai)**753B MoE, MIT, top open-weight on Artificial Analysis index"
   ]
  },
  {
   "url": "https://opensourceforu.com/2026/07/thinking-machines-debuts-inkling-a-975b-open-weights-base-for-fine-tuning",
   "title": "Thinking Machines Debuts Inkling: A 975B Open-Weights Base For Fine-Tuning - Open Source For You",
   "published_at": "2026-07-28T06:47:09",
   "organization": "Thinking Machines",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "A 975B open-weights base model with a 1-million-token context window, built as a Mixture-of-Experts transformer supporting native text, vision, and audio processing.",
   "organization_evidence": [
    "Thinking Machines Debuts Inkling: A 975B Open-Weights Base For Fine-Tuning",
    "Mira Murati\u2019s startup, Thinking Machines, targets enterprise customisation over benchmark leadership with Inkling, its first multimodal open-weights model.",
    "On 15 July 2026, Thinking Machines Lab, a San Francisco-based AI startup founded by Mira Murati, former CTO of OpenAI, launched its first open-weights AI model, **Inkling**."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "On 15 July 2026, Thinking Machines Lab, a San Francisco-based AI startup founded by Mira Murati, former CTO of OpenAI, launched its first open-weights AI model, **Inkling**."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Thinking Machines Debuts Inkling: A 975B Open-Weights Base For Fine-Tuning",
    "The model is built as a Mixture-of-Experts (MoE) transformer with 975 billion total parameters and 41 billion active parameters per query.",
    "It supports up to a 1-million-token context window and was trained from scratch on 45 trillion tokens encompassing a multimodal blend of text, images, audio, and video."
   ]
  },
  {
   "url": "https://aifloxium.online/blog/best-open-source-llms-2026",
   "title": "Best Open-Source LLMs in 2026: 10 Models Tested & Ranked",
   "published_at": "2026-06-24T00:00:00",
   "organization": "DeepSeek",
   "model": "DeepSeek V4",
   "release_date": "2026-02",
   "access_type": "open-weight",
   "license": "MIT license",
   "significance": "Mixture-of-Experts with 1M token context window rivaling top closed models",
   "organization_evidence": [
    "Best Open-Source LLMs in 2026: 10 Models Tested & Ranked",
    "Top 10 Open Models",
    "### \ud83e\udd47 1. DeepSeek V4 \u2014 the best overall open model",
    "DeepSeek stunned everyone again."
   ],
   "model_evidence": [
    "Best Open-Source LLMs in 2026: 10 Models Tested & Ranked",
    "Top 10 Open Models",
    "### \ud83e\udd47 1. DeepSeek V4 \u2014 the best overall open model",
    "The **V4 Preview** was open-sourced under the permissive **MIT license** in early 2026, and it comes in two flagship variants:"
   ],
   "release_date_evidence": [
    "Best Open-Source LLMs in 2026: 10 Models Tested & Ranked",
    "Top 10 Open Models",
    "The **V4 Preview** was open-sourced under the permissive **MIT license** in early 2026, and it comes in two flagship variants:"
   ],
   "access_type_evidence": [
    "Best Open-Source LLMs in 2026: 10 Models Tested & Ranked",
    "\"Ranking open-weight LLMs in 2026 \u2014 DeepSeek V4, GLM-5, Qwen 3.6, Kimi K2.5, Llama 4, Gemma 4, Mistral Large \u2014 benchmarks and license guidelines.\""
   ],
   "license_evidence": [
    "Best Open-Source LLMs in 2026: 10 Models Tested & Ranked",
    "Top 10 Open Models",
    "The **V4 Preview** was open-sourced under the permissive **MIT license** in early 2026, and it comes in two flagship variants:"
   ],
   "significance_evidence": [
    "Best Open-Source LLMs in 2026: 10 Models Tested & Ranked",
    "Top 10 Open Models",
    "- **V4-Pro** \u2014 1.6T total parameters / 49B active (Mixture-of-Experts). This genuinely rivals top closed models on reasoning and math, with a **1M token context window**."
   ]
  },
  {
   "url": "https://aifloxium.online/blog/best-open-source-llms-2026",
   "title": "Best Open-Source LLMs in 2026: 10 Models Tested & Ranked",
   "published_at": "2026-06-24T00:00:00",
   "organization": "MiniMax",
   "model": "MiniMax M3",
   "release_date": "2026-06",
   "access_type": "open-weight",
   "license": null,
   "significance": "Focused heavily on agentic capability, reinforcement learning at scale, native multimodality, and computer use",
   "organization_evidence": [
    "Best Open-Source LLMs in 2026: 10 Models Tested & Ranked",
    "Top 10 Open Models",
    "**MiniMax** focused heavily on agentic capability, reinforcement learning (RL) at scale, native multimodality, and **computer use** (the model can drive a UI). M3 landed in June 2026."
   ],
   "model_evidence": [
    "Best Open-Source LLMs in 2026: 10 Models Tested & Ranked",
    "Top 10 Open Models",
    "**MiniMax** focused heavily on agentic capability, reinforcement learning (RL) at scale, native multimodality, and **computer use** (the model can drive a UI). M3 landed in June 2026."
   ],
   "release_date_evidence": [
    "Best Open-Source LLMs in 2026: 10 Models Tested & Ranked",
    "Top 10 Open Models",
    "**MiniMax** focused heavily on agentic capability, reinforcement learning (RL) at scale, native multimodality, and **computer use** (the model can drive a UI). M3 landed in June 2026."
   ],
   "access_type_evidence": [
    "Best Open-Source LLMs in 2026: 10 Models Tested & Ranked",
    "\"Ranking open-weight LLMs in 2026 \u2014 DeepSeek V4, GLM-5, Qwen 3.6, Kimi K2.5, Llama 4, Gemma 4, Mistral Large \u2014 benchmarks and license guidelines.\""
   ],
   "license_evidence": null,
   "significance_evidence": [
    "Best Open-Source LLMs in 2026: 10 Models Tested & Ranked",
    "Top 10 Open Models",
    "**MiniMax** focused heavily on agentic capability, reinforcement learning (RL) at scale, native multimodality, and **computer use** (the model can drive a UI). M3 landed in June 2026."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/gemma-4-26b-a4b-vs-qwen3-5-9b",
   "title": "Gemma 4 26B A4B (Reasoning) vs Qwen3.5 9B (Reasoning): Model Comparison",
   "published_at": null,
   "organization": "Google",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "25.2B parameters, 3.8B active at inference time; 256k context window; 26 intelligence index",
   "organization_evidence": [
    "# Gemma 4 26B A4B (Reasoning) vs. Qwen3.5 9B (Reasoning)",
    "| | ![Google logo](/img/logos/google_small.svg)Google Gemma 4 26B A4B (Reasoning) | ![Alibaba logo](/img/logos/alibaba_small.svg)Alibaba Qwen3.5 9B (Reasoning) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Gemma 4 26B A4B (Reasoning) vs. Qwen3.5 9B (Reasoning)",
    "| Release Date | April 2026 | March 2026 | Gemma 4 26B A4B (Reasoning) has a more recent release date than Qwen3.5 9B (Reasoning) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Gemma 4 26B A4B (Reasoning) vs. Qwen3.5 9B (Reasoning)",
    "| Intelligence Index | 26 | 22 | Gemma 4 26B A4B (Reasoning) is more intelligent than Qwen3.5 9B (Reasoning) |",
    "| Context Window | 256k tokens~384 A4 pages of size 12 Arial font | 262k tokens~393 A4 pages of size 12 Arial font | Qwen3.5 9B (Reasoning) has a larger context window than Gemma 4 26B A4B (Reasoning) |",
    "| Parameters | 25.2B, 3.8B active at inference time | 9.7B | Gemma 4 26B A4B (Reasoning) has more parameters than Qwen3.5 9B (Reasoning) |"
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/gemma-4-26b-a4b-vs-qwen3-5-9b",
   "title": "Gemma 4 26B A4B (Reasoning) vs Qwen3.5 9B (Reasoning): Model Comparison",
   "published_at": null,
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-03",
   "access_type": null,
   "license": null,
   "significance": "9.7B parameters; 262k context window; 22 intelligence index",
   "organization_evidence": [
    "# Gemma 4 26B A4B (Reasoning) vs. Qwen3.5 9B (Reasoning)",
    "| | ![Google logo](/img/logos/google_small.svg)Google Gemma 4 26B A4B (Reasoning) | ![Alibaba logo](/img/logos/alibaba_small.svg)Alibaba Qwen3.5 9B (Reasoning) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Gemma 4 26B A4B (Reasoning) vs. Qwen3.5 9B (Reasoning)",
    "| Release Date | April 2026 | March 2026 | Gemma 4 26B A4B (Reasoning) has a more recent release date than Qwen3.5 9B (Reasoning) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Gemma 4 26B A4B (Reasoning) vs. Qwen3.5 9B (Reasoning)",
    "| Intelligence Index | 26 | 22 | Gemma 4 26B A4B (Reasoning) is more intelligent than Qwen3.5 9B (Reasoning) |",
    "| Context Window | 256k tokens~384 A4 pages of size 12 Arial font | 262k tokens~393 A4 pages of size 12 Arial font | Qwen3.5 9B (Reasoning) has a larger context window than Gemma 4 26B A4B (Reasoning) |",
    "| Parameters | 25.2B, 3.8B active at inference time | 9.7B | Gemma 4 26B A4B (Reasoning) has more parameters than Qwen3.5 9B (Reasoning) |"
   ]
  },
  {
   "url": "https://fazm.ai/blog/open-source-llm-news-april-2026",
   "title": "Fazm - AI Computer Agent for macOS | Voice-Controlled Desktop Automation",
   "published_at": "2026-04-11T03:50:43",
   "organization": "Mistral",
   "model": null,
   "release_date": "2026-04-08",
   "access_type": null,
   "license": null,
   "significance": "Full permissive license, unlike the original Codestral",
   "organization_evidence": [
    "Mistral releases Codestral 2 under Apache 2.0",
    "Open Source LLM News April 2026: What Happened and Why It Matters",
    "## The Month at a Glance",
    "| Event | Date | Why It Matters |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Mistral releases Codestral 2 under Apache 2.0 | April 8 | Full permissive license, unlike the original Codestral |",
    "Open Source LLM News April 2026: What Happened and Why It Matters",
    "## The Month at a Glance",
    "| Event | Date | Why It Matters |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Mistral releases Codestral 2 under Apache 2.0 | April 8 | Full permissive license, unlike the original Codestral |",
    "Open Source LLM News April 2026: What Happened and Why It Matters",
    "## The Month at a Glance",
    "| Event | Date | Why It Matters |"
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/qwen3-6-plus-vs-qwen3-5-122b-a10b-non-reasoning",
   "title": "Qwen3.6 Plus vs Qwen3.5 122B A10B (Non-reasoning): Model Comparison",
   "published_at": null,
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-02",
   "access_type": null,
   "license": null,
   "significance": "A major open-weight large language model with 122 billion parameters and image input support released in February 2026.",
   "organization_evidence": [
    "# Qwen3.6 Plus vs. Qwen3.5 122B A10B (Non-reasoning)",
    "| Creator | ![Alibaba](/img/logos/alibaba_small.svg)[](https://www.alibabacloud.com/) | ![Alibaba](/img/logos/alibaba_small.svg)[](https://www.alibabacloud.com/) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Qwen3.6 Plus vs. Qwen3.5 122B A10B (Non-reasoning)",
    "| Metric | ![Alibaba logo](/img/logos/alibaba_small.svg)Qwen3.6 Plus | ![Alibaba logo](/img/logos/alibaba_small.svg)Qwen3.5 122B A10B (Non-reasoning) | Analysis |",
    "| Release Date | April, 2026 | February, 2026 | Qwen3.6 Plus has a more recent release date than Qwen3.5 122B A10B (Non-reasoning) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Qwen3.6 Plus vs. Qwen3.5 122B A10B (Non-reasoning)",
    "| Metric | ![Alibaba logo](/img/logos/alibaba_small.svg)Qwen3.6 Plus | ![Alibaba logo](/img/logos/alibaba_small.svg)Qwen3.5 122B A10B (Non-reasoning) | Analysis |",
    "| Open Source (Weights) | No | Yes[](https://huggingface.co/Qwen/Qwen3.5-122B-A10B) | Qwen3.5 122B A10B (Non-reasoning) is open source while Qwen3.6 Plus is proprietary |"
   ]
  },
  {
   "url": "https://arxiv.org/html/2604.07190v1?trk=article-ssr-frontend-pulse_little-text-block",
   "title": "The ATOM Report: Measuring the Open Language Model Ecosystem",
   "published_at": null,
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-02",
   "access_type": null,
   "license": null,
   "significance": "nearly doubling the downloads of the Llama models and capturing massive open model ecosystem adoption",
   "organization_evidence": [
    "The ATOM Report:\nMeasuring the Open Language Model Ecosystem",
    "We document a clear trend where Chinese models overtook their counterparts built in the U.S. in the summer of 2025 and subsequently widened the gap over their western counterparts.",
    "The growth of the Chinese model ecosystem can largely be attributed to Alibaba\u2019s Qwen, which is responsible for almost a billion cumulative downloads by March 2026."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The ATOM Report:\nMeasuring the Open Language Model Ecosystem",
    "we detail our findings across different data sources to show the current key trends across open model usage.",
    "The early adoption numbers of Qwen3.5 (Qwen Team, 2026) since its release in February 2026 are an indication that the dominance of Qwen relative to its peers will continue (see Section 6 and specifically Fig. 15 for early data on Qwen 3.5\u2019s adoption)."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The ATOM Report:\nMeasuring the Open Language Model Ecosystem",
    "we detail our findings across different data sources to show the current key trends across open model usage.",
    "Qwen surpassed Llama in cumulative downloads in September 2025 (325.4M vs. 323.7M) and by March 2026 reached 942.1M, thus nearly doubling the downloads of the Llama models (476.0M)."
   ]
  },
  {
   "url": "https://heybeagle.com/blog/when-does-an-open-weight-model-actually-make-sense",
   "title": "When Does an Open-Weight Model Actually Make Sense?",
   "published_at": "2026-08-04T12:02:06",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "975B open-weight model with a 1 million token context window, native multimodal reasoning, and controllable thinking effort",
   "organization_evidence": [
    "When Does an Open-Weight Model Actually Make Sense?",
    "Thinking Machines Lab shipped Inkling in July 2026 - a 975B open-weight model that admits it isn't the strongest."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "When Does an Open-Weight Model Actually Make Sense?",
    "Thinking Machines Lab shipped its first model on July 15, 2026, and led with a sentence no vendor writes willingly: \"Inkling is not the strongest overall model available today, open or closed.\""
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "When Does an Open-Weight Model Actually Make Sense?",
    "Thinking Machines Lab shipped Inkling in July 2026 - a 975B open-weight model that admits it isn't the strongest."
   ]
  },
  {
   "url": "https://jamesm.blog/ai/state-of-open-weight-models-2026",
   "title": "The State of Open-Weight Models in 2026: Llama, Qwen, Mistral, DeepSeek",
   "published_at": "2026-05-12T08:00:00",
   "organization": "Mistral",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "the EU-jurisdiction frontier choice, scoring 77.6% on SWE-Bench Verified",
   "organization_evidence": [
    "The open-weight model conversation in 2023 was about whether the open ecosystem could keep up with the frontier labs at all.",
    "Mistral has continued to ship through 2025 and into 2026, and Medium 3.5 (released [late April 2026](https://mistral.ai/)) is the EU-jurisdiction frontier choice."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The open-weight model conversation in 2023 was about whether the open ecosystem could keep up with the frontier labs at all.",
    "Mistral has continued to ship through 2025 and into 2026, and Medium 3.5 (released [late April 2026](https://mistral.ai/)) is the EU-jurisdiction frontier choice."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The open-weight model conversation in 2023 was about whether the open ecosystem could keep up with the frontier labs at all.",
    "Mistral has continued to ship through 2025 and into 2026, and Medium 3.5 (released [late April 2026](https://mistral.ai/)) is the EU-jurisdiction frontier choice.",
    "The model scores 77.6% on SWE-Bench Verified, ships under Apache 2.0, and is fully covered by EU data protection regimes - which has become a real procurement consideration for enterprises operating under GDPR, the AI Act, and the various national-level extensions that have appeared since 2025."
   ]
  },
  {
   "url": "https://rejoicehub.com/blogs/what-is-glm",
   "title": "What Is GLM-5.2? The Chinese AI Model Challenging GPT and Claude in 2026",
   "published_at": "2026-06-23T00:00:00",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "753-billion-parameter open-source large language model with a 1-million-token context window that outperforms GPT-5.5 on key coding benchmarks",
   "organization_evidence": [
    "GLM-5.2 is a 753-billion-parameter open-source large language model (LLM) developed by Z.ai (formerly Zhipu AI), released on June 13\u201316, 2026."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "GLM-5.2 is a 753-billion-parameter open-source large language model (LLM) developed by Z.ai (formerly Zhipu AI), released on June 13\u201316, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "In June 2026, GLM-5.2 emerged as one of the most talked-about AI releases of the year outperforming GPT-5.5 on key coding benchmarks, offering a massive 1-million-token context window, and doing all of this as a free, open-source model.",
    "GLM-5.2 is a 753-billion-parameter open-source large language model (LLM) developed by Z.ai (formerly Zhipu AI), released on June 13\u201316, 2026."
   ]
  },
  {
   "url": "https://kompozy.io/reviews/inkling",
   "title": "Inkling Review (2026): Honest Verdict on Thinking Machines' First Open-Weights Model",
   "published_at": "2026-05-21T00:00:00",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "a frontier-scale, natively multimodal open-weights model under Apache 2.0, rated the leading U.S. open-weights model on Artificial Analysis's Intelligence Index at launch, with calibrated answers and controllable thinking effort as real differentiators",
   "organization_evidence": [
    "# Inkling Review (2026): Honest Verdict on Thinking Machines' First Open-Weights Model",
    "Short version up top: Inkling is a landmark open release.",
    "On July 15, 2026, Thinking Machines Lab, the startup led by former OpenAI CTO Mira Murati, published its first in-house model as open weights under Apache 2.0 \u2014 an unusually open move at frontier scale."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Inkling Review (2026): Honest Verdict on Thinking Machines' First Open-Weights Model",
    "Short version up top: Inkling is a landmark open release.",
    "On July 15, 2026, Thinking Machines Lab, the startup led by former OpenAI CTO Mira Murati, published its first in-house model as open weights under Apache 2.0 \u2014 an unusually open move at frontier scale."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Inkling Review (2026): Honest Verdict on Thinking Machines' First Open-Weights Model",
    "Inkling is a genuinely significant first release from Thinking Machines Lab: a frontier-scale, natively multimodal open-weights model under Apache 2.0, rated the leading U.S. open-weights model on Artificial Analysis's Intelligence Index at launch, with calibrated answers and controllable thinking effort as real differentiators."
   ]
  },
  {
   "url": "https://llmrequirements.com/news/2026-07-11-june-open-weight-roundup",
   "title": "June 2026 open-weight roundup: Gemma 4 12B, MiniMax M3, Nemotron 3 Ultra, Kimi K2.7 Code",
   "published_at": "2026-07-11T00:00:00",
   "organization": "Google",
   "model": null,
   "release_date": "2026-06-03",
   "access_type": null,
   "license": null,
   "significance": "An 11.95-billion-parameter dense model that processes text, images, audio, and video natively and runs on any machine with 16 GB of RAM or VRAM",
   "organization_evidence": [
    "# June 2026 open-weight roundup: Gemma 4 12B, MiniMax M3, Nemotron 3 Ultra, Kimi K2.7 Code",
    "Google released Gemma 4 12B on June 3, 2026 under Apache 2.0 (per the official blog and model card)."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# June 2026 open-weight roundup: Gemma 4 12B, MiniMax M3, Nemotron 3 Ultra, Kimi K2.7 Code",
    "Google released Gemma 4 12B on June 3, 2026 under Apache 2.0 (per the official blog and model card)."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# June 2026 open-weight roundup: Gemma 4 12B, MiniMax M3, Nemotron 3 Ultra, Kimi K2.7 Code",
    "It\u2019s an 11.95-billion-parameter dense model that processes text, images, audio, and video natively and runs on any machine with 16 GB of RAM or VRAM (reported, per the Ars Technica hands-on and VentureBeat coverage, June 2026)."
   ]
  },
  {
   "url": "https://llmrequirements.com/news/2026-07-11-june-open-weight-roundup",
   "title": "June 2026 open-weight roundup: Gemma 4 12B, MiniMax M3, Nemotron 3 Ultra, Kimi K2.7 Code",
   "published_at": "2026-07-11T00:00:00",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "A mixture-of-experts model with roughly 428B total parameters and about 23B active per token, built on MiniMax Sparse Attention for 1M-token context",
   "organization_evidence": [
    "# June 2026 open-weight roundup: Gemma 4 12B, MiniMax M3, Nemotron 3 Ultra, Kimi K2.7 Code",
    "MiniMax shipped M3 on June 1, 2026, a mixture-of-experts model with roughly 428B total parameters and about 23B active per token, built on MiniMax Sparse Attention for 1M-token context (reported, per the official Hugging Face repository and MarkTechPost, June 2026)."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# June 2026 open-weight roundup: Gemma 4 12B, MiniMax M3, Nemotron 3 Ultra, Kimi K2.7 Code",
    "MiniMax shipped M3 on June 1, 2026, a mixture-of-experts model with roughly 428B total parameters and about 23B active per token, built on MiniMax Sparse Attention for 1M-token context (reported, per the official Hugging Face repository and MarkTechPost, June 2026)."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# June 2026 open-weight roundup: Gemma 4 12B, MiniMax M3, Nemotron 3 Ultra, Kimi K2.7 Code",
    "MiniMax shipped M3 on June 1, 2026, a mixture-of-experts model with roughly 428B total parameters and about 23B active per token, built on MiniMax Sparse Attention for 1M-token context (reported, per the official Hugging Face repository and MarkTechPost, June 2026)."
   ]
  },
  {
   "url": "https://llmrequirements.com/news/2026-07-11-june-open-weight-roundup",
   "title": "June 2026 open-weight roundup: Gemma 4 12B, MiniMax M3, Nemotron 3 Ultra, Kimi K2.7 Code",
   "published_at": "2026-07-11T00:00:00",
   "organization": "NVIDIA",
   "model": null,
   "release_date": "2026-06-04",
   "access_type": null,
   "license": null,
   "significance": "A 550B-parameter mixture-of-experts model with 55B active per token, using a hybrid Mamba-Transformer architecture with native multi-token prediction and a 1M-token context",
   "organization_evidence": [
    "# June 2026 open-weight roundup: Gemma 4 12B, MiniMax M3, Nemotron 3 Ultra, Kimi K2.7 Code",
    "NVIDIA launched Nemotron 3 Ultra on June 4, 2026 at Computex."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# June 2026 open-weight roundup: Gemma 4 12B, MiniMax M3, Nemotron 3 Ultra, Kimi K2.7 Code",
    "NVIDIA launched Nemotron 3 Ultra on June 4, 2026 at Computex."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# June 2026 open-weight roundup: Gemma 4 12B, MiniMax M3, Nemotron 3 Ultra, Kimi K2.7 Code",
    "It\u2019s a 550B-parameter mixture-of-experts model with 55B active per token, using a hybrid Mamba-Transformer architecture (rather than a pure transformer) with native multi-token prediction and a 1M-token context (reported, per NVIDIA\u2019s release materials and MarkTechPost, June 2026)."
   ]
  },
  {
   "url": "https://llmrequirements.com/news/2026-07-11-june-open-weight-roundup",
   "title": "June 2026 open-weight roundup: Gemma 4 12B, MiniMax M3, Nemotron 3 Ultra, Kimi K2.7 Code",
   "published_at": "2026-07-11T00:00:00",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-06-12",
   "access_type": null,
   "license": null,
   "significance": "A 1T-parameter MoE with 32B active, a 256K context, and a small vision encoder for image and video input",
   "organization_evidence": [
    "# June 2026 open-weight roundup: Gemma 4 12B, MiniMax M3, Nemotron 3 Ultra, Kimi K2.7 Code",
    "Moonshot AI released Kimi K2.7 Code on June 12, 2026 under a Modified MIT license."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# June 2026 open-weight roundup: Gemma 4 12B, MiniMax M3, Nemotron 3 Ultra, Kimi K2.7 Code",
    "Moonshot AI released Kimi K2.7 Code on June 12, 2026 under a Modified MIT license."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# June 2026 open-weight roundup: Gemma 4 12B, MiniMax M3, Nemotron 3 Ultra, Kimi K2.7 Code",
    "It\u2019s a 1T-parameter MoE with 32B active (384 experts, top-8 plus one shared), a 256K context, and a small vision encoder for image and video input (reported, per Moonshot\u2019s model card and Hugging Face repository, June 2026)."
   ]
  },
  {
   "url": "https://huggingface.co/blog/Dharma-AI/specialization-beats-scale",
   "title": "Specialization Beats Scale: A Strategic Variable Most AI Procurement Decisions Overlook",
   "published_at": "2026-05-22T15:25:59",
   "organization": "Dharma",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "A pair of specialized small language models for structured OCR that outperformed every commercial frontier API tested in a well-measured enterprise domain.",
   "organization_evidence": [
    "Specialization Beats Scale: A Strategic Variable Most AI Procurement Decisions Overlook"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Specialization Beats Scale: A Strategic Variable Most AI Procurement Decisions Overlook"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Specialization Beats Scale: A Strategic Variable Most AI Procurement Decisions Overlook",
    "A 3-billion-parameter specialized model outperformed every commercial frontier API tested in a well-measured enterprise domain \u2014 at roughly fifty times lower cost."
   ]
  },
  {
   "url": "https://robofutur.com/en/articles/chinese-open-ai-race-2026",
   "title": "China's open-weight AI race in 2026: who leads and why",
   "published_at": "2026-07-19T00:00:00",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "hybrid-attention reset aimed at long-context, agent-style tool use",
   "organization_evidence": [
    "China's open-weight AI race in 2026: who leads and why",
    "**DeepSeek** is the lab that started the wave with R1 in January 2025 and its low-cost training claims."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "China's open-weight AI race in 2026: who leads and why",
    "It closed out the V3 line in late 2025 and moved to a hybrid-attention reset with the V4 family (V4 Preview, 24 April 2026), aimed at long-context, agent-style tool use."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "China's open-weight AI race in 2026: who leads and why",
    "It closed out the V3 line in late 2025 and moved to a hybrid-attention reset with the V4 family (V4 Preview, 24 April 2026), aimed at long-context, agent-style tool use."
   ]
  },
  {
   "url": "https://robofutur.com/en/articles/chinese-open-ai-race-2026",
   "title": "China's open-weight AI race in 2026: who leads and why",
   "published_at": "2026-07-19T00:00:00",
   "organization": "Moonshot",
   "model": null,
   "release_date": "2026-07-01",
   "access_type": null,
   "license": null,
   "significance": "a 2.8-trillion-parameter mixture-of-experts model \u2014 the largest open-weight release announced to date",
   "organization_evidence": [
    "China's open-weight AI race in 2026: who leads and why",
    "**Kimi** (Moonshot) has become the agentic and coding specialist, with the K2 series ties to top closed models on software-engineering benchmarks."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "China's open-weight AI race in 2026: who leads and why",
    "Its July 2026 K3 is a 2.8-trillion-parameter mixture-of-experts model \u2014 the largest open-weight release announced to date, with weights promised later in the month."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "China's open-weight AI race in 2026: who leads and why",
    "Its July 2026 K3 is a 2.8-trillion-parameter mixture-of-experts model \u2014 the largest open-weight release announced to date, with weights promised later in the month."
   ]
  },
  {
   "url": "https://onthewire.ai/article/daily-update-16-july-2026-america-gets-its-open-weights-giant-grok-bares-its-sou",
   "title": "Daily Update \u2014 16 July 2026: America Gets Its Open-Weights Giant, Grok Bares Its Source, and Xi Speaks Tomorrow",
   "published_at": "2026-07-16T00:00:00",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "a 975-billion-parameter Mixture-of-Experts model supporting a context window of up to 1M tokens, pretrained on 45 trillion tokens of text, images, audio and video",
   "organization_evidence": [
    "# Daily Update \u2014 16 July 2026: America Gets Its Open-Weights Giant, Grok Bares Its Source, and Xi Speaks Tomorrow",
    "Thinking Machines Lab \u2014 the startup founded by former OpenAI chief technology officer Mira Murati \u2014 released **Inkling** yesterday (Wednesday 15 July): a 975-billion-parameter Mixture-of-Experts model with the full weights downloadable from Hugging Face under an Apache 2.0 licence."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Daily Update \u2014 16 July 2026: America Gets Its Open-Weights Giant, Grok Bares Its Source, and Xi Speaks Tomorrow",
    "Thinking Machines Lab \u2014 the startup founded by former OpenAI chief technology officer Mira Murati \u2014 released **Inkling** yesterday (Wednesday 15 July): a 975-billion-parameter Mixture-of-Experts model with the full weights downloadable from Hugging Face under an Apache 2.0 licence."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Daily Update \u2014 16 July 2026: America Gets Its Open-Weights Giant, Grok Bares Its Source, and Xi Speaks Tomorrow",
    "Thinking Machines Lab \u2014 the startup founded by former OpenAI chief technology officer Mira Murati \u2014 released **Inkling** yesterday (Wednesday 15 July): a 975-billion-parameter Mixture-of-Experts model with the full weights downloadable from Hugging Face under an Apache 2.0 licence.",
    "The specs, from the [announcement](https://thinkingmachines.ai/news/introducing-inkling/): \"a Mixture-of-Experts transformer with 975B total parameters, 41B active."
   ]
  },
  {
   "url": "https://venturebeat.com/technology/meet-zaya1-8b-a-super-efficient-open-reasoning-model-trained-on-amd-instinct-mi300-gpus",
   "title": "Meet ZAYA1-8B, a super efficient open reasoning model trained on AMD Instinct MI300 GPUs",
   "published_at": "2026-05-07T18:30:22",
   "organization": "Zyphra",
   "model": "ZAYA1-8B",
   "release_date": "2026-05-07",
   "access_type": "open sourcing",
   "license": "Apache 2.0",
   "significance": "A reasoning, mixture-of-experts language model with just over 8 billion parameters that retains competitive performance on benchmarks against GPT-5-High and DeepSeek-V3.2",
   "organization_evidence": [
    "Even as leading AI providers like OpenAI and Anthropic [battle over the compute](https://x.ai/news/anthropic-compute-partnership) to train and release ever larger, more powerful models, other labs are going in a different direction \u2014 pursuing the development of smaller, more efficient models and often open sourcing them.",
    "The latest worth paying attention to comes from the lesser-known Palo Alto startup [Zyphra](https://www.zyphra.com/), which this week released its new reasoning, mixture-of-experts (MoE) language model, [ZAYA1-8B](https://www.zyphra.com/post/zaya1-8b), with just over 8 billion parameters and only 760 million active \u2014 far fewer than the trillions estimated for the likes of the big labs."
   ],
   "model_evidence": [
    "Even as leading AI providers like OpenAI and Anthropic [battle over the compute](https://x.ai/news/anthropic-compute-partnership) to train and release ever larger, more powerful models, other labs are going in a different direction \u2014 pursuing the development of smaller, more efficient models and often open sourcing them.",
    "The latest worth paying attention to comes from the lesser-known Palo Alto startup [Zyphra](https://www.zyphra.com/), which this week released its new reasoning, mixture-of-experts (MoE) language model, [ZAYA1-8B](https://www.zyphra.com/post/zaya1-8b), with just over 8 billion parameters and only 760 million active \u2014 far fewer than the trillions estimated for the likes of the big labs."
   ],
   "release_date_evidence": [
    "Even as leading AI providers like OpenAI and Anthropic [battle over the compute](https://x.ai/news/anthropic-compute-partnership) to train and release ever larger, more powerful models, other labs are going in a different direction \u2014 pursuing the development of smaller, more efficient models and often open sourcing them.",
    "![ZAYA1-8B Zyphra Cloud test convo](/_next/image?url=https%3A%2F%2Fimages.ctfassets.net%2Fjdtwqhzvc2n1%2Fl5ALS2Kyf2YkqGEhYJ6G8%2Ff9f82c1824ebc72f6982b555367fcba6%2FScreenshot_2026-05-07_at_1.25.08%25C3%25A2__PM.png%3Fw%3D1000%26q%3D100&w=3840&q=75)"
   ],
   "access_type_evidence": [
    "Even as leading AI providers like OpenAI and Anthropic [battle over the compute](https://x.ai/news/anthropic-compute-partnership) to train and release ever larger, more powerful models, other labs are going in a different direction \u2014 pursuing the development of smaller, more efficient models and often open sourcing them.",
    "It can be [downloaded from Hugging Face now](https://huggingface.co/Zyphra/ZAYA1-8B) free of charge under a permissive, standard, enterprise-friendly Apache 2.0 license \u2014 and enterprises and indie developers can begin using and customizing it immediately to suit their needs."
   ],
   "license_evidence": [
    "Even as leading AI providers like OpenAI and Anthropic [battle over the compute](https://x.ai/news/anthropic-compute-partnership) to train and release ever larger, more powerful models, other labs are going in a different direction \u2014 pursuing the development of smaller, more efficient models and often open sourcing them.",
    "It can be [downloaded from Hugging Face now](https://huggingface.co/Zyphra/ZAYA1-8B) free of charge under a permissive, standard, enterprise-friendly Apache 2.0 license \u2014 and enterprises and indie developers can begin using and customizing it immediately to suit their needs."
   ],
   "significance_evidence": [
    "Even as leading AI providers like OpenAI and Anthropic [battle over the compute](https://x.ai/news/anthropic-compute-partnership) to train and release ever larger, more powerful models, other labs are going in a different direction \u2014 pursuing the development of smaller, more efficient models and often open sourcing them.",
    "The latest worth paying attention to comes from the lesser-known Palo Alto startup [Zyphra](https://www.zyphra.com/), which this week released its new reasoning, mixture-of-experts (MoE) language model, [ZAYA1-8B](https://www.zyphra.com/post/zaya1-8b), with just over 8 billion parameters and only 760 million active \u2014 far fewer than the trillions estimated for the likes of the big labs.",
    "Yet, ZAYA1-8B retains competitive performance on third-party benchmarks against GPT-5-High and DeepSeek-V3.2."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/qwen3-6-max-vs-glm-5-1",
   "title": "Qwen3.6 Max Preview vs GLM-5.1 (Reasoning): Model Comparison",
   "published_at": null,
   "organization": "Z AI",
   "model": null,
   "release_date": "April, 2026",
   "access_type": null,
   "license": null,
   "significance": null,
   "organization_evidence": [
    "# Qwen3.6 Max Preview vs. GLM-5.1 (Reasoning)",
    "| Creator | ![Alibaba](/img/logos/alibaba_small.svg)[](https://www.alibabacloud.com/) | ![Z AI](/img/logos/zai_small.svg)[](https://z.ai/) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Qwen3.6 Max Preview vs. GLM-5.1 (Reasoning)",
    "| Release Date | April, 2026 | April, 2026 | Qwen3.6 Max Preview has a more recent release date than GLM-5.1 (Reasoning) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": null
  },
  {
   "url": "https://theplanettools.ai/blog/thinking-machines-inkling-open-weights-model-mira-murati-july-2026",
   "title": "Inkling: Mira Murati's Thinking Machines Ships Its First Open-Weights Model",
   "published_at": "2026-07-16T09:00:00",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "A 975-billion-parameter mixture-of-experts model that activates 41 billion parameters per token, supports a context window up to 1 million tokens, and was pre-trained on 45 trillion tokens of text, image, audio, and video.",
   "organization_evidence": [
    "Inkling: Mira Murati's Thinking Machines Ships Its First Open-Weights Model",
    "Thinking Machines Lab released Inkling on July 15, 2026 \u2014 its first open-weights model."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Inkling: Mira Murati's Thinking Machines Ships Its First Open-Weights Model",
    "Thinking Machines Lab released Inkling on July 15, 2026 \u2014 its first open-weights model."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Inkling: Mira Murati's Thinking Machines Ships Its First Open-Weights Model",
    "Inkling is a 975-billion-parameter mixture-of-experts model that activates 41 billion parameters per token, supports a context window up to 1 million tokens, and was pre-trained on 45 trillion tokens of text, image, audio, and video."
   ]
  },
  {
   "url": "https://the-agent-report.com/2026/07/poolside-laguna-s-2-1-open-weight-coding-model-july-2026",
   "title": "Poolside Ships Laguna S 2.1: A 118B MoE Coding Model That Beats Rivals 10\u00d7 Its Size",
   "published_at": "2026-07-28T06:00:00",
   "organization": "Poolside",
   "model": null,
   "release_date": "2026-07-28",
   "access_type": null,
   "license": null,
   "significance": "a 118-billion-parameter Mixture-of-Experts coding model that activates just 8B parameters per token and beats models 10x to 20x its size on agentic coding benchmarks",
   "organization_evidence": [
    "Poolside Ships Laguna S 2.1: A 118B MoE Coding Model That Beats Rivals 10\u00d7 Its Size",
    "Poolside, a San Francisco lab that has quietly spent three years selling coding models to governments and defense agencies, just changed that."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Poolside Ships Laguna S 2.1: A 118B MoE Coding Model That Beats Rivals 10\u00d7 Its Size",
    "**TL;DR** \u2014 Poolside released Laguna S 2.1 on July 28, a 118-billion-parameter Mixture-of-Experts coding model that activates just 8B parameters per token."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Poolside Ships Laguna S 2.1: A 118B MoE Coding Model That Beats Rivals 10\u00d7 Its Size",
    "**TL;DR** \u2014 Poolside released Laguna S 2.1 on July 28, a 118-billion-parameter Mixture-of-Experts coding model that activates just 8B parameters per token."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/qwen3-6-plus-vs-qwen3-5-27b",
   "title": "Qwen3.6 Plus vs Qwen3.5 27B (Reasoning): Model Comparison",
   "published_at": null,
   "organization": "Alibaba",
   "model": null,
   "release_date": "February, 2026",
   "access_type": null,
   "license": null,
   "significance": null,
   "organization_evidence": [
    "# Qwen3.6 Plus vs. Qwen3.5 27B (Reasoning)",
    "| Metric | ![Alibaba logo](/img/logos/alibaba_small.svg)Qwen3.6 Plus | ![Alibaba logo](/img/logos/alibaba_small.svg)Qwen3.5 27B (Reasoning) | Analysis |",
    "| Creator | ![Alibaba](/img/logos/alibaba_small.svg)[](https://www.alibabacloud.com/) | ![Alibaba](/img/logos/alibaba_small.svg)[](https://www.alibabacloud.com/) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Qwen3.6 Plus vs. Qwen3.5 27B (Reasoning)",
    "| Metric | ![Alibaba logo](/img/logos/alibaba_small.svg)Qwen3.6 Plus | ![Alibaba logo](/img/logos/alibaba_small.svg)Qwen3.5 27B (Reasoning) | Analysis |",
    "| Release Date | April, 2026 | February, 2026 | Qwen3.6 Plus has a more recent release date than Qwen3.5 27B (Reasoning) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": null
  },
  {
   "url": "https://peterhalligan.substack.com/p/china-releases-its-own-ai-model-meet",
   "title": "China releases its own AI model meet \u201cGLM-5.2 an open-weight large language model developed by Chinese AI startup Z.ai",
   "published_at": "2026-06-22T11:35:59",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "outperforms OpenAI's GPT-5.5 on key coding benchmarks like FrontierSWE and SWE-bench Pro",
   "organization_evidence": [
    "# China releases its own AI model meet \u201cGLM-5.2 an open-weight large language model developed by Chinese AI startup Z.ai",
    "\u201c**GLM-5.2** is an open-weight large language model developed by Chinese AI startup **Z.ai** (formerly Zhipu AI), designed specifically for **long-horizon autonomous coding** and agentic workflows."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Launched on **June 13, 2026**, the model features **753 billion total parameters** (with 40 billion active via Mixture-of-Experts) and a **1 million-token context window**."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "- **Benchmark Performance**: GLM-5.2 outperforms **OpenAI\u2019s GPT-5.5** on key coding benchmarks like **FrontierSWE** (74.4% vs. 72.6%) and **SWE-bench Pro** (62.1 vs. 58.6)."
   ]
  },
  {
   "url": "https://opensourceforu.com/2026/08/huawei-open-sources-505b-openpangu-ai-drops-weights-and-code",
   "title": "Huawei Open Sources 505B openPangu AI, Drops Weights And Code - Open Source For You",
   "published_at": "2026-08-04T08:28:10",
   "organization": "Huawei",
   "model": null,
   "release_date": "2026-08",
   "access_type": null,
   "license": null,
   "significance": "505-billion-parameter open-weight language model that completed its entire pretraining run on Ascend 910B NPUs without using Nvidia GPUs",
   "organization_evidence": [
    "# Huawei Open Sources 505B openPangu AI, Drops Weights And Code",
    "Huawei has released the weights, inference code and technical report for openPangu-2.0-Pro, a 505-billion-parameter open-weight language model that it says completed its entire pretraining run on Ascend 910B NPUs without using Nvidia GPUs."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "[![Huawei](https://www.opensourceforu.com/wp-content/uploads/2026/08/Huawei-696x464.jpg)](https://www.opensourceforu.com/wp-content/uploads/2026/08/Huawei.jpg)",
    "Huawei has released the weights, inference code and technical report for its 505B open-weight AI model, offering researchers the first public blueprint for frontier-scale Ascend-native training without Nvidia GPUs."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Huawei Open Sources 505B openPangu AI, Drops Weights And Code",
    "Huawei has released the weights, inference code and technical report for openPangu-2.0-Pro, a 505-billion-parameter open-weight language model that it says completed its entire pretraining run on Ascend 910B NPUs without using Nvidia GPUs."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/kimi-k3-vs-claude-fable-5",
   "title": "Kimi K3 (max) vs Claude Fable 5 (Adaptive Reasoning, Max Effort, Opus 4.8 Fallback): Model Comparison | Artificial Analysis",
   "published_at": null,
   "organization": "Kimi",
   "model": null,
   "release_date": "2026-07",
   "access_type": null,
   "license": null,
   "significance": "1049k tokens context window, 60 Intelligence Index",
   "organization_evidence": [
    "# Kimi K3 (max) vs. Claude Fable 5 (Adaptive Reasoning, Max Effort, Opus 4.8 Fallback)",
    "| | ![Kimi logo](/_next/image?url=%2Fimg%2Flogos%2Fkimi.jpg&w=32&q=75)Kimi Kimi K3 (max) | ![Anthropic logo](/img/logos/anthropic_small.svg)Anthropic Claude Fable 5 (Adaptive Reasoning, Max Effort, Opus 4.8 Fallback) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Kimi K3 (max) vs. Claude Fable 5 (Adaptive Reasoning, Max Effort, Opus 4.8 Fallback)",
    "| Release Date | July 2026 | June 2026 | Kimi K3 (max) has a more recent release date than Claude Fable 5 (Adaptive Reasoning, Max Effort, Opus 4.8 Fallback) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Kimi K3 (max) vs. Claude Fable 5 (Adaptive Reasoning, Max Effort, Opus 4.8 Fallback)",
    "| Intelligence Index | 60 | 62 | Claude Fable 5 (Adaptive Reasoning, Max Effort, Opus 4.8 Fallback) is more intelligent than Kimi K3 (max) |",
    "| Context Window | 1049k tokens~1,573 A4 pages of size 12 Arial font | 1000k tokens~1,500 A4 pages of size 12 Arial font | Kimi K3 (max) has a larger context window than Claude Fable 5 (Adaptive Reasoning, Max Effort, Opus 4.8 Fallback) |"
   ]
  },
  {
   "url": "https://acingai.com/articles/open-source-llm-power-shift",
   "title": "The Open-Source LLM Power Shift: How Qwen, DeepSeek, and Mistral Changed Everything",
   "published_at": "2026-04-13T20:46:30",
   "organization": "Mistral AI",
   "model": null,
   "release_date": "2026-03",
   "access_type": null,
   "license": null,
   "significance": "A 119B MoE combining instruction following, reasoning, vision, and coding in a single model while punching well above its active parameter count",
   "organization_evidence": [
    "AI Research & Industry",
    "# The Open-Source LLM Power Shift: How Qwen, DeepSeek, and Mistral Changed Everything",
    "Mistral ships its biggest release yet with Mistral Large 3 (675B) in December 2025, followed by Mistral Small 4 in March 2026."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "AI Research & Industry",
    "# The Open-Source LLM Power Shift: How Qwen, DeepSeek, and Mistral Changed Everything",
    "Mistral ships its biggest release yet with Mistral Large 3 (675B) in December 2025, followed by Mistral Small 4 in March 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "AI Research & Industry",
    "## Mistral: Europe's Standard-Bearer",
    "- **Mistral Small 4 (March 2026):** A 119B MoE activating just 6B per token, combining instruction following, reasoning, vision, and coding in a single model. Punches well above its active parameter count."
   ]
  },
  {
   "url": "https://acingai.com/articles/open-source-llm-power-shift",
   "title": "The Open-Source LLM Power Shift: How Qwen, DeepSeek, and Mistral Changed Everything",
   "published_at": "2026-04-13T20:46:30",
   "organization": "NVIDIA",
   "model": null,
   "release_date": "2026",
   "access_type": null,
   "license": null,
   "significance": "A hybrid Mamba-Transformer MoE family with high throughput, 1M token context, and open-model licensing",
   "organization_evidence": [
    "AI Research & Industry",
    "# The Open-Source LLM Power Shift: How Qwen, DeepSeek, and Mistral Changed Everything",
    "NVIDIA rolls out the Nemotron 3 family."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "AI Research & Industry",
    "# The Open-Source LLM Power Shift: How Qwen, DeepSeek, and Mistral Changed Everything",
    "NVIDIA rolls out the Nemotron 3 family."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "AI Research & Industry",
    "## NVIDIA Nemotron: The American Entry",
    "The family uses a hybrid Mamba-Transformer MoE architecture, with three tiers:"
   ]
  },
  {
   "url": "https://budgyapp.com/z-ai-glm-5-2-china-open-source-coding-model",
   "title": "Z.ai's GLM-5.2 Coding Model Edges the Closed AI Frontier",
   "published_at": "2026-06-23T00:00:00",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "A 744-billion-parameter sparse mixture-of-experts model supporting a 1 million token context window that beats GPT-5.5 and rivals Claude Opus 4.8 on long-horizon coding benchmarks.",
   "organization_evidence": [
    "Z.ai\u2019s open-source GLM-5.2 coding model lands within a few points of Claude Opus 4.8 and beats GPT-5.5 on long-horizon benchmarks, at one-sixth the API cost.",
    "# Z.ai\u2019s GLM-5.2 Beats GPT-5.5 at One-Sixth the API Price",
    "Z.ai released the model under an unrestricted MIT license and says it carries a 1 million token context window designed for long-running coding tasks."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "GLM-5.2, a 744-billion-parameter open-source language model from Beijing-based Z.ai, launched on June 13, 2026.",
    "# Z.ai\u2019s GLM-5.2 Beats GPT-5.5 at One-Sixth the API Price"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "GLM-5.2, a 744-billion-parameter open-source language model from Beijing-based Z.ai, launched on June 13, 2026.",
    "GLM-5.2 is a sparse mixture-of-experts model with **744 billion total parameters** and roughly 40 billion active at any one time, per Z.ai\u2019s open-source release.",
    "# Z.ai\u2019s GLM-5.2 Beats GPT-5.5 at One-Sixth the API Price"
   ]
  },
  {
   "url": "https://computertech.co/arcee-trinity-large-thinking-review",
   "title": "Arcee Trinity-Large-Thinking Review 2026: The Rare Powerful US-Made Open-Source Reasoning Model - ComputerTech",
   "published_at": "2026-04-03T14:08:15",
   "organization": "Arcee AI",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "A 399B open-weights reasoning model, Apache 2.0 licensed, built in the US, ready for commercial use with zero strings attached.",
   "organization_evidence": [
    "# Arcee Trinity-Large-Thinking Review 2026: The Rare Powerful US-Made Open-Source Reasoning Model",
    "Then Arcee AI \u2014 a 30-person San Francisco lab that just spent $20 million on a single 33-day training run \u2014 dropped Trinity-Large-Thinking."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Arcee Trinity-Large-Thinking is a 399-billion parameter open-source reasoning model built by Arcee AI and released in April 2026.",
    "| **Release Date** | April 2026 |",
    "# Arcee Trinity-Large-Thinking Review 2026: The Rare Powerful US-Made Open-Source Reasoning Model"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "A 399B open-weights reasoning model, Apache 2.0 licensed, built in the US, ready for commercial use with zero strings attached.",
    "# Arcee Trinity-Large-Thinking Review 2026: The Rare Powerful US-Made Open-Source Reasoning Model"
   ]
  },
  {
   "url": "https://miniiot.medium.com/open-source-llms-in-2026-kimi-deepseek-glm-qwen-and-who-wins-what-6e6d7e484043",
   "title": "Open Source LLMs in 2026: Kimi, DeepSeek, GLM, Qwen, and Who Wins What",
   "published_at": "2026-07-09T03:53:58",
   "organization": "Moonshot",
   "model": null,
   "release_date": "2026-04-20",
   "access_type": null,
   "license": null,
   "significance": "1-trillion-parameter, 32-billion-active MoE architecture with native video understanding and a massively upgraded Agent Swarm system",
   "organization_evidence": [
    "Open Source LLMs in 2026: Kimi, DeepSeek, GLM, Qwen, and Who Wins What",
    "Three releases landed in a nine-week stretch this spring and reshaped the whole conversation: Moonshot\u2019s Kimi K2.6 on April 20, DeepSeek V4 on April 24, and Z.ai\u2019s GLM-5.2 on June 13."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Open Source LLMs in 2026: Kimi, DeepSeek, GLM, Qwen, and Who Wins What",
    "Three releases landed in a nine-week stretch this spring and reshaped the whole conversation: Moonshot\u2019s Kimi K2.6 on April 20, DeepSeek V4 on April 24, and Z.ai\u2019s GLM-5.2 on June 13."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Open Source LLMs in 2026: Kimi, DeepSeek, GLM, Qwen, and Who Wins What",
    "Moonshot\u2019s K2.6 keeps the 1-trillion-parameter, 32-billion-active MoE architecture from K2.5 but adds native video understanding and a meaningfully upgraded Agent Swarm system that can spin up to 300 sub-agents to divide a complex task and work it in parallel."
   ]
  },
  {
   "url": "https://miniiot.medium.com/open-source-llms-in-2026-kimi-deepseek-glm-qwen-and-who-wins-what-6e6d7e484043",
   "title": "Open Source LLMs in 2026: Kimi, DeepSeek, GLM, Qwen, and Who Wins What",
   "published_at": "2026-07-09T03:53:58",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "1.6 trillion total parameters with hybrid attention design (Compressed Sparse Attention paired with Heavily Compressed Attention) posting 93.5% on LiveCodeBench",
   "organization_evidence": [
    "Open Source LLMs in 2026: Kimi, DeepSeek, GLM, Qwen, and Who Wins What",
    "Three releases landed in a nine-week stretch this spring and reshaped the whole conversation: Moonshot\u2019s Kimi K2.6 on April 20, DeepSeek V4 on April 24, and Z.ai\u2019s GLM-5.2 on June 13."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Open Source LLMs in 2026: Kimi, DeepSeek, GLM, Qwen, and Who Wins What",
    "Three releases landed in a nine-week stretch this spring and reshaped the whole conversation: Moonshot\u2019s Kimi K2.6 on April 20, DeepSeek V4 on April 24, and Z.ai\u2019s GLM-5.2 on June 13."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Open Source LLMs in 2026: Kimi, DeepSeek, GLM, Qwen, and Who Wins What",
    "DeepSeek\u2019s V4 Pro runs 1.6 trillion total parameters but activates only 49 billion per token, thanks to a hybrid attention design (Compressed Sparse Attention paired with Heavily Compressed Attention) that cuts inference cost at long context dramatically."
   ]
  },
  {
   "url": "https://miniiot.medium.com/open-source-llms-in-2026-kimi-deepseek-glm-qwen-and-who-wins-what-6e6d7e484043",
   "title": "Open Source LLMs in 2026: Kimi, DeepSeek, GLM, Qwen, and Who Wins What",
   "published_at": "2026-07-09T03:53:58",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "roughly 744-billion-parameter MoE model activating about 40 billion parameters per token, leading every open model on SWE-bench Pro and Terminal-Bench 2.1",
   "organization_evidence": [
    "Open Source LLMs in 2026: Kimi, DeepSeek, GLM, Qwen, and Who Wins What",
    "Three releases landed in a nine-week stretch this spring and reshaped the whole conversation: Moonshot\u2019s Kimi K2.6 on April 20, DeepSeek V4 on April 24, and Z.ai\u2019s GLM-5.2 on June 13."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Open Source LLMs in 2026: Kimi, DeepSeek, GLM, Qwen, and Who Wins What",
    "Three releases landed in a nine-week stretch this spring and reshaped the whole conversation: Moonshot\u2019s Kimi K2.6 on April 20, DeepSeek V4 on April 24, and Z.ai\u2019s GLM-5.2 on June 13."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Open Source LLMs in 2026: Kimi, DeepSeek, GLM, Qwen, and Who Wins What",
    "Z.ai\u2019s GLM-5.2 is a roughly 744-billion-parameter MoE model that activates about 40 billion parameters per token, with a one-million-token context window and up to 128K tokens of output."
   ]
  },
  {
   "url": "https://open-techstack.com/blog/kimi-k3-moonshot-open-weights-frontier",
   "title": "Kimi K3: Moonshot's 2.8T Open-Weight Model Shakes Up the AI Race",
   "published_at": "2026-07-19T00:00:00",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-27",
   "access_type": null,
   "license": null,
   "significance": "2.8T parameters, 1M context window, open weights, matches closed models like Claude Fable 5 and GPT-5.6 Sol on key benchmarks.",
   "organization_evidence": [
    "# Kimi K3: Moonshot's 2.8T Open-Weight Model Shakes Up the AI Race",
    "Moonshot AI's Kimi K3: 2.8T parameters, 1M context, open weights July 27.",
    "Moonshot AI dropped Kimi K3 on July 17, and it\u2019s the biggest open-weight release of the year by a wide margin."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Kimi K3: Moonshot's 2.8T Open-Weight Model Shakes Up the AI Race",
    "Moonshot AI's Kimi K3: 2.8T parameters, 1M context, open weights July 27.",
    "The full weights land July 27."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Kimi K3: Moonshot's 2.8T Open-Weight Model Shakes Up the AI Race",
    "Moonshot AI's Kimi K3: 2.8T parameters, 1M context, open weights July 27.",
    "At 2.8 trillion parameters \u2014 larger than DeepSeek V4 Pro (1.6T) and Zhipu\u2019s GLM-5 (744B) combined \u2014 it\u2019s the first model to credibly claim \u201copen frontier intelligence\u201d while matching closed models like Claude Fable 5 and GPT-5.6 Sol on key benchmarks."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/glm-5",
   "title": "GLM-5 - Intelligence, Performance & Price Analysis",
   "published_at": null,
   "organization": "Z AI",
   "model": "GLM-5",
   "release_date": "2026-02",
   "access_type": "open weights",
   "license": null,
   "significance": "Reasoning model released with open weights",
   "organization_evidence": [
    "GLM-5 (Reasoning) Intelligence, Performance & Price Analysis",
    "Z AI has launched a newer model, [GLM-5.1](/models/glm-5-1)."
   ],
   "model_evidence": [
    "GLM-5 (Reasoning) Intelligence, Performance & Price Analysis",
    "\u2022\nGLM-5"
   ],
   "release_date_evidence": [
    "GLM-5 (Reasoning) Intelligence, Performance & Price Analysis",
    "\u2022\nReleased February 2026"
   ],
   "access_type_evidence": [
    "GLM-5 (Reasoning) Intelligence, Performance & Price Analysis",
    "\u2022\nOpen weights model"
   ],
   "license_evidence": null,
   "significance_evidence": [
    "GLM-5 (Reasoning) Intelligence, Performance & Price Analysis",
    "![GLM-5 (Reasoning) logo](/img/logos/zai_small.svg)",
    "\u2022\nGLM-5",
    "\u2022\nOpen weights model",
    "\u2022\nReleased February 2026"
   ]
  },
  {
   "url": "https://gncrypto.news/news/mira-murati-inkling-975b-open-source-multimodal",
   "title": "Mira Murati releases open-source model Inkling",
   "published_at": "2026-07-16T18:33:28",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "a 975-billion-parameter multimodal model with full weights on Hugging Face",
   "organization_evidence": [
    "# Mira Murati releases open-source model Inkling",
    "Thinking Machines Lab released Inkling on July 15 and published the full model weights on Hugging Face under an Apache 2.0 license."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Mira Murati releases open-source model Inkling",
    "Mira Murati released Inkling on July 15, a 975-billion-parameter multimodal model with full weights on Hugging Face under an Apache 2.0 license and fine-tuning on Tinker."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Mira Murati releases open-source model Inkling",
    "Mira Murati released Inkling on July 15, a 975-billion-parameter multimodal model with full weights on Hugging Face under an Apache 2.0 license and fine-tuning on Tinker."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/glm-5-1-vs-grok-4-20",
   "title": "GLM-5.1 (Reasoning) vs Grok 4.20 0309 v2 (Reasoning): Model Comparison",
   "published_at": null,
   "organization": "Z AI",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "A major reasoning large language model with 200k context window and open source weights released in April 2026",
   "organization_evidence": [
    "# GLM-5.1 (Reasoning) vs. Grok 4.20 0309 v2 (Reasoning)",
    "## Model Comparison",
    "| Creator | ![Z AI](/img/logos/zai_small.svg)[](https://z.ai/) | ![xAI](/img/logos/xai.svg)[](https://x.ai/) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# GLM-5.1 (Reasoning) vs. Grok 4.20 0309 v2 (Reasoning)",
    "## Model Comparison",
    "| Release Date | April, 2026 | April, 2026 | Both GLM-5.1 (Reasoning) and Grok 4.20 0309 v2 (Reasoning) have the same release date |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# GLM-5.1 (Reasoning) vs. Grok 4.20 0309 v2 (Reasoning)",
    "## Model Comparison",
    "| Context Window | 200k tokens (~300 A4 pages of size 12 Arial font) | 2000k tokens (~3000 A4 pages of size 12 Arial font) | GLM-5.1 (Reasoning) is smaller than Grok 4.20 0309 v2 (Reasoning) |",
    "| Open Source (Weights) | Yes[](https://huggingface.co/zai-org/GLM-5.1) | No | GLM-5.1 (Reasoning) is open source while Grok 4.20 0309 v2 (Reasoning) is proprietary |"
   ]
  },
  {
   "url": "https://peoplearegeek.com/articles/thinking-machines-inkling-open-weights",
   "title": "Thinking Machines Ships Inkling, a Leading US Open Model",
   "published_at": "2026-07-18T00:00:00",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "A sparse mixture of experts model with 975 billion total parameters and roughly 41 billion active per token, reading a 1 million token context window and trained from scratch on 45 trillion tokens.",
   "organization_evidence": [
    "DevNews",
    "Thinking Machines Lab has released Inkling, and it arrives as the leading open weight model from a US lab."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "DevNews",
    "Announced on July 15, 2026, the weights are already on Hugging Face under an open license, with API access through the Tinker platform."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "DevNews",
    "Thinking Machines Lab released Inkling, a sparse mixture of experts model with 975 billion total parameters and roughly 41 billion active per token."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/claude-opus-4-7-vs-qwen3-6-35b-a3b",
   "title": "Claude Opus 4.7 (Adaptive Reasoning, Max Effort) vs Qwen3.6 35B A3B (Reasoning): Model Comparison",
   "published_at": null,
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "A major reasoning model with 35B parameters, open-source weights, and image input support.",
   "organization_evidence": [
    "# Claude Opus 4.7 (Adaptive Reasoning, Max Effort) vs. Qwen3.6 35B A3B (Reasoning)",
    "| Metric | ![Anthropic logo](/img/logos/anthropic_small.svg)Claude Opus 4.7 (Adaptive Reasoning, Max Effort) | ![Alibaba logo](/img/logos/alibaba_small.svg)Qwen3.6 35B A3B (Reasoning) | Analysis |",
    "| Creator | ![Anthropic](/img/logos/anthropic_small.svg)[](https://www.anthropic.com/) | ![Alibaba](/img/logos/alibaba_small.svg)[](https://www.alibabacloud.com/) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Claude Opus 4.7 (Adaptive Reasoning, Max Effort) vs. Qwen3.6 35B A3B (Reasoning)",
    "| Metric | ![Anthropic logo](/img/logos/anthropic_small.svg)Claude Opus 4.7 (Adaptive Reasoning, Max Effort) | ![Alibaba logo](/img/logos/alibaba_small.svg)Qwen3.6 35B A3B (Reasoning) | Analysis |",
    "| Release Date | April, 2026 | April, 2026 | Both Claude Opus 4.7 (Adaptive Reasoning, Max Effort) and Qwen3.6 35B A3B (Reasoning) have the same release date |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Claude Opus 4.7 (Adaptive Reasoning, Max Effort) vs. Qwen3.6 35B A3B (Reasoning)",
    "| Metric | ![Anthropic logo](/img/logos/anthropic_small.svg)Claude Opus 4.7 (Adaptive Reasoning, Max Effort) | ![Alibaba logo](/img/logos/alibaba_small.svg)Qwen3.6 35B A3B (Reasoning) | Analysis |",
    "| Image Input Support | Yes | Yes | Both Claude Opus 4.7 (Adaptive Reasoning, Max Effort) and Qwen3.6 35B A3B (Reasoning) have image input support |",
    "| Open Source (Weights) | No | Yes[](https://huggingface.co/Qwen/Qwen3.6-35B-A3B) | Qwen3.6 35B A3B (Reasoning) is open source while Claude Opus 4.7 (Adaptive Reasoning, Max Effort) is proprietary |"
   ]
  },
  {
   "url": "https://7minai.com/news/inkling-open-weights-975b",
   "title": "Thinking Machines ships Inkling, a 975B open-weights MoE \u2014 what it means if you want a self-hostable frontier model",
   "published_at": "2026-07-16T00:00:00",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "a 975B-parameter, 41B-active multimodal Mixture-of-Experts with open weights, a 1M-token context",
   "organization_evidence": [
    "# Thinking Machines ships Inkling, a 975B open-weights MoE \u2014 what it means if you want a self-hostable frontier model",
    "Mira Murati's Thinking Machines Lab released Inkling on July 15, 2026: a 975B-parameter, 41B-active multimodal Mixture-of-Experts with open weights, a 1M-token context, and a fine-tuning path on Tinker."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Thinking Machines ships Inkling, a 975B open-weights MoE \u2014 what it means if you want a self-hostable frontier model",
    "Mira Murati's Thinking Machines Lab released Inkling on July 15, 2026: a 975B-parameter, 41B-active multimodal Mixture-of-Experts with open weights, a 1M-token context, and a fine-tuning path on Tinker."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Thinking Machines ships Inkling, a 975B open-weights MoE \u2014 what it means if you want a self-hostable frontier model",
    "Mira Murati's Thinking Machines Lab released Inkling on July 15, 2026: a 975B-parameter, 41B-active multimodal Mixture-of-Experts with open weights, a 1M-token context, and a fine-tuning path on Tinker."
   ]
  },
  {
   "url": "https://google.com/goto?url=CAEShgEB7keqTTXOH8hOEl1_YDS-I6DDS7gGvAMn0Hw35ImD0ocFWs9Siptig_agUIIerdyYI3qyJ6RvBwwAAmmSmD9xjf5HBHOeGE6e3gJ6y6jIUqFBieAncJ_aNSRhJD5SY3aFgb0X98WmBHlogJa1OLToyyHQG3KL07NOErf_Y4w2BO4-7sMEIw%3D%3D",
   "title": "Best Open-Source LLMs April 2026: Benchmarks, Licensing & Deployment Guide",
   "published_at": "2026-04-18T00:00:00",
   "organization": "Zhipu AI",
   "model": "GLM-5.1",
   "release_date": "2026-04-07",
   "access_type": "open-weight",
   "license": "MIT license",
   "significance": "754-billion-parameter Mixture-of-Experts model designed for long-horizon agentic tasks",
   "organization_evidence": [
    "Best Open-Source LLMs in April 2026: GLM-5.1, Gemma 4, Qwen 3.6, Llama 4 & More Compared",
    "1. GLM-5.1 (Zhipu AI) \u2014 Best for Agentic Coding",
    "Released April 7, 2026 under the **MIT license**, GLM-5.1 is a 754-billion-parameter Mixture-of-Experts model designed for long-horizon agentic tasks."
   ],
   "model_evidence": [
    "Best Open-Source LLMs in April 2026: GLM-5.1, Gemma 4, Qwen 3.6, Llama 4 & More Compared",
    "Released April 7, 2026 under the **MIT license**, GLM-5.1 is a 754-billion-parameter Mixture-of-Experts model designed for long-horizon agentic tasks."
   ],
   "release_date_evidence": [
    "Best Open-Source LLMs in April 2026: GLM-5.1, Gemma 4, Qwen 3.6, Llama 4 & More Compared",
    "Released April 7, 2026 under the **MIT license**, GLM-5.1 is a 754-billion-parameter Mixture-of-Experts model designed for long-horizon agentic tasks."
   ],
   "access_type_evidence": [
    "Best Open-Source LLMs in April 2026: GLM-5.1, Gemma 4, Qwen 3.6, Llama 4 & More Compared",
    "In April 2026, six major labs ship open-weight models that match or beat proprietary alternatives on key benchmarks \u2014 and you can run several of them on a single consumer GPU."
   ],
   "license_evidence": [
    "Best Open-Source LLMs in April 2026: GLM-5.1, Gemma 4, Qwen 3.6, Llama 4 & More Compared",
    "Released April 7, 2026 under the **MIT license**, GLM-5.1 is a 754-billion-parameter Mixture-of-Experts model designed for long-horizon agentic tasks."
   ],
   "significance_evidence": [
    "Best Open-Source LLMs in April 2026: GLM-5.1, Gemma 4, Qwen 3.6, Llama 4 & More Compared",
    "Released April 7, 2026 under the **MIT license**, GLM-5.1 is a 754-billion-parameter Mixture-of-Experts model designed for long-horizon agentic tasks."
   ]
  },
  {
   "url": "https://google.com/goto?url=CAEShgEB7keqTTXOH8hOEl1_YDS-I6DDS7gGvAMn0Hw35ImD0ocFWs9Siptig_agUIIerdyYI3qyJ6RvBwwAAmmSmD9xjf5HBHOeGE6e3gJ6y6jIUqFBieAncJ_aNSRhJD5SY3aFgb0X98WmBHlogJa1OLToyyHQG3KL07NOErf_Y4w2BO4-7sMEIw%3D%3D",
   "title": "Best Open-Source LLMs April 2026: Benchmarks, Licensing & Deployment Guide",
   "published_at": "2026-04-18T00:00:00",
   "organization": "Google DeepMind",
   "model": "Gemma 4",
   "release_date": "2026-04-02",
   "access_type": "open",
   "license": "Apache 2.0",
   "significance": "most capable open model you can run on consumer hardware",
   "organization_evidence": [
    "Best Open-Source LLMs in April 2026: GLM-5.1, Gemma 4, Qwen 3.6, Llama 4 & More Compared",
    "2. Gemma 4 (Google DeepMind) \u2014 Best Quality-per-VRAM",
    "Released April 2, 2026 under **Apache 2.0**, Gemma 4 is the most capable open model you can run on consumer hardware."
   ],
   "model_evidence": [
    "Best Open-Source LLMs in April 2026: GLM-5.1, Gemma 4, Qwen 3.6, Llama 4 & More Compared",
    "Released April 2, 2026 under **Apache 2.0**, Gemma 4 is the most capable open model you can run on consumer hardware."
   ],
   "release_date_evidence": [
    "Best Open-Source LLMs in April 2026: GLM-5.1, Gemma 4, Qwen 3.6, Llama 4 & More Compared",
    "Released April 2, 2026 under **Apache 2.0**, Gemma 4 is the most capable open model you can run on consumer hardware."
   ],
   "access_type_evidence": [
    "Best Open-Source LLMs in April 2026: GLM-5.1, Gemma 4, Qwen 3.6, Llama 4 & More Compared",
    "Released April 2, 2026 under **Apache 2.0**, Gemma 4 is the most capable open model you can run on consumer hardware."
   ],
   "license_evidence": [
    "Best Open-Source LLMs in April 2026: GLM-5.1, Gemma 4, Qwen 3.6, Llama 4 & More Compared",
    "Released April 2, 2026 under **Apache 2.0**, Gemma 4 is the most capable open model you can run on consumer hardware."
   ],
   "significance_evidence": [
    "Best Open-Source LLMs in April 2026: GLM-5.1, Gemma 4, Qwen 3.6, Llama 4 & More Compared",
    "Released April 2, 2026 under **Apache 2.0**, Gemma 4 is the most capable open model you can run on consumer hardware."
   ]
  },
  {
   "url": "https://google.com/goto?url=CAEShgEB7keqTTXOH8hOEl1_YDS-I6DDS7gGvAMn0Hw35ImD0ocFWs9Siptig_agUIIerdyYI3qyJ6RvBwwAAmmSmD9xjf5HBHOeGE6e3gJ6y6jIUqFBieAncJ_aNSRhJD5SY3aFgb0X98WmBHlogJa1OLToyyHQG3KL07NOErf_Y4w2BO4-7sMEIw%3D%3D",
   "title": "Best Open-Source LLMs April 2026: Benchmarks, Licensing & Deployment Guide",
   "published_at": "2026-04-18T00:00:00",
   "organization": "Alibaba",
   "model": "Qwen 3.6",
   "release_date": "2026-04-14",
   "access_type": "open-weight",
   "license": "Apache 2.0",
   "significance": "sparse MoE with 35B total parameters but only 3B active per token",
   "organization_evidence": [
    "Best Open-Source LLMs in April 2026: GLM-5.1, Gemma 4, Qwen 3.6, Llama 4 & More Compared",
    "3. Qwen 3.6 (Alibaba) \u2014 Best Efficiency at Scale",
    "Alibaba's Qwen 3.6 generation launched in two forms: the proprietary **Qwen 3.6 Plus** (March 31 \u2013 April 2, 2026) and the open-weight **Qwen 3.6-35B-A3B** (April 14, 2026)."
   ],
   "model_evidence": [
    "Best Open-Source LLMs in April 2026: GLM-5.1, Gemma 4, Qwen 3.6, Llama 4 & More Compared",
    "the open-weight **Qwen 3.6-35B-A3B** (April 14, 2026)."
   ],
   "release_date_evidence": [
    "Best Open-Source LLMs in April 2026: GLM-5.1, Gemma 4, Qwen 3.6, Llama 4 & More Compared",
    "Alibaba's Qwen 3.6 generation launched in two forms: the proprietary **Qwen 3.6 Plus** (March 31 \u2013 April 2, 2026) and the open-weight **Qwen 3.6-35B-A3B** (April 14, 2026)."
   ],
   "access_type_evidence": [
    "Best Open-Source LLMs in April 2026: GLM-5.1, Gemma 4, Qwen 3.6, Llama 4 & More Compared",
    "The open model is a sparse MoE with 35B total parameters but only 3B active per token \u2014 making it one of the most compute-efficient frontier models available."
   ],
   "license_evidence": [
    "Best Open-Source LLMs in April 2026: GLM-5.1, Gemma 4, Qwen 3.6, Llama 4 & More Compared",
    "the open-weight **Qwen 3.6-35B-A3B** (April 14, 2026).",
    "- **License:** Apache 2.0"
   ],
   "significance_evidence": [
    "Best Open-Source LLMs in April 2026: GLM-5.1, Gemma 4, Qwen 3.6, Llama 4 & More Compared",
    "The open model is a sparse MoE with 35B total parameters but only 3B active per token \u2014 making it one of the most compute-efficient frontier models available."
   ]
  },
  {
   "url": "https://google.com/goto?url=CAEShgEB7keqTTXOH8hOEl1_YDS-I6DDS7gGvAMn0Hw35ImD0ocFWs9Siptig_agUIIerdyYI3qyJ6RvBwwAAmmSmD9xjf5HBHOeGE6e3gJ6y6jIUqFBieAncJ_aNSRhJD5SY3aFgb0X98WmBHlogJa1OLToyyHQG3KL07NOErf_Y4w2BO4-7sMEIw%3D%3D",
   "title": "Best Open-Source LLMs April 2026: Benchmarks, Licensing & Deployment Guide",
   "published_at": "2026-04-18T00:00:00",
   "organization": "DeepSeek AI",
   "model": "DeepSeek V4",
   "release_date": null,
   "access_type": "open-weight",
   "license": "DeepSeek open-weight license",
   "significance": "~1 trillion parameter MoE model with ~32-37B active parameters per token",
   "organization_evidence": [
    "Best Open-Source LLMs in April 2026: GLM-5.1, Gemma 4, Qwen 3.6, Llama 4 & More Compared",
    "4. DeepSeek V4 (DeepSeek AI) \u2014 Largest Open-Weight Model",
    "DeepSeek V4 is the most ambitious open-weight release of 2026: a ~1 trillion parameter MoE model with ~32-37B active parameters per token, a 1 million token context window powered by Engram conditional memory, and native multimodal generation (text, image, video)."
   ],
   "model_evidence": [
    "Best Open-Source LLMs in April 2026: GLM-5.1, Gemma 4, Qwen 3.6, Llama 4 & More Compared",
    "DeepSeek V4 is the most ambitious open-weight release of 2026: a ~1 trillion parameter MoE model with ~32-37B active parameters per token, a 1 million token context window powered by Engram conditional memory, and native multimodal generation (text, image, video)."
   ],
   "release_date_evidence": null,
   "access_type_evidence": [
    "Best Open-Source LLMs in April 2026: GLM-5.1, Gemma 4, Qwen 3.6, Llama 4 & More Compared",
    "DeepSeek V4 is the most ambitious open-weight release of 2026: a ~1 trillion parameter MoE model with ~32-37B active parameters per token, a 1 million token context window powered by Engram conditional memory, and native multimodal generation (text, image, video)."
   ],
   "license_evidence": [
    "Best Open-Source LLMs in April 2026: GLM-5.1, Gemma 4, Qwen 3.6, Llama 4 & More Compared",
    "- **License:** DeepSeek open-weight license"
   ],
   "significance_evidence": [
    "Best Open-Source LLMs in April 2026: GLM-5.1, Gemma 4, Qwen 3.6, Llama 4 & More Compared",
    "DeepSeek V4 is the most ambitious open-weight release of 2026: a ~1 trillion parameter MoE model with ~32-37B active parameters per token, a 1 million token context window powered by Engram conditional memory, and native multimodal generation (text, image, video)."
   ]
  },
  {
   "url": "https://google.com/goto?url=CAEShgEB7keqTTXOH8hOEl1_YDS-I6DDS7gGvAMn0Hw35ImD0ocFWs9Siptig_agUIIerdyYI3qyJ6RvBwwAAmmSmD9xjf5HBHOeGE6e3gJ6y6jIUqFBieAncJ_aNSRhJD5SY3aFgb0X98WmBHlogJa1OLToyyHQG3KL07NOErf_Y4w2BO4-7sMEIw%3D%3D",
   "title": "Best Open-Source LLMs April 2026: Benchmarks, Licensing & Deployment Guide",
   "published_at": "2026-04-18T00:00:00",
   "organization": "Meta",
   "model": "Llama 4",
   "release_date": null,
   "access_type": "open model",
   "license": "Meta Llama License",
   "significance": "ships two models: Scout for efficiency and long contexts, and Maverick for raw capability",
   "organization_evidence": [
    "Best Open-Source LLMs in April 2026: GLM-5.1, Gemma 4, Qwen 3.6, Llama 4 & More Compared",
    "5. Llama 4 (Meta) \u2014 Best Multimodal Open Model",
    "Meta's Llama 4 ships two models: **Scout** (109B total, 17B active, 16 experts) for efficiency and long contexts, and **Maverick** (400B total) for raw capability."
   ],
   "model_evidence": [
    "Best Open-Source LLMs in April 2026: GLM-5.1, Gemma 4, Qwen 3.6, Llama 4 & More Compared",
    "Meta's Llama 4 ships two models: **Scout** (109B total, 17B active, 16 experts) for efficiency and long contexts, and **Maverick** (400B total) for raw capability."
   ],
   "release_date_evidence": null,
   "access_type_evidence": [
    "Best Open-Source LLMs in April 2026: GLM-5.1, Gemma 4, Qwen 3.6, Llama 4 & More Compared",
    "Meta's Llama 4 ships two models: **Scout** (109B total, 17B active, 16 experts) for efficiency and long contexts, and **Maverick** (400B total) for raw capability."
   ],
   "license_evidence": [
    "Best Open-Source LLMs in April 2026: GLM-5.1, Gemma 4, Qwen 3.6, Llama 4 & More Compared",
    "- **License:** Meta Llama License (commercial use allowed for <700M MAU)"
   ],
   "significance_evidence": [
    "Best Open-Source LLMs in April 2026: GLM-5.1, Gemma 4, Qwen 3.6, Llama 4 & More Compared",
    "Meta's Llama 4 ships two models: **Scout** (109B total, 17B active, 16 experts) for efficiency and long contexts, and **Maverick** (400B total) for raw capability."
   ]
  },
  {
   "url": "https://google.com/goto?url=CAEShgEB7keqTTXOH8hOEl1_YDS-I6DDS7gGvAMn0Hw35ImD0ocFWs9Siptig_agUIIerdyYI3qyJ6RvBwwAAmmSmD9xjf5HBHOeGE6e3gJ6y6jIUqFBieAncJ_aNSRhJD5SY3aFgb0X98WmBHlogJa1OLToyyHQG3KL07NOErf_Y4w2BO4-7sMEIw%3D%3D",
   "title": "Best Open-Source LLMs April 2026: Benchmarks, Licensing & Deployment Guide",
   "published_at": "2026-04-18T00:00:00",
   "organization": "Mistral AI",
   "model": "Mistral Small 4",
   "release_date": "2026-03-16",
   "access_type": "open-weight",
   "license": "Apache 2.0",
   "significance": "119B-parameter MoE that unifies four previously separate products into a single deployment",
   "organization_evidence": [
    "Best Open-Source LLMs in April 2026: GLM-5.1, Gemma 4, Qwen 3.6, Llama 4 & More Compared",
    "6. Mistral Small 4 (Mistral AI) \u2014 Best Unified Model",
    "Released March 16, 2026 under **Apache 2.0**, Mistral Small 4 is a 119B-parameter MoE that unifies four previously separate products (Mistral Small, Magistral, Pixtral, and Devstral) into a single deployment with configurable reasoning effort."
   ],
   "model_evidence": [
    "Best Open-Source LLMs in April 2026: GLM-5.1, Gemma 4, Qwen 3.6, Llama 4 & More Compared",
    "Released March 16, 2026 under **Apache 2.0**, Mistral Small 4 is a 119B-parameter MoE that unifies four previously separate products (Mistral Small, Magistral, Pixtral, and Devstral) into a single deployment with configurable reasoning effort."
   ],
   "release_date_evidence": [
    "Best Open-Source LLMs in April 2026: GLM-5.1, Gemma 4, Qwen 3.6, Llama 4 & More Compared",
    "Released March 16, 2026 under **Apache 2.0**, Mistral Small 4 is a 119B-parameter MoE that unifies four previously separate products (Mistral Small, Magistral, Pixtral, and Devstral) into a single deployment with configurable reasoning effort."
   ],
   "access_type_evidence": [
    "Best Open-Source LLMs in April 2026: GLM-5.1, Gemma 4, Qwen 3.6, Llama 4 & More Compared",
    "Released March 16, 2026 under **Apache 2.0**, Mistral Small 4 is a 119B-parameter MoE that unifies four previously separate products (Mistral Small, Magistral, Pixtral, and Devstral) into a single deployment with configurable reasoning effort."
   ],
   "license_evidence": [
    "Best Open-Source LLMs in April 2026: GLM-5.1, Gemma 4, Qwen 3.6, Llama 4 & More Compared",
    "Released March 16, 2026 under **Apache 2.0**, Mistral Small 4 is a 119B-parameter MoE that unifies four previously separate products (Mistral Small, Magistral, Pixtral, and Devstral) into a single deployment with configurable reasoning effort."
   ],
   "significance_evidence": [
    "Best Open-Source LLMs in April 2026: GLM-5.1, Gemma 4, Qwen 3.6, Llama 4 & More Compared",
    "Released March 16, 2026 under **Apache 2.0**, Mistral Small 4 is a 119B-parameter MoE that unifies four previously separate products (Mistral Small, Magistral, Pixtral, and Devstral) into a single deployment with configurable reasoning effort."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/gemma-4-31b-vs-gpt-5-4-pro",
   "title": "Gemma 4 31B (Reasoning) vs GPT-5.4 Pro (xhigh): Model Comparison",
   "published_at": null,
   "organization": "Google",
   "model": "Gemma 4 31B (Reasoning)",
   "release_date": "2026-04",
   "access_type": "open source",
   "license": null,
   "significance": "Reasoning model with 31B parameters and 256k context window",
   "organization_evidence": [
    "# Gemma 4 31B (Reasoning) vs. GPT-5.4 Pro (xhigh)",
    "## Model Comparison",
    "| | ![Google logo](/img/logos/google_small.svg)Google Gemma 4 31B (Reasoning) | ![OpenAI logo](/img/logos/openai_small.svg)OpenAI GPT-5.4 Pro (xhigh) | |"
   ],
   "model_evidence": [
    "# Gemma 4 31B (Reasoning) vs. GPT-5.4 Pro (xhigh)",
    "Comparison between Gemma 4 31B (Reasoning) and GPT-5.4 Pro (xhigh) across intelligence, price, speed, context window and more.",
    "## Model Comparison",
    "| | ![Google logo](/img/logos/google_small.svg)Google Gemma 4 31B (Reasoning) | ![OpenAI logo](/img/logos/openai_small.svg)OpenAI GPT-5.4 Pro (xhigh) | |"
   ],
   "release_date_evidence": [
    "# Gemma 4 31B (Reasoning) vs. GPT-5.4 Pro (xhigh)",
    "## Model Comparison",
    "| Release Date | April 2026 | March 2026 | Gemma 4 31B (Reasoning) has a more recent release date than GPT-5.4 Pro (xhigh) |"
   ],
   "access_type_evidence": [
    "# Gemma 4 31B (Reasoning) vs. GPT-5.4 Pro (xhigh)",
    "## Model Comparison",
    "| Open Source (Weights) | Yes | No | Gemma 4 31B (Reasoning) is open source while GPT-5.4 Pro (xhigh) is proprietary |"
   ],
   "license_evidence": null,
   "significance_evidence": [
    "# Gemma 4 31B (Reasoning) vs. GPT-5.4 Pro (xhigh)",
    "Comparison between Gemma 4 31B (Reasoning) and GPT-5.4 Pro (xhigh) across intelligence, price, speed, context window and more.",
    "## Model Comparison",
    "| Context Window | 256k tokens~384 A4 pages of size 12 Arial font | 1050k tokens~1,575 A4 pages of size 12 Arial font | GPT-5.4 Pro (xhigh) has a larger context window than Gemma 4 31B (Reasoning) |"
   ]
  },
  {
   "url": "https://leftbrainmarketing.net/ai-tooling/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - leftbrainmarketing.net",
   "published_at": "2026-07-15T10:56:40",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor",
   "organization_evidence": [
    "## Four Frontier-Class Open Models in Eight Weeks",
    "The Chinese open field has grown from a single lab two years ago to four major players: DeepSeek, Z.ai, Moonshot, and Alibaba, each with distinct strategic focuses.",
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## Four Frontier-Class Open Models in Eight Weeks",
    "Between late April and mid-June 2026, Chinese AI labs released four major open-weight models: **DeepSeek V4** on April 24, **MiniMax M3** on June 1, and **Kimi K2.7-Code** and **GLM-5.2** within days of each other in mid-June.",
    "APR 24",
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## Four Frontier-Class Open Models in Eight Weeks",
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor"
   ]
  },
  {
   "url": "https://leftbrainmarketing.net/ai-tooling/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - leftbrainmarketing.net",
   "published_at": "2026-07-15T10:56:40",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "cheap 1M-token context, native multimodal, modified-MIT",
   "organization_evidence": [
    "## Four Frontier-Class Open Models in Eight Weeks",
    "JUN 01"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## Four Frontier-Class Open Models in Eight Weeks",
    "Between late April and mid-June 2026, Chinese AI labs released four major open-weight models: **DeepSeek V4** on April 24, **MiniMax M3** on June 1, and **Kimi K2.7-Code** and **GLM-5.2** within days of each other in mid-June.",
    "JUN 01"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## Four Frontier-Class Open Models in Eight Weeks"
   ]
  },
  {
   "url": "https://leftbrainmarketing.net/ai-tooling/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - leftbrainmarketing.net",
   "published_at": "2026-07-15T10:56:40",
   "organization": "Moonshot",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "agent-run specialist, ~30% fewer thinking tokens than K2.6",
   "organization_evidence": [
    "## Four Frontier-Class Open Models in Eight Weeks",
    "The Chinese open field has grown from a single lab two years ago to four major players: DeepSeek, Z.ai, Moonshot, and Alibaba, each with distinct strategic focuses.",
    "JUN 13"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## Four Frontier-Class Open Models in Eight Weeks",
    "Between late April and mid-June 2026, Chinese AI labs released four major open-weight models: **DeepSeek V4** on April 24, **MiniMax M3** on June 1, and **Kimi K2.7-Code** and **GLM-5.2** within days of each other in mid-June.",
    "JUN 13"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## Four Frontier-Class Open Models in Eight Weeks"
   ]
  },
  {
   "url": "https://leftbrainmarketing.net/ai-tooling/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - leftbrainmarketing.net",
   "published_at": "2026-07-15T10:56:40",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "753B MoE, MIT, top open-weight on Artificial Analysis index",
   "organization_evidence": [
    "## Four Frontier-Class Open Models in Eight Weeks",
    "The Chinese open field has grown from a single lab two years ago to four major players: DeepSeek, Z.ai, Moonshot, and Alibaba, each with distinct strategic focuses.",
    "JUN 13\u201316"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## Four Frontier-Class Open Models in Eight Weeks",
    "Between late April and mid-June 2026, Chinese AI labs released four major open-weight models: **DeepSeek V4** on April 24, **MiniMax M3** on June 1, and **Kimi K2.7-Code** and **GLM-5.2** within days of each other in mid-June.",
    "JUN 13\u201316"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## Four Frontier-Class Open Models in Eight Weeks"
   ]
  },
  {
   "url": "https://ai-master.dev/out/computingforgeeks-computingforgeeks",
   "title": "Open Source LLM Comparison Table (2026) | ComputingForGeeks",
   "published_at": "2026-03-28T07:53:41",
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-02",
   "access_type": null,
   "license": null,
   "significance": "A major new natively multimodal model-family generation with 256K context windows and strong performance that rivals closed-source alternatives.",
   "organization_evidence": [
    "This reference covers every major open-source and open-weight large language model available as of March 2026, with verified benchmark scores, license terms, hardware requirements, and hands-on performance data from real self-hosting tests.",
    "| **Qwen 3.5 397B-A17B** | Alibaba | 397B | 17B | MoE | 256K tokens | Text + Image | Apache 2.0 | Feb 2026 |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "This reference covers every major open-source and open-weight large language model available as of March 2026, with verified benchmark scores, license terms, hardware requirements, and hands-on performance data from real self-hosting tests.",
    "| **Qwen 3.5 397B-A17B** | Alibaba | 397B | 17B | MoE | 256K tokens | Text + Image | Apache 2.0 | Feb 2026 |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "This reference covers every major open-source and open-weight large language model available as of March 2026, with verified benchmark scores, license terms, hardware requirements, and hands-on performance data from real self-hosting tests.",
    "Qwen 3.5, released in waves from February to March 2026, represents a generational leap."
   ]
  },
  {
   "url": "https://ai-master.dev/out/computingforgeeks-computingforgeeks",
   "title": "Open Source LLM Comparison Table (2026) | ComputingForGeeks",
   "published_at": "2026-03-28T07:53:41",
   "organization": "Zhipu AI",
   "model": null,
   "release_date": "2026-02",
   "access_type": null,
   "license": null,
   "significance": "A 744B total parameter MoE model trained entirely on Huawei Ascend chips, posting 77.8% on SWE-bench Verified.",
   "organization_evidence": [
    "This reference covers every major open-source and open-weight large language model available as of March 2026, with verified benchmark scores, license terms, hardware requirements, and hands-on performance data from real self-hosting tests.",
    "| **GLM-5** | Zhipu AI | 744B | 40B | MoE | 205K tokens | Text + Image | MIT | Feb 2026 |",
    "GLM-5, released in February 2026 by Chinese AI lab Zhipu AI, is one of the most significant open-source releases of the year."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "This reference covers every major open-source and open-weight large language model available as of March 2026, with verified benchmark scores, license terms, hardware requirements, and hands-on performance data from real self-hosting tests.",
    "| **GLM-5** | Zhipu AI | 744B | 40B | MoE | 205K tokens | Text + Image | MIT | Feb 2026 |",
    "GLM-5, released in February 2026 by Chinese AI lab Zhipu AI, is one of the most significant open-source releases of the year."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "This reference covers every major open-source and open-weight large language model available as of March 2026, with verified benchmark scores, license terms, hardware requirements, and hands-on performance data from real self-hosting tests.",
    "GLM-5, released in February 2026 by Chinese AI lab Zhipu AI, is one of the most significant open-source releases of the year.",
    "At 744B total parameters with 40B active, it\u2019s a large MoE model under the MIT license with no usage restrictions.",
    "The model posts 50.4% on Humanity\u2019s Last Exam and 77.8% on SWE-bench Verified, the latter being the strongest coding benchmark result among open models."
   ]
  },
  {
   "url": "https://ai-master.dev/out/computingforgeeks-computingforgeeks",
   "title": "Open Source LLM Comparison Table (2026) | ComputingForGeeks",
   "published_at": "2026-03-28T07:53:41",
   "organization": "Mistral AI",
   "model": null,
   "release_date": "2026-03",
   "access_type": null,
   "license": null,
   "significance": "Unifies instruction following, reasoning, and multimodal capabilities under Apache 2.0 at 6B active parameters.",
   "organization_evidence": [
    "This reference covers every major open-source and open-weight large language model available as of March 2026, with verified benchmark scores, license terms, hardware requirements, and hands-on performance data from real self-hosting tests.",
    "| **Mistral Small 4** | Mistral AI | 119B | 6B | MoE (128e, 4 active) | 256K tokens | Text + Image | Apache 2.0 | Mar 2026 |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "This reference covers every major open-source and open-weight large language model available as of March 2026, with verified benchmark scores, license terms, hardware requirements, and hands-on performance data from real self-hosting tests.",
    "| **Mistral Small 4** | Mistral AI | 119B | 6B | MoE (128e, 4 active) | 256K tokens | Text + Image | Apache 2.0 | Mar 2026 |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "This reference covers every major open-source and open-weight large language model available as of March 2026, with verified benchmark scores, license terms, hardware requirements, and hands-on performance data from real self-hosting tests.",
    "Mistral Small 4, released in March 2026, is the most interesting recent addition."
   ]
  },
  {
   "url": "https://ai-master.dev/out/computingforgeeks-computingforgeeks",
   "title": "Open Source LLM Comparison Table (2026) | ComputingForGeeks",
   "published_at": "2026-03-28T07:53:41",
   "organization": "Microsoft",
   "model": null,
   "release_date": "2026-03-04",
   "access_type": null,
   "license": null,
   "significance": "Adds image understanding to the reasoning pipeline with dynamic resolution support under the MIT license.",
   "organization_evidence": [
    "This reference covers every major open-source and open-weight large language model available as of March 2026, with verified benchmark scores, license terms, hardware requirements, and hands-on performance data from real self-hosting tests.",
    "| **Phi-4 Reasoning Vision** | Microsoft | 15B | 15B | Dense | 16K tokens | Text + Image | MIT | Mar 2026 |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "This reference covers every major open-source and open-weight large language model available as of March 2026, with verified benchmark scores, license terms, hardware requirements, and hands-on performance data from real self-hosting tests.",
    "| **Phi-4 Reasoning Vision** | Microsoft | 15B | 15B | Dense | 16K tokens | Text + Image | MIT | Mar 2026 |",
    "Phi-4 Reasoning Vision (15B), released March 4, 2026, adds image understanding to the reasoning pipeline."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "This reference covers every major open-source and open-weight large language model available as of March 2026, with verified benchmark scores, license terms, hardware requirements, and hands-on performance data from real self-hosting tests.",
    "Phi-4 Reasoning Vision (15B), released March 4, 2026, adds image understanding to the reasoning pipeline."
   ]
  },
  {
   "url": "https://blockchain-council.org/ai/meet-kimi-k3",
   "title": "Meet KIMI K3",
   "published_at": "2026-07-20T07:00:07",
   "organization": "Moonshot AI",
   "model": "KIMI K3",
   "release_date": "2026-07-16",
   "access_type": "open-weight",
   "license": "Modified MIT license",
   "significance": "the most powerful open-weight large language model ever released at 2.8 trillion total parameters",
   "organization_evidence": [
    "This guide covers the complete picture: what KIMI K3 is, how its architecture works, what it benchmarks, how it is priced, how it compares to the frontier, and why it matters for the global AI industry.",
    "On July 16, 2026, Moonshot AI launched **KIMI K3**, the most powerful open-weight large language model ever released."
   ],
   "model_evidence": [
    "This guide covers the complete picture: what KIMI K3 is, how its architecture works, what it benchmarks, how it is priced, how it compares to the frontier, and why it matters for the global AI industry.",
    "On July 16, 2026, Moonshot AI launched **KIMI K3**, the most powerful open-weight large language model ever released."
   ],
   "release_date_evidence": [
    "This guide covers the complete picture: what KIMI K3 is, how its architecture works, what it benchmarks, how it is priced, how it compares to the frontier, and why it matters for the global AI industry.",
    "On July 16, 2026, Moonshot AI launched **KIMI K3**, the most powerful open-weight large language model ever released."
   ],
   "access_type_evidence": [
    "This guide covers the complete picture: what KIMI K3 is, how its architecture works, what it benchmarks, how it is priced, how it compares to the frontier, and why it matters for the global AI industry.",
    "On July 16, 2026, Moonshot AI launched **KIMI K3**, the most powerful open-weight large language model ever released."
   ],
   "license_evidence": [
    "This guide covers the complete picture: what KIMI K3 is, how its architecture works, what it benchmarks, how it is priced, how it compares to the frontier, and why it matters for the global AI industry.",
    "**KIMI K3** is not merely a parameter count story.",
    "It introduces a new architecture built on Kimi Delta Attention, a native vision and multimodal capability, an always-on reasoning mode called thinking mode, a one-million-token context window, and an open-weight release scheduled for July 27, 2026, under a Modified MIT license."
   ],
   "significance_evidence": [
    "This guide covers the complete picture: what KIMI K3 is, how its architecture works, what it benchmarks, how it is priced, how it compares to the frontier, and why it matters for the global AI industry.",
    "On July 16, 2026, Moonshot AI launched **KIMI K3**, the most powerful open-weight large language model ever released.",
    "At 2.8 trillion total parameters, it is the first open 3-trillion-class model in AI history, surpassing DeepSeek's previous record of 1.6 trillion parameters and arriving as a direct challenger to the most advanced proprietary systems from Anthropic and OpenAI."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/gemma-4-26b-a4b-vs-qwen3-5-27b-non-reasoning",
   "title": "Gemma 4 26B A4B (Reasoning) vs Qwen3.5 27B (Non-reasoning): Model Comparison",
   "published_at": null,
   "organization": "Google",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "25.2B parameters with 3.8B active at inference time, 256k context window, and image input support",
   "organization_evidence": [
    "# Gemma 4 26B A4B (Reasoning) vs. Qwen3.5 27B (Non-reasoning)",
    "| Creator | ![Google](/img/logos/google_small.svg)[](https://deepmind.google/technologies/gemini/) | ![Alibaba](/img/logos/alibaba_small.svg)[](https://www.alibabacloud.com/) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Gemma 4 26B A4B (Reasoning) vs. Qwen3.5 27B (Non-reasoning)",
    "| Release Date | April, 2026 | February, 2026 | Gemma 4 26B A4B (Reasoning) has a more recent release date than Qwen3.5 27B (Non-reasoning) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Gemma 4 26B A4B (Reasoning) vs. Qwen3.5 27B (Non-reasoning)",
    "| Context Window | 256k tokens (~384 A4 pages of size 12 Arial font) | 262k tokens (~393 A4 pages of size 12 Arial font) | Gemma 4 26B A4B (Reasoning) is smaller than Qwen3.5 27B (Non-reasoning) |",
    "| Parameters | 25.2B, 3.8B active at inference time | 27.8B | Gemma 4 26B A4B (Reasoning) is smaller than Qwen3.5 27B (Non-reasoning) |",
    "| Image Input Support | Yes | Yes | Both Gemma 4 26B A4B (Reasoning) and Qwen3.5 27B (Non-reasoning) have image input support |"
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/gemma-4-26b-a4b-vs-qwen3-5-27b-non-reasoning",
   "title": "Gemma 4 26B A4B (Reasoning) vs Qwen3.5 27B (Non-reasoning): Model Comparison",
   "published_at": null,
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-02",
   "access_type": null,
   "license": null,
   "significance": "27.8B parameters, 262k context window, and image input support",
   "organization_evidence": [
    "# Gemma 4 26B A4B (Reasoning) vs. Qwen3.5 27B (Non-reasoning)",
    "| Creator | ![Google](/img/logos/google_small.svg)[](https://deepmind.google/technologies/gemini/) | ![Alibaba](/img/logos/alibaba_small.svg)[](https://www.alibabacloud.com/) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Gemma 4 26B A4B (Reasoning) vs. Qwen3.5 27B (Non-reasoning)",
    "| Release Date | April, 2026 | February, 2026 | Gemma 4 26B A4B (Reasoning) has a more recent release date than Qwen3.5 27B (Non-reasoning) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Gemma 4 26B A4B (Reasoning) vs. Qwen3.5 27B (Non-reasoning)",
    "| Context Window | 256k tokens (~384 A4 pages of size 12 Arial font) | 262k tokens (~393 A4 pages of size 12 Arial font) | Gemma 4 26B A4B (Reasoning) is smaller than Qwen3.5 27B (Non-reasoning) |",
    "| Parameters | 25.2B, 3.8B active at inference time | 27.8B | Gemma 4 26B A4B (Reasoning) is smaller than Qwen3.5 27B (Non-reasoning) |",
    "| Image Input Support | Yes | Yes | Both Gemma 4 26B A4B (Reasoning) and Qwen3.5 27B (Non-reasoning) have image input support |"
   ]
  },
  {
   "url": "https://frontiernews.ai/news/article/why-chinese-ai-labs-just-caught-up-to-openais-fron-50fca749",
   "title": "Why Chinese AI Labs Just Caught Up to OpenAI's Frontier Models",
   "published_at": "2026-07-05T12:22:42",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "All-round leader that scored 51 on the Artificial Analysis Intelligence Index and achieved 62.1% on SWE-bench Pro",
   "organization_evidence": [
    "# Why Chinese AI Labs Just Caught Up to OpenAI's Frontier Models",
    "When Z.ai released GLM-5.2 in mid-June 2026, the open-weight model landed fourth overall on the industry's headline intelligence benchmark, ahead of every proprietary system except OpenAI's GPT-5.5, Anthropic's Claude Opus 4.8, and Fable 5."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Why Chinese AI Labs Just Caught Up to OpenAI's Frontier Models",
    "Then Z.ai closed the quarter with GLM-5.2 on June 13."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Why Chinese AI Labs Just Caught Up to OpenAI's Frontier Models",
    "GLM-5.2 is the new all-round leader."
   ]
  },
  {
   "url": "https://frontiernews.ai/news/article/why-chinese-ai-labs-just-caught-up-to-openais-fron-50fca749",
   "title": "Why Chinese AI Labs Just Caught Up to OpenAI's Frontier Models",
   "published_at": "2026-07-05T12:22:42",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-04-20",
   "access_type": null,
   "license": null,
   "significance": "Trillion-parameter agentic specialist and natively multimodal member handling text, images, and video",
   "organization_evidence": [
    "# Why Chinese AI Labs Just Caught Up to OpenAI's Frontier Models",
    "Moonshot AI shipped Kimi K2.6 on April 20, 2026."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Why Chinese AI Labs Just Caught Up to OpenAI's Frontier Models",
    "Moonshot AI shipped Kimi K2.6 on April 20, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Why Chinese AI Labs Just Caught Up to OpenAI's Frontier Models",
    "Kimi K2.6 is the trillion-parameter agentic specialist."
   ]
  },
  {
   "url": "https://frontiernews.ai/news/article/why-chinese-ai-labs-just-caught-up-to-openais-fron-50fca749",
   "title": "Why Chinese AI Labs Just Caught Up to OpenAI's Frontier Models",
   "published_at": "2026-07-05T12:22:42",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "Efficiency and long-context champion featuring a hybrid attention stack",
   "organization_evidence": [
    "# Why Chinese AI Labs Just Caught Up to OpenAI's Frontier Models",
    "DeepSeek followed with DeepSeek V4 on April 24."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Why Chinese AI Labs Just Caught Up to OpenAI's Frontier Models",
    "DeepSeek followed with DeepSeek V4 on April 24."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Why Chinese AI Labs Just Caught Up to OpenAI's Frontier Models",
    "DeepSeek V4-Pro is the efficiency and long-context champion."
   ]
  },
  {
   "url": "https://agenticbrew.ai/news/69de7353-e30a-45b8-ad68-59f2aad0ddca/thinking-machines-lab-launches-inkling-open-weights-model",
   "title": "Thinking Machines Lab launches Inkling open-weights model",
   "published_at": "2026-07-16T17:22:23",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "a 975B-parameter multimodal Mixture-of-Experts model with 41B active parameters, a 1M-token context window",
   "organization_evidence": [
    "Thinking Machines Lab released Inkling on July 15, 2026 - a 975B-parameter multimodal Mixture-of-Experts model with 41B active parameters, a 1M-token context window, and Apache 2.0 licensing."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Thinking Machines Lab released Inkling on July 15, 2026 - a 975B-parameter multimodal Mixture-of-Experts model with 41B active parameters, a 1M-token context window, and Apache 2.0 licensing."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Thinking Machines Lab released Inkling on July 15, 2026 - a 975B-parameter multimodal Mixture-of-Experts model with 41B active parameters, a 1M-token context window, and Apache 2.0 licensing."
   ]
  },
  {
   "url": "https://taskade.com/blog/open-source-llms",
   "title": "10 Best Open-Source LLMs in July 2026 (Ranked for Real Work)",
   "published_at": "2026-05-23T08:00:00",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-16",
   "access_type": null,
   "license": null,
   "significance": "largest open-weight model released to date with 2.8-trillion-parameter mixture-of-experts, native vision, and 1-million-token context window",
   "organization_evidence": [
    "# 10 Best Open-Source LLMs in July 2026 (Ranked for Real Work)",
    "## What Changed Since May 2026",
    "- **July 16, 2026** \u2014 Moonshot shipped **Kimi K3**, a **2.8-trillion-parameter** open-weight mixture-of-experts model (only ~16 of 896 experts active per token) with native vision and a **1-million-token** context window."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# 10 Best Open-Source LLMs in July 2026 (Ranked for Real Work)",
    "## What Changed Since May 2026",
    "- **July 16, 2026** \u2014 Moonshot shipped **Kimi K3**, a **2.8-trillion-parameter** open-weight mixture-of-experts model (only ~16 of 896 experts active per token) with native vision and a **1-million-token** context window."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# 10 Best Open-Source LLMs in July 2026 (Ranked for Real Work)",
    "## What Changed Since May 2026",
    "- **July 16, 2026** \u2014 Moonshot shipped **Kimi K3**, a **2.8-trillion-parameter** open-weight mixture-of-experts model (only ~16 of 896 experts active per token) with native vision and a **1-million-token** context window. It is the **largest open-weight model released to date**, and early independent reviews place it at frontier level for agentic coding; full open weights are scheduled for July 27."
   ]
  },
  {
   "url": "https://taskade.com/blog/open-source-llms",
   "title": "10 Best Open-Source LLMs in July 2026 (Ranked for Real Work)",
   "published_at": "2026-05-23T08:00:00",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-06-12",
   "access_type": null,
   "license": null,
   "significance": "+21.8% over K2.6 on Kimi Code Bench v2, plus a HighSpeed variant with ~6x faster inference",
   "organization_evidence": [
    "# 10 Best Open-Source LLMs in July 2026 (Ranked for Real Work)",
    "## What Changed Since May 2026",
    "- **June 12, 2026** \u2014 Moonshot shipped **Kimi K2.7 Code**: +21.8% over K2.6 on Kimi Code Bench v2, plus a HighSpeed variant with ~6\u00d7 faster inference."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# 10 Best Open-Source LLMs in July 2026 (Ranked for Real Work)",
    "## What Changed Since May 2026",
    "- **June 12, 2026** \u2014 Moonshot shipped **Kimi K2.7 Code**: +21.8% over K2.6 on Kimi Code Bench v2, plus a HighSpeed variant with ~6\u00d7 faster inference."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# 10 Best Open-Source LLMs in July 2026 (Ranked for Real Work)",
    "## What Changed Since May 2026",
    "- **June 12, 2026** \u2014 Moonshot shipped **Kimi K2.7 Code**: +21.8% over K2.6 on Kimi Code Bench v2, plus a HighSpeed variant with ~6\u00d7 faster inference."
   ]
  },
  {
   "url": "https://taskade.com/blog/open-source-llms",
   "title": "10 Best Open-Source LLMs in July 2026 (Ranked for Real Work)",
   "published_at": "2026-05-23T08:00:00",
   "organization": "Zhipu AI",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "91.2% on GPQA Diamond and long-horizon coding matching or beating GPT-5.5 at roughly 1/6 the cost",
   "organization_evidence": [
    "# 10 Best Open-Source LLMs in July 2026 (Ranked for Real Work)",
    "## What Changed Since May 2026",
    "- **June 13, 2026** \u2014 Zhipu released **GLM-5.2**; MIT weights landed on Hugging Face around June 17. The new #1."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# 10 Best Open-Source LLMs in July 2026 (Ranked for Real Work)",
    "## What Changed Since May 2026",
    "- **June 13, 2026** \u2014 Zhipu released **GLM-5.2**; MIT weights landed on Hugging Face around June 17. The new #1."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# 10 Best Open-Source LLMs in July 2026 (Ranked for Real Work)",
    "## 1. GLM-5.2: The New Open-Source Frontier",
    "**Benchmark snapshot (June 2026 release):** GPQA Diamond **91.2%** \u00b7 matches or beats GPT-5.5 on long-horizon coding tasks at roughly **1/6 the cost**."
   ]
  },
  {
   "url": "https://taskade.com/blog/open-source-llms",
   "title": "10 Best Open-Source LLMs in July 2026 (Ranked for Real Work)",
   "published_at": "2026-05-23T08:00:00",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-06",
   "access_type": null,
   "license": null,
   "significance": "top open-weight score of 59.0% on SWE-bench Pro, 1 million context window, and native multimodal",
   "organization_evidence": [
    "# 10 Best Open-Source LLMs in July 2026 (Ranked for Real Work)",
    "## What Changed Since May 2026",
    "- **June 2026** \u2014 **MiniMax M3** posted 59.0% on SWE-bench Pro, the top open-weight score."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# 10 Best Open-Source LLMs in July 2026 (Ranked for Real Work)",
    "## What Changed Since May 2026",
    "- **June 2026** \u2014 **MiniMax M3** posted 59.0% on SWE-bench Pro, the top open-weight score."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# 10 Best Open-Source LLMs in July 2026 (Ranked for Real Work)",
    "## 4. MiniMax M3: The Open-Weight SWE-bench Pro Leader",
    "**Benchmark snapshot:** SWE-bench Pro **59.0%** \u2014 the top open-weight score as of July 2026, edging past the record Kimi K2.6 set in April (58.6%)."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/qwen3-6-plus-vs-qwen3-5-397b-a17b-non-reasoning",
   "title": "Qwen3.6 Plus vs Qwen3.5 397B A17B (Non-reasoning): Model Comparison",
   "published_at": null,
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-02",
   "access_type": null,
   "license": null,
   "significance": "Qwen3.5 397B A17B (Non-reasoning) is open source",
   "organization_evidence": [
    "# Qwen3.6 Plus vs. Qwen3.5 397B A17B (Non-reasoning)",
    "| Metric | ![Alibaba logo](/img/logos/alibaba_small.svg)Qwen3.6 Plus | ![Alibaba logo](/img/logos/alibaba_small.svg)Qwen3.5 397B A17B (Non-reasoning) | Analysis |",
    "| Creator | ![Alibaba](/img/logos/alibaba_small.svg)[](https://www.alibabacloud.com/) | ![Alibaba](/img/logos/alibaba_small.svg)[](https://www.alibabacloud.com/) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Qwen3.6 Plus vs. Qwen3.5 397B A17B (Non-reasoning)",
    "| Metric | ![Alibaba logo](/img/logos/alibaba_small.svg)Qwen3.6 Plus | ![Alibaba logo](/img/logos/alibaba_small.svg)Qwen3.5 397B A17B (Non-reasoning) | Analysis |",
    "| Release Date | April, 2026 | February, 2026 | Qwen3.6 Plus has a more recent release date than Qwen3.5 397B A17B (Non-reasoning) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Qwen3.6 Plus vs. Qwen3.5 397B A17B (Non-reasoning)",
    "| Metric | ![Alibaba logo](/img/logos/alibaba_small.svg)Qwen3.6 Plus | ![Alibaba logo](/img/logos/alibaba_small.svg)Qwen3.5 397B A17B (Non-reasoning) | Analysis |",
    "| Open Source (Weights) | No | Yes[](https://huggingface.co/Qwen/Qwen3.5-397B-A17B) | Qwen3.5 397B A17B (Non-reasoning) is open source while Qwen3.6 Plus is proprietary |"
   ]
  },
  {
   "url": "https://af.net/realtime/new-open-source-llm-releases-in-april-2026-key-models-and-innovations",
   "title": "New Open Source LLM Releases in April 2026: Key Models and Innovations | AIFOD | AI FOR DEVELOPING COUNTRIES FORUM",
   "published_at": null,
   "organization": "Meta",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "ax\u00e9es sur l'efficacit\u00e9 et la scalabilit\u00e9",
   "organization_evidence": [
    "Avril 2026 marque un mois significatif pour la communaut\u00e9 IA open source, avec des acteurs majeurs comme Meta, Alibaba et Google d\u00e9voilant de nouvelles it\u00e9rations de leurs mod\u00e8les de langage de grande taille (LLM).",
    "Meta a introduit Llama 4 en deux variantes, Scout et Maverick, ax\u00e9es sur l'efficacit\u00e9 et la scalabilit\u00e9."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Avril 2026 marque un mois significatif pour la communaut\u00e9 IA open source, avec des acteurs majeurs comme Meta, Alibaba et Google d\u00e9voilant de nouvelles it\u00e9rations de leurs mod\u00e8les de langage de grande taille (LLM)."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Meta a introduit Llama 4 en deux variantes, Scout et Maverick, ax\u00e9es sur l'efficacit\u00e9 et la scalabilit\u00e9."
   ]
  },
  {
   "url": "https://af.net/realtime/new-open-source-llm-releases-in-april-2026-key-models-and-innovations",
   "title": "New Open Source LLM Releases in April 2026: Key Models and Innovations | AIFOD | AI FOR DEVELOPING COUNTRIES FORUM",
   "published_at": null,
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "un mod\u00e8le optimis\u00e9 pour les t\u00e2ches multilingues, excelling particuli\u00e8rement dans les langues asiatiques",
   "organization_evidence": [
    "Avril 2026 marque un mois significatif pour la communaut\u00e9 IA open source, avec des acteurs majeurs comme Meta, Alibaba et Google d\u00e9voilant de nouvelles it\u00e9rations de leurs mod\u00e8les de langage de grande taille (LLM).",
    "Pendant ce temps, Alibaba a lanc\u00e9 Qwen 3, un mod\u00e8le optimis\u00e9 pour les t\u00e2ches multilingues, excelling particuli\u00e8rement dans les langues asiatiques."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Avril 2026 marque un mois significatif pour la communaut\u00e9 IA open source, avec des acteurs majeurs comme Meta, Alibaba et Google d\u00e9voilant de nouvelles it\u00e9rations de leurs mod\u00e8les de langage de grande taille (LLM)."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Pendant ce temps, Alibaba a lanc\u00e9 Qwen 3, un mod\u00e8le optimis\u00e9 pour les t\u00e2ches multilingues, excelling particuli\u00e8rement dans les langues asiatiques."
   ]
  },
  {
   "url": "https://af.net/realtime/new-open-source-llm-releases-in-april-2026-key-models-and-innovations",
   "title": "New Open Source LLM Releases in April 2026: Key Models and Innovations | AIFOD | AI FOR DEVELOPING COUNTRIES FORUM",
   "published_at": null,
   "organization": "Google",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "offrant des fen\u00eatres contextuelles \u00e9largies et des capacit\u00e9s de fine-tuning am\u00e9lior\u00e9es",
   "organization_evidence": [
    "Avril 2026 marque un mois significatif pour la communaut\u00e9 IA open source, avec des acteurs majeurs comme Meta, Alibaba et Google d\u00e9voilant de nouvelles it\u00e9rations de leurs mod\u00e8les de langage de grande taille (LLM).",
    "Google a rejoint la course avec Gemma 3n, offrant des fen\u00eatres contextuelles \u00e9largies et des capacit\u00e9s de fine-tuning am\u00e9lior\u00e9es."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Avril 2026 marque un mois significatif pour la communaut\u00e9 IA open source, avec des acteurs majeurs comme Meta, Alibaba et Google d\u00e9voilant de nouvelles it\u00e9rations de leurs mod\u00e8les de langage de grande taille (LLM)."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Google a rejoint la course avec Gemma 3n, offrant des fen\u00eatres contextuelles \u00e9largies et des capacit\u00e9s de fine-tuning am\u00e9lior\u00e9es."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/qwen3-8-27b-vs-claude-fable-5",
   "title": "Qwen3.8 27B (xhigh) vs Claude Fable 5 (Adaptive Reasoning, Max Effort, Opus 4.8 Fallback): Model Comparison | Artificial Analysis",
   "published_at": null,
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-08",
   "access_type": null,
   "license": null,
   "significance": "27B model with reasoning, image input support, 256k context window, and open source weights",
   "organization_evidence": [
    "Comparison between Qwen3.8 27B (xhigh) and Claude Fable 5 (Adaptive Reasoning, Max Effort, Opus 4.8 Fallback) across intelligence, price, speed, context window and more.",
    "| | ![Alibaba logo](/img/logos/alibaba_small.svg)Alibaba Qwen3.8 27B (xhigh) | ![Anthropic logo](/img/logos/anthropic_small.svg)Anthropic Claude Fable 5 (Adaptive Reasoning, Max Effort, Opus 4.8 Fallback) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## Model Comparison",
    "| Release Date | August 2026 | June 2026 | Qwen3.8 27B (xhigh) has a more recent release date than Claude Fable 5 (Adaptive Reasoning, Max Effort, Opus 4.8 Fallback) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## Model Comparison",
    "| | ![Alibaba logo](/img/logos/alibaba_small.svg)Alibaba Qwen3.8 27B (xhigh) | ![Anthropic logo](/img/logos/anthropic_small.svg)Anthropic Claude Fable 5 (Adaptive Reasoning, Max Effort, Opus 4.8 Fallback) | |",
    "| Intelligence Index | 52 | 62 | Claude Fable 5 (Adaptive Reasoning, Max Effort, Opus 4.8 Fallback) is more intelligent than Qwen3.8 27B (xhigh) |",
    "| Context Window | 256k tokens~384 A4 pages of size 12 Arial font | 1000k tokens~1,500 A4 pages of size 12 Arial font | Claude Fable 5 (Adaptive Reasoning, Max Effort, Opus 4.8 Fallback) has a larger context window than Qwen3.8 27B (xhigh) |",
    "| Reasoning | Yes | Yes | Both Qwen3.8 27B (xhigh) and Claude Fable 5 (Adaptive Reasoning, Max Effort, Opus 4.8 Fallback) have reasoning |",
    "| Image Input Support | Yes | Yes | Both Qwen3.8 27B (xhigh) and Claude Fable 5 (Adaptive Reasoning, Max Effort, Opus 4.8 Fallback) have image input support |",
    "| Open Source (Weights) | Yes | No | Qwen3.8 27B (xhigh) is open source while Claude Fable 5 (Adaptive Reasoning, Max Effort, Opus 4.8 Fallback) is proprietary |"
   ]
  },
  {
   "url": "https://simplemondays.com/technology/understanding-china-s-rapid-ai-deployment-four-frontier-open-models",
   "title": "Understanding China\u2019s Rapid AI Deployment: Four Frontier Open Models - Simple Mondays",
   "published_at": "2026-07-15T06:14:42",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor",
   "organization_evidence": [
    "Understanding China\u2019s Rapid AI Deployment: Four Frontier Open Models",
    "The production line \u2014 spring 2026",
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Understanding China\u2019s Rapid AI Deployment: Four Frontier Open Models",
    "The production line \u2014 spring 2026",
    "APR 24"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Understanding China\u2019s Rapid AI Deployment: Four Frontier Open Models",
    "The production line \u2014 spring 2026",
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor"
   ]
  },
  {
   "url": "https://simplemondays.com/technology/understanding-china-s-rapid-ai-deployment-four-frontier-open-models",
   "title": "Understanding China\u2019s Rapid AI Deployment: Four Frontier Open Models - Simple Mondays",
   "published_at": "2026-07-15T06:14:42",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "cheap 1M-token context, native multimodal, modified-MIT",
   "organization_evidence": [
    "Understanding China\u2019s Rapid AI Deployment: Four Frontier Open Models",
    "The production line \u2014 spring 2026",
    "**MiniMax M3**cheap 1M-token context, native multimodal, modified-MIT"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Understanding China\u2019s Rapid AI Deployment: Four Frontier Open Models",
    "The production line \u2014 spring 2026",
    "JUN 01"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Understanding China\u2019s Rapid AI Deployment: Four Frontier Open Models",
    "The production line \u2014 spring 2026",
    "**MiniMax M3**cheap 1M-token context, native multimodal, modified-MIT"
   ]
  },
  {
   "url": "https://simplemondays.com/technology/understanding-china-s-rapid-ai-deployment-four-frontier-open-models",
   "title": "Understanding China\u2019s Rapid AI Deployment: Four Frontier Open Models - Simple Mondays",
   "published_at": "2026-07-15T06:14:42",
   "organization": "Moonshot",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "agent-run specialist, ~30% fewer thinking tokens than K2.6",
   "organization_evidence": [
    "Understanding China\u2019s Rapid AI Deployment: Four Frontier Open Models",
    "The production line \u2014 spring 2026",
    "**Kimi K2.7-Code (Moonshot)**agent-run specialist, ~30% fewer thinking tokens than K2.6"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Understanding China\u2019s Rapid AI Deployment: Four Frontier Open Models",
    "The production line \u2014 spring 2026",
    "JUN 13"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Understanding China\u2019s Rapid AI Deployment: Four Frontier Open Models",
    "The production line \u2014 spring 2026",
    "**Kimi K2.7-Code (Moonshot)**agent-run specialist, ~30% fewer thinking tokens than K2.6"
   ]
  },
  {
   "url": "https://simplemondays.com/technology/understanding-china-s-rapid-ai-deployment-four-frontier-open-models",
   "title": "Understanding China\u2019s Rapid AI Deployment: Four Frontier Open Models - Simple Mondays",
   "published_at": "2026-07-15T06:14:42",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "753B MoE, MIT, top open-weight on Artificial Analysis index",
   "organization_evidence": [
    "Understanding China\u2019s Rapid AI Deployment: Four Frontier Open Models",
    "The production line \u2014 spring 2026",
    "**GLM-5.2 (Z.ai)**753B MoE, MIT, top open-weight on Artificial Analysis index"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Understanding China\u2019s Rapid AI Deployment: Four Frontier Open Models",
    "The production line \u2014 spring 2026",
    "JUN 13\u201316"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Understanding China\u2019s Rapid AI Deployment: Four Frontier Open Models",
    "The production line \u2014 spring 2026",
    "**GLM-5.2 (Z.ai)**753B MoE, MIT, top open-weight on Artificial Analysis index"
   ]
  },
  {
   "url": "https://particle.news/story/thinking-machines-releases-inkling-an-openweights-975b-multimodal-model",
   "title": "Thinking Machines Releases Inkling, an Open\u2011Weights 975B Multimodal Model",
   "published_at": "2026-07-15T00:00:00",
   "organization": "Thinking Machines",
   "model": "Inkling",
   "release_date": "2026-07-15",
   "access_type": "open-weights",
   "license": null,
   "significance": "a mixture-of-experts system with 975 billion total parameters and roughly 41 billion active parameters for any given task that processes text, images, audio and video, supports very long context windows, and offers a controllable 'thinking effort' setting",
   "organization_evidence": [
    "## Overview",
    "- [Thinking Machines](/entity/thinking-machines-lab-inc/g/11x0gy281q?ref=cluster:17882141) publicly released Inkling on Wednesday, July 15, 2026, as its first general\u2011purpose open\u2011weights model."
   ],
   "model_evidence": [
    "## Overview",
    "- [Thinking Machines](/entity/thinking-machines-lab-inc/g/11x0gy281q?ref=cluster:17882141) publicly released Inkling on Wednesday, July 15, 2026, as its first general\u2011purpose open\u2011weights model."
   ],
   "release_date_evidence": [
    "## Overview",
    "- [Thinking Machines](/entity/thinking-machines-lab-inc/g/11x0gy281q?ref=cluster:17882141) publicly released Inkling on Wednesday, July 15, 2026, as its first general\u2011purpose open\u2011weights model."
   ],
   "access_type_evidence": [
    "## Overview",
    "- [Thinking Machines](/entity/thinking-machines-lab-inc/g/11x0gy281q?ref=cluster:17882141) publicly released Inkling on Wednesday, July 15, 2026, as its first general\u2011purpose open\u2011weights model."
   ],
   "license_evidence": null,
   "significance_evidence": [
    "## Overview",
    "- Inkling is a mixture\u2011of\u2011experts system with 975 billion total parameters and roughly 41 billion active parameters for any given task.",
    "- The model processes text, images, audio and video, supports very long context windows (reports cite up to 1 million tokens), and offers a controllable \u2018thinking effort\u2019 setting to trade compute for performance."
   ]
  },
  {
   "url": "https://blog.ibvl.in/index.php/2026/02/25/a-dream-of-spring-for-open-weight-llms-10-architectures-from-jan-feb-2026",
   "title": "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
   "published_at": "2026-02-25T13:26:56",
   "organization": "Arcee AI",
   "model": null,
   "release_date": "2026-01-27",
   "access_type": null,
   "license": null,
   "significance": "A 400B parameter Mixture-of-Experts flagship model with 13B active parameters.",
   "organization_evidence": [
    "In this article, I will walk you through the ten main releases in chronological order, with a focus on the architecture similarities and differences:Arcee AI\u2019s Trinity Large (Jan 27, 2026)",
    "1. Arcee AI\u2019s Trinity Large: A New US-Based Start-Up Sharing Open-Weight Models"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "In this article, I will walk you through the ten main releases in chronological order, with a focus on the architecture similarities and differences:Arcee AI\u2019s Trinity Large (Jan 27, 2026)",
    "On January 27, Arcee AI (a company I hadn\u2019t had on my radar up to then) began releasing versions of their open-weight 400B Trinity Large LLMs on the model hub, along with two smaller variants:"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "1. Arcee AI\u2019s Trinity Large: A New US-Based Start-Up Sharing Open-Weight Models",
    "Their flagship large model is a 400B param Mixture-of-Experts (MoE) with 13B active parameters."
   ]
  },
  {
   "url": "https://blog.ibvl.in/index.php/2026/02/25/a-dream-of-spring-for-open-weight-llms-10-architectures-from-jan-feb-2026",
   "title": "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
   "published_at": "2026-02-25T13:26:56",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-01-27",
   "access_type": null,
   "license": null,
   "significance": "A 1-trillion-parameter multimodal model with vision support that set a new open-weight performance ceiling.",
   "organization_evidence": [
    "In this article, I will walk you through the ten main releases in chronological order, with a focus on the architecture similarities and differences:Arcee AI\u2019s Trinity Large (Jan 27, 2026)Moonshot AI\u2019s Kimi K2.5 (Jan 27, 2026)",
    "2. Moonshot AI\u2019s Kimi K2.5: A DeepSeek-Like Model at a 1-Trillion-Parameter Scale"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "In this article, I will walk you through the ten main releases in chronological order, with a focus on the architecture similarities and differences:Arcee AI\u2019s Trinity Large (Jan 27, 2026)Moonshot AI\u2019s Kimi K2.5 (Jan 27, 2026)",
    "While Arcee Trinity essentially matched the modeling performance of the older GLM-4.5 model, Kimi K2.5 is an open-weight model that set a new open-weight performance ceiling at the time of its release on Jan 27.\u200b"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "2. Moonshot AI\u2019s Kimi K2.5: A DeepSeek-Like Model at a 1-Trillion-Parameter Scale",
    "While Arcee Trinity essentially matched the modeling performance of the older GLM-4.5 model, Kimi K2.5 is an open-weight model that set a new open-weight performance ceiling at the time of its release on Jan 27.\u200b",
    "The good modeling performance is no surprise when compared to, e.g., Arcee Trinity or GLM-4.5 covered earlier, since (similar to its K2 predecessor), Kimi K2.5 is a 1-trillion-parameter model and thus 2.5x larger than Trinity and 2.8x larger than GLM-4.5.",
    "However, K2 was a pure text model, and Kimi K2.5 is now a multimodal model with vision support."
   ]
  },
  {
   "url": "https://blog.ibvl.in/index.php/2026/02/25/a-dream-of-spring-for-open-weight-llms-10-architectures-from-jan-feb-2026",
   "title": "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
   "published_at": "2026-02-25T13:26:56",
   "organization": "StepFun",
   "model": null,
   "release_date": "2026-02-01",
   "access_type": null,
   "license": null,
   "significance": "A 196B parameter model with fast tokens/sec throughput.",
   "organization_evidence": [
    "In this article, I will walk you through the ten main releases in chronological order, with a focus on the architecture similarities and differences:Arcee AI\u2019s Trinity Large (Jan 27, 2026)Moonshot AI\u2019s Kimi K2.5 (Jan 27, 2026)StepFun Step 3.5 Flash (Feb 1, 2026)",
    "3. StepFun\u2019s Step 3.5 Flash: Good Performance at Great Tokens/Sec Throughput"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "In this article, I will walk you through the ten main releases in chronological order, with a focus on the architecture similarities and differences:Arcee AI\u2019s Trinity Large (Jan 27, 2026)Moonshot AI\u2019s Kimi K2.5 (Jan 27, 2026)StepFun Step 3.5 Flash (Feb 1, 2026)"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "3. StepFun\u2019s Step 3.5 Flash: Good Performance at Great Tokens/Sec Throughput",
    "Step 3.5 Flash is a 196B parameter model that is more than 3x smaller than the recent DeepSeek V3.2 model (671B) while being slightly ahead in modeling performance benchmarks."
   ]
  },
  {
   "url": "https://blog.ibvl.in/index.php/2026/02/25/a-dream-of-spring-for-open-weight-llms-10-architectures-from-jan-feb-2026",
   "title": "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
   "published_at": "2026-02-25T13:26:56",
   "organization": "Qwen team",
   "model": null,
   "release_date": "2026-02-03",
   "access_type": null,
   "license": null,
   "significance": "An 80B attention-hybrid coding model with 3B active parameters outperforming much larger models.",
   "organization_evidence": [
    "In this article, I will walk you through the ten main releases in chronological order, with a focus on the architecture similarities and differences:Arcee AI\u2019s Trinity Large (Jan 27, 2026)Moonshot AI\u2019s Kimi K2.5 (Jan 27, 2026)StepFun Step 3.5 Flash (Feb 1, 2026)Qwen3-Coder-Next (Feb 3, 2026)",
    "4. Qwen3-Coder-Next: An Attention-Hybrid for Coding",
    "In early February 2026, the Qwen3 team shared the 80B Qwen3-Coder-Next model (3B parameters active), which made big headlines for outperforming much larger models like DeepSeek V3.2 (37B active) and Kimi K2.5 and GLM-4.7 (both 32B active) on coding tasks."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "In this article, I will walk you through the ten main releases in chronological order, with a focus on the architecture similarities and differences:Arcee AI\u2019s Trinity Large (Jan 27, 2026)Moonshot AI\u2019s Kimi K2.5 (Jan 27, 2026)StepFun Step 3.5 Flash (Feb 1, 2026)Qwen3-Coder-Next (Feb 3, 2026)",
    "In early February 2026, the Qwen3 team shared the 80B Qwen3-Coder-Next model (3B parameters active), which made big headlines for outperforming much larger models like DeepSeek V3.2 (37B active) and Kimi K2.5 and GLM-4.7 (both 32B active) on coding tasks."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "4. Qwen3-Coder-Next: An Attention-Hybrid for Coding",
    "In early February 2026, the Qwen3 team shared the 80B Qwen3-Coder-Next model (3B parameters active), which made big headlines for outperforming much larger models like DeepSeek V3.2 (37B active) and Kimi K2.5 and GLM-4.7 (both 32B active) on coding tasks."
   ]
  },
  {
   "url": "https://blog.ibvl.in/index.php/2026/02/25/a-dream-of-spring-for-open-weight-llms-10-architectures-from-jan-feb-2026",
   "title": "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
   "published_at": "2026-02-25T13:26:56",
   "organization": "z.AI",
   "model": null,
   "release_date": "2026-02-12",
   "access_type": null,
   "license": null,
   "significance": "A 744B parameter Mixture-of-Experts flagship model on par with major proprietary offerings.",
   "organization_evidence": [
    "In this article, I will walk you through the ten main releases in chronological order, with a focus on the architecture similarities and differences:Arcee AI\u2019s Trinity Large (Jan 27, 2026)Moonshot AI\u2019s Kimi K2.5 (Jan 27, 2026)StepFun Step 3.5 Flash (Feb 1, 2026)Qwen3-Coder-Next (Feb 3, 2026)z.AI\u2019s GLM-5 (Feb 12, 2026)",
    "5. z.AI\u2019s GLM-5: A New Flagship Open-Weight Model"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "In this article, I will walk you through the ten main releases in chronological order, with a focus on the architecture similarities and differences:Arcee AI\u2019s Trinity Large (Jan 27, 2026)Moonshot AI\u2019s Kimi K2.5 (Jan 27, 2026)StepFun Step 3.5 Flash (Feb 1, 2026)Qwen3-Coder-Next (Feb 3, 2026)z.AI\u2019s GLM-5 (Feb 12, 2026)",
    "The GLM-5 release on February 12th was a big deal, because at the time of its release it appeared to be on par with the major flagship LLM offerings, including GPT-5.2 extra-high, Gemini Pro 3, and Claude 4.6 Opus."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "5. z.AI\u2019s GLM-5: A New Flagship Open-Weight Model",
    "The GLM-5 release on February 12th was a big deal, because at the time of its release it appeared to be on par with the major flagship LLM offerings, including GPT-5.2 extra-high, Gemini Pro 3, and Claude 4.6 Opus.",
    "GLM-5 shows a major modeling performance improvement based on the benchmark shown in Figure 18 above.",
    "Like GLM-4.7, all the other models discussed so far, GLM-5 is a Mixture-of-Experts model."
   ]
  },
  {
   "url": "https://blog.ibvl.in/index.php/2026/02/25/a-dream-of-spring-for-open-weight-llms-10-architectures-from-jan-feb-2026",
   "title": "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
   "published_at": "2026-02-25T13:26:56",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-02-12",
   "access_type": null,
   "license": null,
   "significance": "A 230B parameter strong coding model with high popularity.",
   "organization_evidence": [
    "In this article, I will walk you through the ten main releases in chronological order, with a focus on the architecture similarities and differences:Arcee AI\u2019s Trinity Large (Jan 27, 2026)Moonshot AI\u2019s Kimi K2.5 (Jan 27, 2026)StepFun Step 3.5 Flash (Feb 1, 2026)Qwen3-Coder-Next (Feb 3, 2026)z.AI\u2019s GLM-5 (Feb 12, 2026)MiniMax M2.5 (Feb 12, 2026)",
    "6. MiniMax M2.5: A Strong Coder with \u201cOnly\u201d 230B Parameters"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "In this article, I will walk you through the ten main releases in chronological order, with a focus on the architecture similarities and differences:Arcee AI\u2019s Trinity Large (Jan 27, 2026)Moonshot AI\u2019s Kimi K2.5 (Jan 27, 2026)StepFun Step 3.5 Flash (Feb 1, 2026)Qwen3-Coder-Next (Feb 3, 2026)z.AI\u2019s GLM-5 (Feb 12, 2026)MiniMax M2.5 (Feb 12, 2026)",
    "The aforementioned GLM-5 and Kimi K2.5 are popular open-weight models, but according to OpenRouter statistics, they pale in comparison to MiniMax M2.5, which was released on February 12 as well."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "6. MiniMax M2.5: A Strong Coder with \u201cOnly\u201d 230B Parameters",
    "Architecture-wise, MiniMax M2.5 is a 230B model with a fairly classic design: just plain Grouped Query Attention, no sliding window attention or other efficiency improvements."
   ]
  },
  {
   "url": "https://blog.ibvl.in/index.php/2026/02/25/a-dream-of-spring-for-open-weight-llms-10-architectures-from-jan-feb-2026",
   "title": "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
   "published_at": "2026-02-25T13:26:56",
   "organization": "Nanbeige",
   "model": null,
   "release_date": "2026-02-13",
   "access_type": null,
   "license": null,
   "significance": "A strong small LLM successor targeting on-device use cases.",
   "organization_evidence": [
    "In this article, I will walk you through the ten main releases in chronological order, with a focus on the architecture similarities and differences:Arcee AI\u2019s Trinity Large (Jan 27, 2026)Moonshot AI\u2019s Kimi K2.5 (Jan 27, 2026)StepFun Step 3.5 Flash (Feb 1, 2026)Qwen3-Coder-Next (Feb 3, 2026)z.AI\u2019s GLM-5 (Feb 12, 2026)MiniMax M2.5 (Feb 12, 2026)Nanbeige 4.1 3B (Feb 13, 2026)",
    "7. Nanbeige 4.1 3B: A Strong Llama 3 Successor"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "In this article, I will walk you through the ten main releases in chronological order, with a focus on the architecture similarities and differences:Arcee AI\u2019s Trinity Large (Jan 27, 2026)Moonshot AI\u2019s Kimi K2.5 (Jan 27, 2026)StepFun Step 3.5 Flash (Feb 1, 2026)Qwen3-Coder-Next (Feb 3, 2026)z.AI\u2019s GLM-5 (Feb 12, 2026)MiniMax M2.5 (Feb 12, 2026)Nanbeige 4.1 3B (Feb 13, 2026)"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "7. Nanbeige 4.1 3B: A Strong Llama 3 Successor",
    "Why I am mentioning all this is that Nanbeige 4.1 3B seems to target the \u201csmall\u201d LLM on-device use case that Qwen3 is so popular for."
   ]
  },
  {
   "url": "https://blog.ibvl.in/index.php/2026/02/25/a-dream-of-spring-for-open-weight-llms-10-architectures-from-jan-feb-2026",
   "title": "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
   "published_at": "2026-02-25T13:26:56",
   "organization": "Qwen team",
   "model": null,
   "release_date": "2026-02-15",
   "access_type": null,
   "license": null,
   "significance": "A 397B parameter Mixture-of-Experts model featuring hybrid attention and native multimodal support.",
   "organization_evidence": [
    "In this article, I will walk you through the ten main releases in chronological order, with a focus on the architecture similarities and differences:Arcee AI\u2019s Trinity Large (Jan 27, 2026)Moonshot AI\u2019s Kimi K2.5 (Jan 27, 2026)StepFun Step 3.5 Flash (Feb 1, 2026)Qwen3-Coder-Next (Feb 3, 2026)z.AI\u2019s GLM-5 (Feb 12, 2026)MiniMax M2.5 (Feb 12, 2026)Nanbeige 4.1 3B (Feb 13, 2026)Qwen 3.5 (Feb 15, 2026)",
    "8. Qwen3.5 and the Continutation of Hybrid Attention",
    "However, the Qwen team just released a new Qwen3.5 model variant on February 15.\u200bQwen3.5"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "In this article, I will walk you through the ten main releases in chronological order, with a focus on the architecture similarities and differences:Arcee AI\u2019s Trinity Large (Jan 27, 2026)Moonshot AI\u2019s Kimi K2.5 (Jan 27, 2026)StepFun Step 3.5 Flash (Feb 1, 2026)Qwen3-Coder-Next (Feb 3, 2026)z.AI\u2019s GLM-5 (Feb 12, 2026)MiniMax M2.5 (Feb 12, 2026)Nanbeige 4.1 3B (Feb 13, 2026)Qwen 3.5 (Feb 15, 2026)",
    "However, the Qwen team just released a new Qwen3.5 model variant on February 15.\u200bQwen3.5"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "8. Qwen3.5 and the Continutation of Hybrid Attention",
    "Qwen3.5 397B-A17B, a Mixture-of-Experts (MoE) with 397B parameters (17B active per token), is a step up from the largest Qwen3 model, which is 235B parameters in size.",
    "Architecture-wise, Qwen3.5 adopts the hybrid attention model (featuring Gated DeltaNet) that Qwen3-Next and Qwen3-Coder-Next (section 4) used.",
    "Besides scaling up the model size, as shown in the figure above, Qwen3.5 now also includes multimodal support (previously, it was only available in separate Qwen3-VL models)."
   ]
  },
  {
   "url": "https://blog.ibvl.in/index.php/2026/02/25/a-dream-of-spring-for-open-weight-llms-10-architectures-from-jan-feb-2026",
   "title": "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
   "published_at": "2026-02-25T13:26:56",
   "organization": "Ant Group",
   "model": null,
   "release_date": "2026-02-16",
   "access_type": null,
   "license": null,
   "significance": "A 1-trillion-parameter LLM with hybrid attention offering high efficiency in long contexts.",
   "organization_evidence": [
    "In this article, I will walk you through the ten main releases in chronological order, with a focus on the architecture similarities and differences:Arcee AI\u2019s Trinity Large (Jan 27, 2026)Moonshot AI\u2019s Kimi K2.5 (Jan 27, 2026)StepFun Step 3.5 Flash (Feb 1, 2026)Qwen3-Coder-Next (Feb 3, 2026)z.AI\u2019s GLM-5 (Feb 12, 2026)MiniMax M2.5 (Feb 12, 2026)Nanbeige 4.1 3B (Feb 13, 2026)Qwen 3.5 (Feb 15, 2026)Ant Group\u2019s Ling 2.5 1T & Ring 2.5 1T (Feb 16, 2026)",
    "9. Ant Group\u2019s Ling 2.5 1T with Lightning Attention"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "In this article, I will walk you through the ten main releases in chronological order, with a focus on the architecture similarities and differences:Arcee AI\u2019s Trinity Large (Jan 27, 2026)Moonshot AI\u2019s Kimi K2.5 (Jan 27, 2026)StepFun Step 3.5 Flash (Feb 1, 2026)Qwen3-Coder-Next (Feb 3, 2026)z.AI\u2019s GLM-5 (Feb 12, 2026)MiniMax M2.5 (Feb 12, 2026)Nanbeige 4.1 3B (Feb 13, 2026)Qwen 3.5 (Feb 15, 2026)Ant Group\u2019s Ling 2.5 1T & Ring 2.5 1T (Feb 16, 2026)"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "9. Ant Group\u2019s Ling 2.5 1T with Lightning Attention",
    "Ling 2.5 (and the reasoning variant Ring 2.5) are 1-trillion-parameter LLMs with a hybrid attention architecture in a similar spirit to Qwen3.5 and Qwen3-Next.",
    "Ling 2.5 is not the strongest model in terms of absolute benchmark performance, but its selling point is very good efficiency in long contexts (due to the hybrid attention)."
   ]
  },
  {
   "url": "https://blog.ibvl.in/index.php/2026/02/25/a-dream-of-spring-for-open-weight-llms-10-architectures-from-jan-feb-2026",
   "title": "A Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026",
   "published_at": "2026-02-25T13:26:56",
   "organization": "Cohere",
   "model": null,
   "release_date": "2026-02-17",
   "access_type": null,
   "license": null,
   "significance": "A 3.35B parameter multilingual small LLM with parallel transformer blocks.",
   "organization_evidence": [
    "In this article, I will walk you through the ten main releases in chronological order, with a focus on the architecture similarities and differences:Arcee AI\u2019s Trinity Large (Jan 27, 2026)Moonshot AI\u2019s Kimi K2.5 (Jan 27, 2026)StepFun Step 3.5 Flash (Feb 1, 2026)Qwen3-Coder-Next (Feb 3, 2026)z.AI\u2019s GLM-5 (Feb 12, 2026)MiniMax M2.5 (Feb 12, 2026)Nanbeige 4.1 3B (Feb 13, 2026)Qwen 3.5 (Feb 15, 2026)Ant Group\u2019s Ling 2.5 1T & Ring 2.5 1T (Feb 16, 2026)Cohere\u2019s Tiny Aya (Feb 17, 2026)",
    "10. Tiny Aya: A 3.35B Model with Strong Multilingual Support"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "In this article, I will walk you through the ten main releases in chronological order, with a focus on the architecture similarities and differences:Arcee AI\u2019s Trinity Large (Jan 27, 2026)Moonshot AI\u2019s Kimi K2.5 (Jan 27, 2026)StepFun Step 3.5 Flash (Feb 1, 2026)Qwen3-Coder-Next (Feb 3, 2026)z.AI\u2019s GLM-5 (Feb 12, 2026)MiniMax M2.5 (Feb 12, 2026)Nanbeige 4.1 3B (Feb 13, 2026)Qwen 3.5 (Feb 15, 2026)Ant Group\u2019s Ling 2.5 1T & Ring 2.5 1T (Feb 16, 2026)Cohere\u2019s Tiny Aya (Feb 17, 2026)",
    "Released on February 17, Tiny Aya is a new, \u201csmall\u201d LLM by Cohere that is said to be the \u201cmost capable multilingual open-weight model\u201d at the 3B parameter size class."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "10. Tiny Aya: A 3.35B Model with Strong Multilingual Support",
    "Released on February 17, Tiny Aya is a new, \u201csmall\u201d LLM by Cohere that is said to be the \u201cmost capable multilingual open-weight model\u201d at the 3B parameter size class.",
    "That aside, Aya is a 3.35B parameter model that comes in several flavors that are useful forpersonal and (non-commercial) research use: tiny-aya-base (base model) tiny-aya-global (best balance across languages and regions) tiny-aya-fire (optimized for South Asian languages)tiny-aya-water (optimized for European and Asia Pacific languages)tiny-aya-earth (optimized for West Asian and African languages)"
   ]
  },
  {
   "url": "https://ventureatlas.org/product/moonshot-ai-kimi-k-series-open-weight-models-k2-k2-5-k2-6-k3",
   "title": "Kimi K-series open-weight models (K2 / K2.5 / K2.6 / K3) - Moonshot AI",
   "published_at": null,
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-16",
   "access_type": null,
   "license": null,
   "significance": "~2.8T-parameter MoE with 1M-token context and Kimi Delta Attention, the largest open-weight model ever released",
   "organization_evidence": [
    "# Kimi K-series open-weight models (K2 / K2.5 / K2.6 / K3)",
    "Generally Available[Moonshot AI](/company/moonshot-ai)",
    "Moonshot's open-weight frontier line under a modified MIT license: K2 (July 2025, 1T MoE), K2 Thinking (November 2025 agentic-benchmark leader), K2.5 (vision via MoonViT), K2.6, K2.7-Code, and K3 (July 16, 2026) - a ~2.8T-parameter MoE with 1M-token context and Kimi Delta Attention, the largest open-weight model ever released."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Kimi K-series open-weight models (K2 / K2.5 / K2.6 / K3)",
    "Moonshot's open-weight frontier line under a modified MIT license: K2 (July 2025, 1T MoE), K2 Thinking (November 2025 agentic-benchmark leader), K2.5 (vision via MoonViT), K2.6, K2.7-Code, and K3 (July 16, 2026) - a ~2.8T-parameter MoE with 1M-token context and Kimi Delta Attention, the largest open-weight model ever released."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Kimi K-series open-weight models (K2 / K2.5 / K2.6 / K3)",
    "Moonshot's open-weight frontier line under a modified MIT license: K2 (July 2025, 1T MoE), K2 Thinking (November 2025 agentic-benchmark leader), K2.5 (vision via MoonViT), K2.6, K2.7-Code, and K3 (July 16, 2026) - a ~2.8T-parameter MoE with 1M-token context and Kimi Delta Attention, the largest open-weight model ever released."
   ]
  },
  {
   "url": "https://news-area.com/2026/06/30/meituan-open-sources-longcat-2-0-the-1-6t-near-frontier-agentic-coding-model-thats-been-leading-openrouter-trained-entirely-on-chinese-chips",
   "title": "Meituan open sources LongCat-2.0, the 1.6T, near-frontier agentic coding model that\u2019s been leading OpenRouter \u2014 trained entirely on Chinese chips",
   "published_at": "2026-06-30T03:39:00",
   "organization": "Meituan",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "1.6-trillion-parameter Mixture-of-Experts agentic coding model with a native 1-million-token context window that leads OpenRouter and ranks high on software engineering benchmarks",
   "organization_evidence": [
    "# Meituan open sources LongCat-2.0, the 1.6T, near-frontier agentic coding model that\u2019s been leading OpenRouter \u2014 trained entirely on Chinese chips",
    "A few hours ago, Chinese delivery app company Meituan officially unveiled LongCat-2.0 on GitHub , Hugging Face , and its native platform, unmasking the model as the computational engine behind \"Owl Alpha,\" the anonymous stealth model that has spent the last two months commanding global developer charts on OpenRouter."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Meituan open sources LongCat-2.0, the 1.6T, near-frontier agentic coding model that\u2019s been leading OpenRouter \u2014 trained entirely on Chinese chips",
    "Developed to fundamentally disrupt closed-source enterprise dominance in autonomous software engineering, the 1.6-trillion-parameter Mixture-of-Experts (MoE) system brings a native 1-million-token context window to the public domain under a highly permissive, enterprise grade, commercially viable MIT license."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/claude-opus-5-xhigh-vs-kimi-k3-low",
   "title": "Claude Opus 5 (Adaptive Reasoning, Xhigh Effort) vs Kimi K3 (low): Model Comparison",
   "published_at": null,
   "organization": "Kimi",
   "model": "Kimi K3 (low)",
   "release_date": "2026-07",
   "access_type": "open source",
   "license": null,
   "significance": "Large context window of 1049k tokens with open weights",
   "organization_evidence": [
    "# Claude Opus 5 (Adaptive Reasoning, Xhigh Effort) vs. Kimi K3 (low)",
    "## Model Comparison",
    "| | ![Anthropic logo](/img/logos/anthropic_small.svg)Anthropic Claude Opus 5 (Adaptive Reasoning, Xhigh Effort) | ![Kimi logo](/_next/image?url=%2Fimg%2Flogos%2Fkimi.jpg&w=32&q=75)Kimi Kimi K3 (low) | |"
   ],
   "model_evidence": [
    "# Claude Opus 5 (Adaptive Reasoning, Xhigh Effort) vs. Kimi K3 (low)",
    "## Model Comparison",
    "| | ![Anthropic logo](/img/logos/anthropic_small.svg)Anthropic Claude Opus 5 (Adaptive Reasoning, Xhigh Effort) | ![Kimi logo](/_next/image?url=%2Fimg%2Flogos%2Fkimi.jpg&w=32&q=75)Kimi Kimi K3 (low) | |",
    "| Intelligence Index | 63 | 48 | Claude Opus 5 (Adaptive Reasoning, Xhigh Effort) is more intelligent than Kimi K3 (low) |"
   ],
   "release_date_evidence": [
    "# Claude Opus 5 (Adaptive Reasoning, Xhigh Effort) vs. Kimi K3 (low)",
    "## Model Comparison",
    "| | ![Anthropic logo](/img/logos/anthropic_small.svg)Anthropic Claude Opus 5 (Adaptive Reasoning, Xhigh Effort) | ![Kimi logo](/_next/image?url=%2Fimg%2Flogos%2Fkimi.jpg&w=32&q=75)Kimi Kimi K3 (low) | |",
    "| Release Date | July 2026 | July 2026 | Claude Opus 5 (Adaptive Reasoning, Xhigh Effort) has a more recent release date than Kimi K3 (low) |"
   ],
   "access_type_evidence": [
    "# Claude Opus 5 (Adaptive Reasoning, Xhigh Effort) vs. Kimi K3 (low)",
    "## Model Comparison",
    "| | ![Anthropic logo](/img/logos/anthropic_small.svg)Anthropic Claude Opus 5 (Adaptive Reasoning, Xhigh Effort) | ![Kimi logo](/_next/image?url=%2Fimg%2Flogos%2Fkimi.jpg&w=32&q=75)Kimi Kimi K3 (low) | |",
    "| Open Source (Weights) | No | Yes | Kimi K3 (low) is open source while Claude Opus 5 (Adaptive Reasoning, Xhigh Effort) is proprietary |"
   ],
   "license_evidence": null,
   "significance_evidence": [
    "# Claude Opus 5 (Adaptive Reasoning, Xhigh Effort) vs. Kimi K3 (low)",
    "## Model Comparison",
    "| | ![Anthropic logo](/img/logos/anthropic_small.svg)Anthropic Claude Opus 5 (Adaptive Reasoning, Xhigh Effort) | ![Kimi logo](/_next/image?url=%2Fimg%2Flogos%2Fkimi.jpg&w=32&q=75)Kimi Kimi K3 (low) | |",
    "| Context Window | 1000k tokens~1,500 A4 pages of size 12 Arial font | 1049k tokens~1,573 A4 pages of size 12 Arial font | Kimi K3 (low) has a larger context window than Claude Opus 5 (Adaptive Reasoning, Xhigh Effort) |",
    "| Open Source (Weights) | No | Yes | Kimi K3 (low) is open source while Claude Opus 5 (Adaptive Reasoning, Xhigh Effort) is proprietary |"
   ]
  },
  {
   "url": "https://elfysworld.com/general/the-speed-of-innovation-four-frontier-ai-models-in-eight-weeks",
   "title": "The Speed Of Innovation: Four Frontier AI Models In Eight Weeks - ELFY'S WORLD",
   "published_at": "2026-07-15T06:21:57",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor",
   "organization_evidence": [
    "Over eight weeks in 2026, Chinese labs released four frontier-class open-weight AI models, marking a notable increase in AI development activity.",
    "These releases include DeepSeek V4 on April 24, MiniMax M3 on June 1, and Kimi K2.7-Code along with GLM-5.2 in mid-June."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "These releases include DeepSeek V4 on April 24, MiniMax M3 on June 1, and Kimi K2.7-Code along with GLM-5.2 in mid-June."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor"
   ]
  },
  {
   "url": "https://elfysworld.com/general/the-speed-of-innovation-four-frontier-ai-models-in-eight-weeks",
   "title": "The Speed Of Innovation: Four Frontier AI Models In Eight Weeks - ELFY'S WORLD",
   "published_at": "2026-07-15T06:21:57",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "cheap 1M-token context, native multimodal, modified-MIT",
   "organization_evidence": [
    "Over eight weeks in 2026, Chinese labs released four frontier-class open-weight AI models, marking a notable increase in AI development activity.",
    "These releases include DeepSeek V4 on April 24, MiniMax M3 on June 1, and Kimi K2.7-Code along with GLM-5.2 in mid-June."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "These releases include DeepSeek V4 on April 24, MiniMax M3 on June 1, and Kimi K2.7-Code along with GLM-5.2 in mid-June."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "**MiniMax M3**cheap 1M-token context, native multimodal, modified-MIT"
   ]
  },
  {
   "url": "https://elfysworld.com/general/the-speed-of-innovation-four-frontier-ai-models-in-eight-weeks",
   "title": "The Speed Of Innovation: Four Frontier AI Models In Eight Weeks - ELFY'S WORLD",
   "published_at": "2026-07-15T06:21:57",
   "organization": "Moonshot",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "agent-run specialist, ~30% fewer thinking tokens than K2.6",
   "organization_evidence": [
    "Over eight weeks in 2026, Chinese labs released four frontier-class open-weight AI models, marking a notable increase in AI development activity.",
    "JUN 13\n\n**Kimi K2.7-Code (Moonshot)**agent-run specialist, ~30% fewer thinking tokens than K2.6"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "These releases include DeepSeek V4 on April 24, MiniMax M3 on June 1, and Kimi K2.7-Code along with GLM-5.2 in mid-June.",
    "JUN 13\n\n**Kimi K2.7-Code (Moonshot)**agent-run specialist, ~30% fewer thinking tokens than K2.6"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "JUN 13\n\n**Kimi K2.7-Code (Moonshot)**agent-run specialist, ~30% fewer thinking tokens than K2.6"
   ]
  },
  {
   "url": "https://elfysworld.com/general/the-speed-of-innovation-four-frontier-ai-models-in-eight-weeks",
   "title": "The Speed Of Innovation: Four Frontier AI Models In Eight Weeks - ELFY'S WORLD",
   "published_at": "2026-07-15T06:21:57",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "753B MoE, MIT, top open-weight on Artificial Analysis index",
   "organization_evidence": [
    "Over eight weeks in 2026, Chinese labs released four frontier-class open-weight AI models, marking a notable increase in AI development activity.",
    "JUN 13\u201316\n\n**GLM-5.2 (Z.ai)**753B MoE, MIT, top open-weight on Artificial Analysis index"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "These releases include DeepSeek V4 on April 24, MiniMax M3 on June 1, and Kimi K2.7-Code along with GLM-5.2 in mid-June.",
    "JUN 13\u201316\n\n**GLM-5.2 (Z.ai)**753B MoE, MIT, top open-weight on Artificial Analysis index"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "JUN 13\u201316\n\n**GLM-5.2 (Z.ai)**753B MoE, MIT, top open-weight on Artificial Analysis index"
   ]
  },
  {
   "url": "https://wan27.org/blog/inkling-thinking-machines-explained",
   "title": "What Is Inkling? Thinking Machines Lab's First Open Model Explained (2026) | Wan 2.7",
   "published_at": "2026-07-15T00:00:00",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "A 975B open-weight MoE model (41B active) and multimodal reasoning engine that natively reads text, images, video frames, and audio.",
   "organization_evidence": [
    "What Is Inkling? Thinking Machines Lab's First Open Model Explained (2026)",
    "On July 15, 2026, Thinking Machines Lab \u2014 the roughly 200-person startup that ex-OpenAI CTO Mira Murati has kept in near-total stealth for over a year \u2014 dropped **Inkling**, a 975-billion-parameter open-weight model, with the full weights on Hugging Face the same afternoon."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "What Is Inkling? Thinking Machines Lab's First Open Model Explained (2026)",
    "On July 15, 2026, Thinking Machines Lab \u2014 the roughly 200-person startup that ex-OpenAI CTO Mira Murati has kept in near-total stealth for over a year \u2014 dropped **Inkling**, a 975-billion-parameter open-weight model, with the full weights on Hugging Face the same afternoon."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "What Is Inkling? Thinking Machines Lab's First Open Model Explained (2026)",
    "Inkling is a 975B open-weight MoE model (41B active) from Thinking Machines Lab."
   ]
  },
  {
   "url": "https://fireworks.ai/blog/best-open-source-llms-may-2026?from_theconsensus=1",
   "title": "Best Open Source LLMs of May 2026: We Reviewed 7 Models",
   "published_at": "2026-05-19T00:00:00",
   "organization": "MiniMax AI",
   "model": null,
   "release_date": "2026-02-12",
   "access_type": null,
   "license": null,
   "significance": "Real-world productivity and software engineering workflows with 80.2% on SWE-Bench Verified",
   "organization_evidence": [
    "This roundup compares the top open source LLMs available today, breaking down their trade-offs and ideal applications so you can make an informed decision without running your own eval suite.",
    "## The best open source LLMs at a glance",
    "| MiniMax-M2.5 | February 12, 2026 | 229B total (10B active) | 196,608 tokens | Real-world productivity and software engineering workflows | [Try in playground](https://app.fireworks.ai/playground?model=accounts/fireworks/models/minimax-m2p5&_gl=1*145r14b*_gcl_au*MTY5MjA2MTk2OC4xNzcwMjYzODcy) |",
    "## MiniMax-M2.5",
    "\u2022**Corporate Sponsor:** MiniMax AI"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "This roundup compares the top open source LLMs available today, breaking down their trade-offs and ideal applications so you can make an informed decision without running your own eval suite.",
    "## The best open source LLMs at a glance",
    "| MiniMax-M2.5 | February 12, 2026 | 229B total (10B active) | 196,608 tokens | Real-world productivity and software engineering workflows | [Try in playground](https://app.fireworks.ai/playground?model=accounts/fireworks/models/minimax-m2p5&_gl=1*145r14b*_gcl_au*MTY5MjA2MTk2OC4xNzcwMjYzODcy) |",
    "## MiniMax-M2.5",
    "\u2022**Release Date:** February 12, 2026"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "This roundup compares the top open source LLMs available today, breaking down their trade-offs and ideal applications so you can make an informed decision without running your own eval suite.",
    "## The best open source LLMs at a glance",
    "| MiniMax-M2.5 | February 12, 2026 | 229B total (10B active) | 196,608 tokens | Real-world productivity and software engineering workflows | [Try in playground](https://app.fireworks.ai/playground?model=accounts/fireworks/models/minimax-m2p5&_gl=1*145r14b*_gcl_au*MTY5MjA2MTk2OC4xNzcwMjYzODcy) |",
    "## MiniMax-M2.5",
    "MiniMax-M2.5 is an agent-native MoE model trained with reinforcement learning across hundreds of thousands of complex real-world environments."
   ]
  },
  {
   "url": "https://fireworks.ai/blog/best-open-source-llms-may-2026?from_theconsensus=1",
   "title": "Best Open Source LLMs of May 2026: We Reviewed 7 Models",
   "published_at": "2026-05-19T00:00:00",
   "organization": "Zhipu AI",
   "model": null,
   "release_date": "2026-02-12",
   "access_type": null,
   "license": null,
   "significance": "Complex systems engineering and long-horizon agentic tasks with up to 128K output tokens per inference pass",
   "organization_evidence": [
    "This roundup compares the top open source LLMs available today, breaking down their trade-offs and ideal applications so you can make an informed decision without running your own eval suite.",
    "## The best open source LLMs at a glance",
    "| GLM-5 | February 12, 2026 | 744B total (40B active) | 200K input tokens | Complex systems engineering and long-horizon agentic tasks | [Try in playground](https://app.fireworks.ai/playground?model=accounts/fireworks/models/glm-5) |",
    "## GLM-5"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "This roundup compares the top open source LLMs available today, breaking down their trade-offs and ideal applications so you can make an informed decision without running your own eval suite.",
    "## The best open source LLMs at a glance",
    "| GLM-5 | February 12, 2026 | 744B total (40B active) | 200K input tokens | Complex systems engineering and long-horizon agentic tasks | [Try in playground](https://app.fireworks.ai/playground?model=accounts/fireworks/models/glm-5) |",
    "## GLM-5",
    "\u2022**Release Date:** February 12, 2026"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "This roundup compares the top open source LLMs available today, breaking down their trade-offs and ideal applications so you can make an informed decision without running your own eval suite.",
    "## The best open source LLMs at a glance",
    "| GLM-5 | February 12, 2026 | 744B total (40B active) | 200K input tokens | Complex systems engineering and long-horizon agentic tasks | [Try in playground](https://app.fireworks.ai/playground?model=accounts/fireworks/models/glm-5) |",
    "## GLM-5",
    "GLM-5 is Zhipu AI's largest foundation model, built for complex systems engineering and long-horizon agentic workflows."
   ]
  },
  {
   "url": "https://fireworks.ai/blog/best-open-source-llms-may-2026?from_theconsensus=1",
   "title": "Best Open Source LLMs of May 2026: We Reviewed 7 Models",
   "published_at": "2026-05-19T00:00:00",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-01-27",
   "access_type": null,
   "license": null,
   "significance": "Front-end visual coding and distributed market research with native multimodality",
   "organization_evidence": [
    "This roundup compares the top open source LLMs available today, breaking down their trade-offs and ideal applications so you can make an informed decision without running your own eval suite.",
    "## The best open source LLMs at a glance",
    "| Kimi K2.5 | January 27, 2026 | 1 Trillion total (32B active) | 256K tokens | Front-end visual coding and distributed market research | [Try in playground](https://app.fireworks.ai/playground?model=accounts/fireworks/models/kimi-k2p5) |",
    "## Kimi K2.5",
    "\u2022**Corporate Sponsor:** Moonshot AI"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "This roundup compares the top open source LLMs available today, breaking down their trade-offs and ideal applications so you can make an informed decision without running your own eval suite.",
    "## The best open source LLMs at a glance",
    "| Kimi K2.5 | January 27, 2026 | 1 Trillion total (32B active) | 256K tokens | Front-end visual coding and distributed market research | [Try in playground](https://app.fireworks.ai/playground?model=accounts/fireworks/models/kimi-k2p5) |",
    "## Kimi K2.5",
    "\u2022**Release Date:** January 27, 2026"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "This roundup compares the top open source LLMs available today, breaking down their trade-offs and ideal applications so you can make an informed decision without running your own eval suite.",
    "## The best open source LLMs at a glance",
    "| Kimi K2.5 | January 27, 2026 | 1 Trillion total (32B active) | 256K tokens | Front-end visual coding and distributed market research | [Try in playground](https://app.fireworks.ai/playground?model=accounts/fireworks/models/kimi-k2p5) |",
    "## Kimi K2.5",
    "Kimi K2.5 is the multimodal successor to K2 Thinking, adding native vision through a 400M-parameter MoonViT encoder jointly pre-trained on 15 trillion mixed tokens."
   ]
  },
  {
   "url": "https://singularitymoments.com/content/mira-muratis-thinking-machines-drops-inkling-a-975b-parameter-open-model",
   "title": "Mira Murati\u2019s Thinking Machines drops Inkling, a 975B-parameter open model",
   "published_at": "2026-07-15T19:47:51",
   "organization": "Thinking Machines",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "a 975B-parameter open multimodal model built on a sparse Mixture-of-Experts architecture containing 256 routed experts",
   "organization_evidence": [
    "INTERVIEW RECAP",
    "# Mira Murati\u2019s Thinking Machines drops Inkling, a 975B-parameter open model",
    "Thinking Machines Lab, the high-profile artificial intelligence startup founded by former OpenAI executives, has officially launched its first multimodal model, Inkling."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "INTERVIEW RECAP",
    "# Mira Murati\u2019s Thinking Machines drops Inkling, a 975B-parameter open model",
    "** Jul 15, 2026 ** Sara Quinn"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "INTERVIEW RECAP",
    "# Mira Murati\u2019s Thinking Machines drops Inkling, a 975B-parameter open model",
    "Thinking Machines Lab, the high-profile artificial intelligence startup founded by former OpenAI executives, has officially launched its first multimodal model, Inkling.",
    "Breaking a year and a half of silent development, the company has released the weights of its new 975-billion-parameter system under an open-source model.",
    "Inkling is built on a sparse Mixture-of-Experts architecture containing 256 routed experts."
   ]
  },
  {
   "url": "https://techieus.com/general/signal-s-four-frontier-models-a-testament-to-china-s-speed-in-ai-development",
   "title": "Signal\u2019s Four Frontier Models: A Testament To China\u2019s Speed In AI Development - TechieUS",
   "published_at": "2026-07-15T06:16:58",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor",
   "organization_evidence": [
    "Signal\u2019s Four Frontier Models: A Testament To China\u2019s Speed In AI Development",
    "The Chinese open-weight ecosystem has expanded from a single lab two years ago to four major players: DeepSeek, Z.ai, Moonshot, and Alibaba, each with distinct strategic focuses.",
    "### The production line \u2014 spring 2026",
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Signal\u2019s Four Frontier Models: A Testament To China\u2019s Speed In AI Development",
    "From late April to mid-June 2026, Chinese research labs introduced four major open-weight AI models, with DeepSeek V4 launched on April 24, followed by MiniMax M3 on June 1, and Kimi K2.7-Code and GLM-5.2 released within days of each other in mid-June.",
    "### The production line \u2014 spring 2026",
    "APR 24"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Signal\u2019s Four Frontier Models: A Testament To China\u2019s Speed In AI Development",
    "### The production line \u2014 spring 2026",
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor"
   ]
  },
  {
   "url": "https://techieus.com/general/signal-s-four-frontier-models-a-testament-to-china-s-speed-in-ai-development",
   "title": "Signal\u2019s Four Frontier Models: A Testament To China\u2019s Speed In AI Development - TechieUS",
   "published_at": "2026-07-15T06:16:58",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "cheap 1M-token context, native multimodal, modified-MIT",
   "organization_evidence": [
    "Signal\u2019s Four Frontier Models: A Testament To China\u2019s Speed In AI Development",
    "From late April to mid-June 2026, Chinese research labs introduced four major open-weight AI models, with DeepSeek V4 launched on April 24, followed by MiniMax M3 on June 1, and Kimi K2.7-Code and GLM-5.2 released within days of each other in mid-June.",
    "### The production line \u2014 spring 2026",
    "**MiniMax M3**cheap 1M-token context, native multimodal, modified-MIT"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Signal\u2019s Four Frontier Models: A Testament To China\u2019s Speed In AI Development",
    "From late April to mid-June 2026, Chinese research labs introduced four major open-weight AI models, with DeepSeek V4 launched on April 24, followed by MiniMax M3 on June 1, and Kimi K2.7-Code and GLM-5.2 released within days of each other in mid-June.",
    "### The production line \u2014 spring 2026",
    "JUN 01"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Signal\u2019s Four Frontier Models: A Testament To China\u2019s Speed In AI Development",
    "### The production line \u2014 spring 2026",
    "**MiniMax M3**cheap 1M-token context, native multimodal, modified-MIT"
   ]
  },
  {
   "url": "https://techieus.com/general/signal-s-four-frontier-models-a-testament-to-china-s-speed-in-ai-development",
   "title": "Signal\u2019s Four Frontier Models: A Testament To China\u2019s Speed In AI Development - TechieUS",
   "published_at": "2026-07-15T06:16:58",
   "organization": "Moonshot",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "agent-run specialist, ~30% fewer thinking tokens than K2.6",
   "organization_evidence": [
    "Signal\u2019s Four Frontier Models: A Testament To China\u2019s Speed In AI Development",
    "The Chinese open-weight ecosystem has expanded from a single lab two years ago to four major players: DeepSeek, Z.ai, Moonshot, and Alibaba, each with distinct strategic focuses.",
    "### The production line \u2014 spring 2026",
    "**Kimi K2.7-Code (Moonshot)**agent-run specialist, ~30% fewer thinking tokens than K2.6"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Signal\u2019s Four Frontier Models: A Testament To China\u2019s Speed In AI Development",
    "From late April to mid-June 2026, Chinese research labs introduced four major open-weight AI models, with DeepSeek V4 launched on April 24, followed by MiniMax M3 on June 1, and Kimi K2.7-Code and GLM-5.2 released within days of each other in mid-June.",
    "### The production line \u2014 spring 2026",
    "JUN 13"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Signal\u2019s Four Frontier Models: A Testament To China\u2019s Speed In AI Development",
    "### The production line \u2014 spring 2026",
    "**Kimi K2.7-Code (Moonshot)**agent-run specialist, ~30% fewer thinking tokens than K2.6"
   ]
  },
  {
   "url": "https://techieus.com/general/signal-s-four-frontier-models-a-testament-to-china-s-speed-in-ai-development",
   "title": "Signal\u2019s Four Frontier Models: A Testament To China\u2019s Speed In AI Development - TechieUS",
   "published_at": "2026-07-15T06:16:58",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-06-16",
   "access_type": null,
   "license": null,
   "significance": "753B MoE, MIT, top open-weight on Artificial Analysis index",
   "organization_evidence": [
    "Signal\u2019s Four Frontier Models: A Testament To China\u2019s Speed In AI Development",
    "The Chinese open-weight ecosystem has expanded from a single lab two years ago to four major players: DeepSeek, Z.ai, Moonshot, and Alibaba, each with distinct strategic focuses.",
    "### The production line \u2014 spring 2026",
    "**GLM-5.2 (Z.ai)**753B MoE, MIT, top open-weight on Artificial Analysis index"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Signal\u2019s Four Frontier Models: A Testament To China\u2019s Speed In AI Development",
    "From late April to mid-June 2026, Chinese research labs introduced four major open-weight AI models, with DeepSeek V4 launched on April 24, followed by MiniMax M3 on June 1, and Kimi K2.7-Code and GLM-5.2 released within days of each other in mid-June.",
    "### The production line \u2014 spring 2026",
    "JUN 13\u201316"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Signal\u2019s Four Frontier Models: A Testament To China\u2019s Speed In AI Development",
    "### The production line \u2014 spring 2026",
    "**GLM-5.2 (Z.ai)**753B MoE, MIT, top open-weight on Artificial Analysis index"
   ]
  },
  {
   "url": "https://eesel.ai/blog/inkling",
   "title": "Inkling explained: Thinking Machines' open-weights AI model",
   "published_at": "2026-07-20T00:00:00",
   "organization": "Thinking Machines Lab",
   "model": "Inkling",
   "release_date": "2026-07-15",
   "access_type": "open-weights",
   "license": "Apache 2.0",
   "significance": "a Mixture-of-Experts (MoE) transformer with 975B total parameters, native audio support, and a 1M-token context window",
   "organization_evidence": [
    "# Inkling explained: Thinking Machines' open-weights AI model",
    "Inkling explained: Thinking Machines' open-weights AI model",
    "![Illustration of Inkling, Thinking Machines Lab's open-weights AI model](/_next/image?url=https%3A%2F%2Fcdn-public.eesel.ai%2F314c3443-850d-416d-8721-d2e52c7ac2e2%2F585703e2-5508-4d19-8a3a-24b54e4a1f15%2Fcdd6118883054d5d8ac3bac58f6bd9f6.webp&w=1680&q=75)"
   ],
   "model_evidence": [
    "# Inkling explained: Thinking Machines' open-weights AI model",
    "Inkling explained: Thinking Machines' open-weights AI model",
    "Inkling is Thinking Machines Lab's [first public model](https://thinkingmachines.ai/news/introducing-inkling/), and the lab is refreshingly upfront that it isn't trying to win the leaderboard."
   ],
   "release_date_evidence": [
    "# Inkling explained: Thinking Machines' open-weights AI model",
    "Inkling explained: Thinking Machines' open-weights AI model",
    "- **Released:** July 15, 2026."
   ],
   "access_type_evidence": [
    "# Inkling explained: Thinking Machines' open-weights AI model",
    "Inkling explained: Thinking Machines' open-weights AI model",
    "![Illustration of Inkling, Thinking Machines Lab's open-weights AI model](/_next/image?url=https%3A%2F%2Fcdn-public.eesel.ai%2F314c3443-850d-416d-8721-d2e52c7ac2e2%2F585703e2-5508-4d19-8a3a-24b54e4a1f15%2Fcdd6118883054d5d8ac3bac58f6bd9f6.webp&w=1680&q=75)"
   ],
   "license_evidence": [
    "# Inkling explained: Thinking Machines' open-weights AI model",
    "Inkling explained: Thinking Machines' open-weights AI model",
    "- **License:** **Apache 2.0**, so commercial use is allowed and the weights are [on Hugging Face](https://huggingface.co/thinkingmachines/Inkling)."
   ],
   "significance_evidence": [
    "# Inkling explained: Thinking Machines' open-weights AI model",
    "Inkling explained: Thinking Machines' open-weights AI model",
    "- **Architecture:** a Mixture-of-Experts (MoE) transformer, **975B total parameters** with **41B active** per token (it routes each token through 6 of 256 experts, plus 2 shared experts)."
   ]
  },
  {
   "url": "https://versely.studio/blog/qwen-3-5-alibaba-open-source-surprise-2026",
   "title": "Qwen 3.5: How Alibaba Became the Open-Source Surprise of 2026",
   "published_at": "2026-05-15T15:23:25",
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-02-16",
   "access_type": null,
   "license": null,
   "significance": "A sparse Mixture-of-Experts with 397 billion total parameters and roughly 17 billion active per token that takes 50% of global open-source LLM downloads inside six weeks.",
   "organization_evidence": [
    "AI News",
    "Qwen 3.5: How Alibaba Became the Open-Source Surprise of 2026",
    "When Alibaba dropped Qwen3.5-397B-A17B on Hugging Face on February 16, 2026, nobody outside China expected an open-weight Mixture-of-Experts to take 50% of global open-source LLM downloads inside six weeks."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "AI News",
    "Qwen 3.5: How Alibaba Became the Open-Source Surprise of 2026",
    "When Alibaba dropped Qwen3.5-397B-A17B on Hugging Face on February 16, 2026, nobody outside China expected an open-weight Mixture-of-Experts to take 50% of global open-source LLM downloads inside six weeks."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "AI News",
    "Qwen 3.5: How Alibaba Became the Open-Source Surprise of 2026",
    "When Alibaba dropped Qwen3.5-397B-A17B on Hugging Face on February 16, 2026, nobody outside China expected an open-weight Mixture-of-Experts to take 50% of global open-source LLM downloads inside six weeks.",
    "The flagship Qwen3.5-397B-A17B is a sparse Mixture-of-Experts with 397 billion total parameters and roughly 17 billion active per token."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/gemma-4-26b-a4b-vs-mimo-v2-pro",
   "title": "Gemma 4 26B A4B (Reasoning) vs MiMo-V2-Pro: Model Comparison",
   "published_at": null,
   "organization": "Google",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "Gemma 4 26B A4B (Reasoning) features a 256k context window and image input support.",
   "organization_evidence": [
    "# Gemma 4 26B A4B (Reasoning) vs. MiMo-V2-Pro",
    "| Metric | ![Google logo](/img/logos/google_small.svg)Gemma 4 26B A4B (Reasoning) | ![Xiaomi logo](/img/logos/xiaomi_small.svg)MiMo-V2-Pro | Analysis |",
    "| Creator | ![Google](/img/logos/google_small.svg)[](https://deepmind.google/technologies/gemini/) | ![Xiaomi](/img/logos/xiaomi_small.svg)[](https://huggingface.co/XiaomiMiMo) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Gemma 4 26B A4B (Reasoning) vs. MiMo-V2-Pro",
    "| Metric | ![Google logo](/img/logos/google_small.svg)Gemma 4 26B A4B (Reasoning) | ![Xiaomi logo](/img/logos/xiaomi_small.svg)MiMo-V2-Pro | Analysis |",
    "| Release Date | April, 2026 | March, 2026 | Gemma 4 26B A4B (Reasoning) has a more recent release date than MiMo-V2-Pro |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Gemma 4 26B A4B (Reasoning) vs. MiMo-V2-Pro",
    "| Metric | ![Google logo](/img/logos/google_small.svg)Gemma 4 26B A4B (Reasoning) | ![Xiaomi logo](/img/logos/xiaomi_small.svg)MiMo-V2-Pro | Analysis |",
    "| Context Window | 256k tokens (~384 A4 pages of size 12 Arial font) | 1000k tokens (~1500 A4 pages of size 12 Arial font) | Gemma 4 26B A4B (Reasoning) is smaller than MiMo-V2-Pro |",
    "| Image Input Support | Yes | No | Gemma 4 26B A4B (Reasoning) has image input support while MiMo-V2-Pro does not |"
   ]
  },
  {
   "url": "https://frontiernews.ai/news/article/chinese-ai-labs-are-quietly-building-the-next-gene-70a051d9",
   "title": "Chinese AI Labs Are Quietly Building the Next Generation of Open-Weight Models",
   "published_at": "2026-06-15T06:33:57",
   "organization": "Z.ai",
   "model": "GLM-5.2",
   "release_date": "2026-06-13",
   "access_type": "open-weight",
   "license": "MIT or modified MIT licenses",
   "significance": "features a 1-million-token context window and two reasoning effort levels",
   "organization_evidence": [
    "# Chinese AI Labs Are Quietly Building the Next Generation of Open-Weight Models",
    "Z.ai launched GLM-5.2 on June 13, 2026, featuring a 1-million-token context window and two reasoning effort levels, while Moonshot AI's Kimi K2.6, released in April 2026, offers native multimodal abilities and improved agent coordination."
   ],
   "model_evidence": [
    "# Chinese AI Labs Are Quietly Building the Next Generation of Open-Weight Models",
    "Z.ai launched GLM-5.2 on June 13, 2026, featuring a 1-million-token context window and two reasoning effort levels, while Moonshot AI's Kimi K2.6, released in April 2026, offers native multimodal abilities and improved agent coordination."
   ],
   "release_date_evidence": [
    "# Chinese AI Labs Are Quietly Building the Next Generation of Open-Weight Models",
    "Z.ai launched GLM-5.2 on June 13, 2026, featuring a 1-million-token context window and two reasoning effort levels, while Moonshot AI's Kimi K2.6, released in April 2026, offers native multimodal abilities and improved agent coordination."
   ],
   "access_type_evidence": [
    "# Chinese AI Labs Are Quietly Building the Next Generation of Open-Weight Models",
    "Chinese AI companies are releasing open-weight large language models with capabilities that rival or exceed Western alternatives, marking a significant shift in the global AI race."
   ],
   "license_evidence": [
    "# Chinese AI Labs Are Quietly Building the Next Generation of Open-Weight Models",
    "Both GLM-5.2 and Kimi K2 are released as open-weight models under MIT or modified MIT licenses."
   ],
   "significance_evidence": [
    "# Chinese AI Labs Are Quietly Building the Next Generation of Open-Weight Models",
    "Z.ai launched GLM-5.2 on June 13, 2026, featuring a 1-million-token context window and two reasoning effort levels, while Moonshot AI's Kimi K2.6, released in April 2026, offers native multimodal abilities and improved agent coordination."
   ]
  },
  {
   "url": "https://frontiernews.ai/news/article/chinese-ai-labs-are-quietly-building-the-next-gene-70a051d9",
   "title": "Chinese AI Labs Are Quietly Building the Next Generation of Open-Weight Models",
   "published_at": "2026-06-15T06:33:57",
   "organization": "Moonshot AI",
   "model": "Kimi K2.6",
   "release_date": "2026-04",
   "access_type": "open-weight",
   "license": "MIT or modified MIT licenses",
   "significance": "offers native multimodal abilities and improved agent coordination",
   "organization_evidence": [
    "# Chinese AI Labs Are Quietly Building the Next Generation of Open-Weight Models",
    "Z.ai launched GLM-5.2 on June 13, 2026, featuring a 1-million-token context window and two reasoning effort levels, while Moonshot AI's Kimi K2.6, released in April 2026, offers native multimodal abilities and improved agent coordination."
   ],
   "model_evidence": [
    "# Chinese AI Labs Are Quietly Building the Next Generation of Open-Weight Models",
    "Z.ai launched GLM-5.2 on June 13, 2026, featuring a 1-million-token context window and two reasoning effort levels, while Moonshot AI's Kimi K2.6, released in April 2026, offers native multimodal abilities and improved agent coordination."
   ],
   "release_date_evidence": [
    "# Chinese AI Labs Are Quietly Building the Next Generation of Open-Weight Models",
    "Z.ai launched GLM-5.2 on June 13, 2026, featuring a 1-million-token context window and two reasoning effort levels, while Moonshot AI's Kimi K2.6, released in April 2026, offers native multimodal abilities and improved agent coordination."
   ],
   "access_type_evidence": [
    "# Chinese AI Labs Are Quietly Building the Next Generation of Open-Weight Models",
    "Chinese AI companies are releasing open-weight large language models with capabilities that rival or exceed Western alternatives, marking a significant shift in the global AI race."
   ],
   "license_evidence": [
    "# Chinese AI Labs Are Quietly Building the Next Generation of Open-Weight Models",
    "Both GLM-5.2 and Kimi K2 are released as open-weight models under MIT or modified MIT licenses."
   ],
   "significance_evidence": [
    "# Chinese AI Labs Are Quietly Building the Next Generation of Open-Weight Models",
    "Z.ai launched GLM-5.2 on June 13, 2026, featuring a 1-million-token context window and two reasoning effort levels, while Moonshot AI's Kimi K2.6, released in April 2026, offers native multimodal abilities and improved agent coordination."
   ]
  },
  {
   "url": "https://siliconangle.com/2026/07/15/mira-muratis-thinking-machines-drops-inkling-open-weights-model-anyone-can-access",
   "title": "Mira Murati\u2019s Thinking Machines drops Inkling, an open-weights model anyone can access",
   "published_at": "2026-07-15T00:00:00",
   "organization": "Thinking Machines",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "A mixture-of-experts model with 975 billion parameters trained from scratch on about 45 trillion tokens of text, image, audio, and video.",
   "organization_evidence": [
    "Mira Murati\u2019s [Thinking Machines Lab Inc.](https://thinkingmachines.ai/) today launched its first foundation model with the release of [Inkling](https://thinkingmachines.ai/inkling/), making its full open weights available to developers so they can fine-tune it as they wish.",
    "Inkling is the first model fully trained from scratch by Thinking Machines, coming after a year in which the company mostly made headlines for its [sizable funding rounds](https://siliconangle.com/2025/07/15/thinking-machines-led-former-openai-cto-mira-murati-raises-2b-seed-funding/) and its partnership with Nvidia Corp."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Our first model, Inkling. Trained from scratch, weights are open, fine-tunable on Tinker today. [https://t.co/m7q5RsX0Ud](https://t.co/m7q5RsX0Ud)",
    "\u2014 Mira Murati (@miramurati) [July 15, 2026](https://x.com/miramurati/status/2077455974743593100?ref_src=twsrc%5Etfw)"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "In a blog post, Thinking Machines [explained](https://thinkingmachines.ai/news/introducing-inkling/) that Inkling is a mixture-of-experts model that features 975 billion parameters, although for the average prompt it will only draw on a small fraction of that number \u2013 about 41 billion \u2013 in order to process tasks faster and keep costs low.",
    "The company said the model was trained on about 45 trillion tokens of text, image, audio and video and can reason natively across all four inputs."
   ]
  },
  {
   "url": "https://pbxscience.com/2-8-trillion-parameters-moonshot-ais-kimi-k3-becomes-worlds-largest-open-source-model-earns-musks-praise",
   "title": "2.8 Trillion Parameters: Moonshot AI\u2019s Kimi K3 Becomes World\u2019s Largest Open-Source Model, Earns Musk\u2019s Praise",
   "published_at": "2026-07-19T00:33:07",
   "organization": "Moonshot AI",
   "model": "Kimi K3",
   "release_date": "2026-07-16",
   "access_type": "open-source",
   "license": "modified MIT-style license",
   "significance": "2.8 trillion total parameters, making it the largest open-weight language model ever released with a 1-million-token context window and native visual understanding.",
   "organization_evidence": [
    "Chinese AI startup Moonshot AI released its newest flagship model, Kimi K3, on July 16, 2026, and it is already being called a milestone for open-source artificial intelligence.",
    "2.8 Trillion Parameters: Moonshot AI\u2019s Kimi K3 Becomes World\u2019s Largest Open-Source Model, Earns Musk\u2019s Praise"
   ],
   "model_evidence": [
    "Chinese AI startup Moonshot AI released its newest flagship model, Kimi K3, on July 16, 2026, and it is already being called a milestone for open-source artificial intelligence.",
    "2.8 Trillion Parameters: Moonshot AI\u2019s Kimi K3 Becomes World\u2019s Largest Open-Source Model, Earns Musk\u2019s Praise"
   ],
   "release_date_evidence": [
    "Chinese AI startup Moonshot AI released its newest flagship model, Kimi K3, on July 16, 2026, and it is already being called a milestone for open-source artificial intelligence.",
    "2.8 Trillion Parameters: Moonshot AI\u2019s Kimi K3 Becomes World\u2019s Largest Open-Source Model, Earns Musk\u2019s Praise"
   ],
   "access_type_evidence": [
    "Chinese AI startup Moonshot AI released its newest flagship model, Kimi K3, on July 16, 2026, and it is already being called a milestone for open-source artificial intelligence.",
    "2.8 Trillion Parameters: Moonshot AI\u2019s Kimi K3 Becomes World\u2019s Largest Open-Source Model, Earns Musk\u2019s Praise"
   ],
   "license_evidence": [
    "Moonshot has said it will publish K3\u2019s complete model weights to the community on July 27, 2026, under a modified MIT-style license, allowing developers worldwide to download and run the model themselves.",
    "2.8 Trillion Parameters: Moonshot AI\u2019s Kimi K3 Becomes World\u2019s Largest Open-Source Model, Earns Musk\u2019s Praise"
   ],
   "significance_evidence": [
    "With 2.8 trillion total parameters, K3 is now the largest open-weight language model ever released, surpassing every other open model on the market, including DeepSeek\u2019s V4 Pro and Zhipu AI\u2019s GLM 5 series.",
    "The model features a 1-million-token context window and native visual understanding, positioning it for long-horizon coding, agentic workflows, and knowledge-heavy tasks.",
    "2.8 Trillion Parameters: Moonshot AI\u2019s Kimi K3 Becomes World\u2019s Largest Open-Source Model, Earns Musk\u2019s Praise"
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/qwen3-6-35b-a3b-vs-gemma-4-26b-a4b",
   "title": "Qwen3.6 35B A3B (Reasoning) vs Gemma 4 26B A4B (Reasoning): Model Comparison",
   "published_at": null,
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "Intelligence Index 32 with 36B total parameters and reasoning support",
   "organization_evidence": [
    "# Qwen3.6 35B A3B (Reasoning) vs. Gemma 4 26B A4B (Reasoning)",
    "| | ![Alibaba logo](/img/logos/alibaba_small.svg)Alibaba Qwen3.6 35B A3B (Reasoning) | ![Google logo](/img/logos/google_small.svg)Google Gemma 4 26B A4B (Reasoning) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Qwen3.6 35B A3B (Reasoning) vs. Gemma 4 26B A4B (Reasoning)",
    "| Release Date | April 2026 | April 2026 | Qwen3.6 35B A3B (Reasoning) has a more recent release date than Gemma 4 26B A4B (Reasoning) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Qwen3.6 35B A3B (Reasoning) vs. Gemma 4 26B A4B (Reasoning)",
    "| Intelligence Index | 32 | 26 | Qwen3.6 35B A3B (Reasoning) is more intelligent than Gemma 4 26B A4B (Reasoning) |",
    "| Parameters | 36B, 3B active at inference time | 25.2B, 3.8B active at inference time | Qwen3.6 35B A3B (Reasoning) has more parameters than Gemma 4 26B A4B (Reasoning) |",
    "| Reasoning | Yes | Yes | Both Qwen3.6 35B A3B (Reasoning) and Gemma 4 26B A4B (Reasoning) have reasoning |"
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/qwen3-6-35b-a3b-vs-gemma-4-26b-a4b",
   "title": "Qwen3.6 35B A3B (Reasoning) vs Gemma 4 26B A4B (Reasoning): Model Comparison",
   "published_at": null,
   "organization": "Google",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "Intelligence Index 26 with 25.2B total parameters and reasoning support",
   "organization_evidence": [
    "# Qwen3.6 35B A3B (Reasoning) vs. Gemma 4 26B A4B (Reasoning)",
    "| | ![Alibaba logo](/img/logos/alibaba_small.svg)Alibaba Qwen3.6 35B A3B (Reasoning) | ![Google logo](/img/logos/google_small.svg)Google Gemma 4 26B A4B (Reasoning) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Qwen3.6 35B A3B (Reasoning) vs. Gemma 4 26B A4B (Reasoning)",
    "| Release Date | April 2026 | April 2026 | Qwen3.6 35B A3B (Reasoning) has a more recent release date than Gemma 4 26B A4B (Reasoning) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Qwen3.6 35B A3B (Reasoning) vs. Gemma 4 26B A4B (Reasoning)",
    "| Intelligence Index | 32 | 26 | Qwen3.6 35B A3B (Reasoning) is more intelligent than Gemma 4 26B A4B (Reasoning) |",
    "| Parameters | 36B, 3B active at inference time | 25.2B, 3.8B active at inference time | Qwen3.6 35B A3B (Reasoning) has more parameters than Gemma 4 26B A4B (Reasoning) |",
    "| Reasoning | Yes | Yes | Both Qwen3.6 35B A3B (Reasoning) and Gemma 4 26B A4B (Reasoning) have reasoning |"
   ]
  },
  {
   "url": "https://frontiernews.ai/news/article/chinas-ai-labs-just-released-models-larger-than-gp-b4f1292a",
   "title": "China's AI Labs Just Released Models Larger Than GPT-5.6, and They're Giving Away the Code",
   "published_at": "2026-07-20T17:42:21",
   "organization": "Moonshot",
   "model": null,
   "release_date": "2026-07-20",
   "access_type": null,
   "license": null,
   "significance": "world's largest open-source AI system at 2.8 trillion parameters",
   "organization_evidence": [
    "# China's AI Labs Just Released Models Larger Than GPT-5.6, and They're Giving Away the Code",
    "Moonshot's Kimi K3 launched on July 20, 2026, at 2.8 trillion parameters, positioning itself as the world's largest open-source AI system."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# China's AI Labs Just Released Models Larger Than GPT-5.6, and They're Giving Away the Code",
    "Moonshot's Kimi K3 launched on July 20, 2026, at 2.8 trillion parameters, positioning itself as the world's largest open-source AI system."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# China's AI Labs Just Released Models Larger Than GPT-5.6, and They're Giving Away the Code",
    "Moonshot's Kimi K3 launched on July 20, 2026, at 2.8 trillion parameters, positioning itself as the world's largest open-source AI system."
   ]
  },
  {
   "url": "https://frontiernews.ai/news/article/chinas-ai-labs-just-released-models-larger-than-gp-b4f1292a",
   "title": "China's AI Labs Just Released Models Larger Than GPT-5.6, and They're Giving Away the Code",
   "published_at": "2026-07-20T17:42:21",
   "organization": "Alibaba",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "frontier-level model at 2.4 trillion parameters, second only to Fable 5",
   "organization_evidence": [
    "# China's AI Labs Just Released Models Larger Than GPT-5.6, and They're Giving Away the Code",
    "Alibaba followed with a preview of Qwen3.8 at 2.4 trillion parameters, describing it as \"second only to Fable 5.\""
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# China's AI Labs Just Released Models Larger Than GPT-5.6, and They're Giving Away the Code",
    "**Chinese AI startups Moonshot and Alibaba have released frontier-level models at 2.8 and 2.4 trillion parameters respectively, and both plan to publish their full model weights publicly, collapsing two of Silicon Valley's traditional advantages: scale and secrecy.**",
    "Alibaba followed with a preview of Qwen3.8 at 2.4 trillion parameters, describing it as \"second only to Fable 5.\""
   ]
  },
  {
   "url": "https://frontiernews.ai/news/article/the-open-source-llm-race-just-got-fiercer-why-llam-5191e884",
   "title": "The Open-Source LLM Race Just Got Fiercer: Why Llama 4, Qwen 3.5, and Mistral Are Reshaping AI in 2026",
   "published_at": "2026-07-06T12:24:48",
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-02",
   "access_type": null,
   "license": null,
   "significance": "Tops out at a 397-billion-total and 17-billion-active mixture-of-experts model built from 256 experts with a native 262,144-token context window, native vision-language support, and coverage of 201 languages.",
   "organization_evidence": [
    "# The Open-Source LLM Race Just Got Fiercer: Why Llama 4, Qwen 3.5, and Mistral Are Reshaping AI in 2026",
    "As of July 2026, three families dominate conversations whenever engineers ask which open-weight model to build on: Meta's Llama 4, Alibaba's Qwen 3.5, and Mistral AI's Large 3.",
    "Alibaba's Qwen 3.5, rolled out in February 2026, tops out at a 397-billion-total and 17-billion-active mixture-of-experts (MoE) model built from 256 experts."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# The Open-Source LLM Race Just Got Fiercer: Why Llama 4, Qwen 3.5, and Mistral Are Reshaping AI in 2026",
    "Alibaba's Qwen 3.5, rolled out in February 2026, tops out at a 397-billion-total and 17-billion-active mixture-of-experts (MoE) model built from 256 experts."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# The Open-Source LLM Race Just Got Fiercer: Why Llama 4, Qwen 3.5, and Mistral Are Reshaping AI in 2026",
    "Alibaba's Qwen 3.5, rolled out in February 2026, tops out at a 397-billion-total and 17-billion-active mixture-of-experts (MoE) model built from 256 experts.",
    "It ships with a native 262,144-token context window that extends toward roughly one million tokens, native vision-language support, and coverage of an astonishing 201 languages."
   ]
  },
  {
   "url": "https://frontiernews.ai/news/article/the-open-source-llm-race-just-got-fiercer-why-llam-5191e884",
   "title": "The Open-Source LLM Race Just Got Fiercer: Why Llama 4, Qwen 3.5, and Mistral Are Reshaping AI in 2026",
   "published_at": "2026-07-06T12:24:48",
   "organization": "Mistral AI",
   "model": null,
   "release_date": "2026-03",
   "access_type": null,
   "license": null,
   "significance": "A lean 119-billion-total model that activates only 6 billion parameters per token.",
   "organization_evidence": [
    "# The Open-Source LLM Race Just Got Fiercer: Why Llama 4, Qwen 3.5, and Mistral Are Reshaping AI in 2026",
    "As of July 2026, three families dominate conversations whenever engineers ask which open-weight model to build on: Meta's Llama 4, Alibaba's Qwen 3.5, and Mistral AI's Large 3.",
    "In March 2026, Mistral added Mistral Small 4, a remarkably lean 119-billion-total model that activates only 6 billion parameters per token."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# The Open-Source LLM Race Just Got Fiercer: Why Llama 4, Qwen 3.5, and Mistral Are Reshaping AI in 2026",
    "In March 2026, Mistral added Mistral Small 4, a remarkably lean 119-billion-total model that activates only 6 billion parameters per token."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# The Open-Source LLM Race Just Got Fiercer: Why Llama 4, Qwen 3.5, and Mistral Are Reshaping AI in 2026",
    "In March 2026, Mistral added Mistral Small 4, a remarkably lean 119-billion-total model that activates only 6 billion parameters per token."
   ]
  },
  {
   "url": "https://lithosai.com/blog/open-weight-models-no-longer-catching-up",
   "title": "Open Weight Models Are No Longer Just Catching Up | LithosAI",
   "published_at": null,
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-16",
   "access_type": null,
   "license": null,
   "significance": "A Mixture-of-Experts model with 2.8 trillion total parameters and a native 1M-token context window that sets the frontier on agentic benchmarks.",
   "organization_evidence": [
    "# Open Weight Models Are No Longer Just Catching Up",
    "Kimi K3, the open-weight model released yesterday by Moonshot AI, breaks this pattern."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Open Weight Models Are No Longer Just Catching Up",
    "July 17, 2026 \u00b7 5 min read",
    "Kimi K3, the open-weight model released yesterday by Moonshot AI, breaks this pattern."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Open Weight Models Are No Longer Just Catching Up",
    "Kimi K3, the open-weight model released yesterday by Moonshot AI, breaks this pattern.",
    "It is a Mixture-of-Experts model with 2.8 trillion total parameters and a native 1M-token context window, and on the benchmarks that measure sustained agentic work, **it now sets the frontier rather than chasing it**."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/glm-5-1-vs-mimo-v2-pro",
   "title": "GLM-5.1 (Reasoning) vs MiMo-V2-Pro: Model Comparison",
   "published_at": null,
   "organization": "Z AI",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "Reasoning model featuring a 200k token context window and open source weights",
   "organization_evidence": [
    "# GLM-5.1 (Reasoning) vs. MiMo-V2-Pro",
    "## Model Comparison",
    "| Creator | ![Z AI](/img/logos/zai_small.svg)[](https://z.ai/) | ![Xiaomi](/img/logos/xiaomi_small.svg)[](https://huggingface.co/XiaomiMiMo) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# GLM-5.1 (Reasoning) vs. MiMo-V2-Pro",
    "## Model Comparison",
    "| Release Date | April, 2026 | March, 2026 | GLM-5.1 (Reasoning) has a more recent release date than MiMo-V2-Pro |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# GLM-5.1 (Reasoning) vs. MiMo-V2-Pro",
    "## Model Comparison",
    "| Context Window | 200k tokens (~300 A4 pages of size 12 Arial font) | 1000k tokens (~1500 A4 pages of size 12 Arial font) | GLM-5.1 (Reasoning) is smaller than MiMo-V2-Pro |",
    "| Open Source (Weights) | Yes[](https://huggingface.co/zai-org/GLM-5.1) | No | GLM-5.1 (Reasoning) is open source while MiMo-V2-Pro is proprietary |"
   ]
  },
  {
   "url": "https://shaam.blog/articles/inkling-small-open-weights-model-2026",
   "title": "Inkling-Small (2026): The Open-Weights Model That Matches Its 975B Sibling at a Quarter the Size",
   "published_at": "2026-08-03T23:13:47",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-30",
   "access_type": null,
   "license": null,
   "significance": "a 276-billion-parameter Mixture-of-Experts model that activates only 12 billion per token, yet matches or beats its 975-billion-parameter sibling Inkling on reasoning and agentic coding benchmarks while costing roughly a quarter as much to run",
   "organization_evidence": [
    "Inkling-Small is Thinking Machines' 276B open-weights model that matches the 975B Inkling on reasoning and coding.",
    "# Inkling-Small (2026): The Open-Weights Model That Matches Its 975B Sibling at a Quarter the Size"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Released on July 30, 2026 under the Apache 2.0 license with full weights on Hugging Face, it is the first credible proof that the \"train a giant, then distill a smaller version fast\" approach scales as a repeatable playbook \u2014 not a one-off.",
    "# Inkling-Small (2026): The Open-Weights Model That Matches Its 975B Sibling at a Quarter the Size"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "**Verdict:** Inkling-Small is the most efficient open-weights model Thinking Machines Lab has shipped \u2014 a 276-billion-parameter Mixture-of-Experts model that activates only 12 billion per token, yet matches or beats its 975-billion-parameter sibling Inkling on reasoning and agentic coding benchmarks while costing roughly a quarter as much to run.",
    "# Inkling-Small (2026): The Open-Weights Model That Matches Its 975B Sibling at a Quarter the Size"
   ]
  },
  {
   "url": "https://shaam.blog/articles/inkling-small-open-weights-model-2026",
   "title": "Inkling-Small (2026): The Open-Weights Model That Matches Its 975B Sibling at a Quarter the Size",
   "published_at": "2026-08-03T23:13:47",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "a 975-billion-parameter MoE model",
   "organization_evidence": [
    "The model is the smaller sibling of Inkling, a 975-billion-parameter MoE model (41 billion active) that Thinking Machines released just two weeks earlier on July 15, 2026.",
    "# Inkling-Small (2026): The Open-Weights Model That Matches Its 975B Sibling at a Quarter the Size"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The model is the smaller sibling of Inkling, a 975-billion-parameter MoE model (41 billion active) that Thinking Machines released just two weeks earlier on July 15, 2026.",
    "# Inkling-Small (2026): The Open-Weights Model That Matches Its 975B Sibling at a Quarter the Size"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The model is the smaller sibling of Inkling, a 975-billion-parameter MoE model (41 billion active) that Thinking Machines released just two weeks earlier on July 15, 2026.",
    "# Inkling-Small (2026): The Open-Weights Model That Matches Its 975B Sibling at a Quarter the Size"
   ]
  },
  {
   "url": "https://venturebeat.com/technology/cohere-cracks-lossless-quantization-and-native-citations-with-first-full-apache-2-0-licensed-open-model-command-a",
   "title": "Cohere cracks lossless quantization and native citations with first full Apache 2.0 licensed open model Command A+",
   "published_at": "2026-05-20T21:16:49",
   "organization": "Cohere",
   "model": null,
   "release_date": "2026-05",
   "access_type": null,
   "license": null,
   "significance": "A 218-billion-parameter language model engineered specifically for complex reasoning, multimodal document processing, and agentic workflows.",
   "organization_evidence": [
    "Canadian AI lab [Cohere](https://cohere.com/) made waves recently by [announcing a merger with German AI startup Aleph Alpha](https://cohere.com/blog/cohere-alephalpha-join-forces), but now it has even more in store for enterprise builders around the globe: today, the firm [co-founded by former Googler and \"Attention Is All You Need\" co-author Aidan Gomez](https://venturebeat.com/ai/openai-rival-cohere-ai-has-flown-under-the-radar-that-may-be-about-to-change) unveiled [Command A+](https://cohere.com/blog/command-a-plus), a highly optimized, 218-billion-parameter language model engineered specifically for complex reasoning, multimodal document processing, and agentic workflows."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "By releasing the model weights free on the [popular AI code sharing repository Hugging Face](https://huggingface.co/CohereLabs/command-a-plus-05-2026-w4a4) under a [highly permissive Apache 2.0 open-source license](https://x.com/aidangomez/status/2057142232860258527) \u2014 a first for the company, according to [a post by Gomez, now Cohere's CEO, on X](https://x.com/aidangomez/status/2057142232860258527?s=20) \u2014 Cohere is making a calculated bet on \"sovereign AI\"\u2014the thesis that enterprises, governments, and developers should have the ability to run, control, and adapt frontier-grade AI entirely within their own secure environments, without sacrificing performance.",
    "Canadian AI lab [Cohere](https://cohere.com/) made waves recently by [announcing a merger with German AI startup Aleph Alpha](https://cohere.com/blog/cohere-alephalpha-join-forces), but now it has even more in store for enterprise builders around the globe: today, the firm [co-founded by former Googler and \"Attention Is All You Need\" co-author Aidan Gomez](https://venturebeat.com/ai/openai-rival-cohere-ai-has-flown-under-the-radar-that-may-be-about-to-change) unveiled [Command A+](https://cohere.com/blog/command-a-plus), a highly optimized, 218-billion-parameter language model engineered specifically for complex reasoning, multimodal document processing, and agentic workflows."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Canadian AI lab [Cohere](https://cohere.com/) made waves recently by [announcing a merger with German AI startup Aleph Alpha](https://cohere.com/blog/cohere-alephalpha-join-forces), but now it has even more in store for enterprise builders around the globe: today, the firm [co-founded by former Googler and \"Attention Is All You Need\" co-author Aidan Gomez](https://venturebeat.com/ai/openai-rival-cohere-ai-has-flown-under-the-radar-that-may-be-about-to-change) unveiled [Command A+](https://cohere.com/blog/command-a-plus), a highly optimized, 218-billion-parameter language model engineered specifically for complex reasoning, multimodal document processing, and agentic workflows."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/deepseek-v4-pro-vs-gpt-5-5",
   "title": "DeepSeek V4 Pro 0813 (Reasoning, Max Effort) vs GPT-5.5 (xhigh): Model Comparison",
   "published_at": null,
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-08",
   "access_type": null,
   "license": null,
   "significance": "A major large language model featuring reasoning capabilities, an intelligence index of 53, and a 1M token context window.",
   "organization_evidence": [
    "DeepSeek V4 Pro 0813 (Reasoning, Max Effort) vs. GPT-5.5 (xhigh)",
    "| | ![DeepSeek logo](/img/logos/deepseek_small.svg)DeepSeek DeepSeek V4 Pro 0813 (Reasoning, Max Effort) | ![OpenAI logo](/img/logos/openai_small.svg)OpenAI GPT-5.5 (xhigh) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "DeepSeek V4 Pro 0813 (Reasoning, Max Effort) vs. GPT-5.5 (xhigh)",
    "| Release Date | August 2026 | April 2026 | DeepSeek V4 Pro 0813 (Reasoning, Max Effort) has a more recent release date than GPT-5.5 (xhigh) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "DeepSeek V4 Pro 0813 (Reasoning, Max Effort) vs. GPT-5.5 (xhigh)",
    "| Intelligence Index | 53 | 56 | GPT-5.5 (xhigh) is more intelligent than DeepSeek V4 Pro 0813 (Reasoning, Max Effort) |",
    "| Context Window | 1000k tokens~1,500 A4 pages of size 12 Arial font | 922k tokens~1,383 A4 pages of size 12 Arial font | DeepSeek V4 Pro 0813 (Reasoning, Max Effort) has a larger context window than GPT-5.5 (xhigh) |",
    "| Reasoning | Yes | Yes | Both DeepSeek V4 Pro 0813 (Reasoning, Max Effort) and GPT-5.5 (xhigh) have reasoning |"
   ]
  },
  {
   "url": "https://nanomachines.net/business-markets/the-timeline-of-china-s-four-frontier-ai-models-launch-in-record-time",
   "title": "The Timeline Of China\u2019s Four Frontier AI Models Launch In Record Time - NanoMachines",
   "published_at": "2026-07-15T06:19:58",
   "organization": "DeepSeek",
   "model": "DeepSeek V4",
   "release_date": "2026-04-24",
   "access_type": "open-weight",
   "license": "MIT",
   "significance": "1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor",
   "organization_evidence": [
    "Starting with DeepSeek V4 on April 24, followed by MiniMax M3 on June 1, and then Kimi K2.7-Code and GLM-5.2 within days of each other in mid-June, Chinese laboratories have exhibited a rapid development cycle.",
    "### The production line \u2014 spring 2026"
   ],
   "model_evidence": [
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor",
    "### The production line \u2014 spring 2026"
   ],
   "release_date_evidence": [
    "Starting with DeepSeek V4 on April 24, followed by MiniMax M3 on June 1, and then Kimi K2.7-Code and GLM-5.2 within days of each other in mid-June, Chinese laboratories have exhibited a rapid development cycle.",
    "### The production line \u2014 spring 2026"
   ],
   "access_type_evidence": [
    "In just eight weeks, Chinese AI labs launched four frontier-class open-weight models, demonstrating an accelerated development pace.",
    "### The production line \u2014 spring 2026"
   ],
   "license_evidence": [
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor",
    "### The production line \u2014 spring 2026"
   ],
   "significance_evidence": [
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor",
    "### The production line \u2014 spring 2026"
   ]
  },
  {
   "url": "https://nanomachines.net/business-markets/the-timeline-of-china-s-four-frontier-ai-models-launch-in-record-time",
   "title": "The Timeline Of China\u2019s Four Frontier AI Models Launch In Record Time - NanoMachines",
   "published_at": "2026-07-15T06:19:58",
   "organization": "MiniMax",
   "model": "MiniMax M3",
   "release_date": "2026-06-01",
   "access_type": "open-weight",
   "license": "modified-MIT",
   "significance": "cheap 1M-token context, native multimodal, modified-MIT",
   "organization_evidence": [
    "Starting with DeepSeek V4 on April 24, followed by MiniMax M3 on June 1, and then Kimi K2.7-Code and GLM-5.2 within days of each other in mid-June, Chinese laboratories have exhibited a rapid development cycle.",
    "### The production line \u2014 spring 2026"
   ],
   "model_evidence": [
    "### The production line \u2014 spring 2026"
   ],
   "release_date_evidence": [
    "Starting with DeepSeek V4 on April 24, followed by MiniMax M3 on June 1, and then Kimi K2.7-Code and GLM-5.2 within days of each other in mid-June, Chinese laboratories have exhibited a rapid development cycle.",
    "### The production line \u2014 spring 2026"
   ],
   "access_type_evidence": [
    "In just eight weeks, Chinese AI labs launched four frontier-class open-weight models, demonstrating an accelerated development pace.",
    "### The production line \u2014 spring 2026"
   ],
   "license_evidence": [
    "### The production line \u2014 spring 2026"
   ],
   "significance_evidence": [
    "### The production line \u2014 spring 2026"
   ]
  },
  {
   "url": "https://nanomachines.net/business-markets/the-timeline-of-china-s-four-frontier-ai-models-launch-in-record-time",
   "title": "The Timeline Of China\u2019s Four Frontier AI Models Launch In Record Time - NanoMachines",
   "published_at": "2026-07-15T06:19:58",
   "organization": "Moonshot",
   "model": "Kimi K2.7-Code",
   "release_date": "2026-06-13",
   "access_type": "open-weight",
   "license": null,
   "significance": "agent-run specialist, ~30% fewer thinking tokens than K2.6",
   "organization_evidence": [
    "Chinese organizations such as Z.ai, Moonshot, and Alibaba have each introduced models targeting different aspects of AI performance, including cost efficiency, long-term stability, and accessibility.",
    "### The production line \u2014 spring 2026"
   ],
   "model_evidence": [
    "### The production line \u2014 spring 2026"
   ],
   "release_date_evidence": [
    "Starting with DeepSeek V4 on April 24, followed by MiniMax M3 on June 1, and then Kimi K2.7-Code and GLM-5.2 within days of each other in mid-June, Chinese laboratories have exhibited a rapid development cycle.",
    "### The production line \u2014 spring 2026"
   ],
   "access_type_evidence": [
    "In just eight weeks, Chinese AI labs launched four frontier-class open-weight models, demonstrating an accelerated development pace.",
    "### The production line \u2014 spring 2026"
   ],
   "license_evidence": null,
   "significance_evidence": [
    "### The production line \u2014 spring 2026"
   ]
  },
  {
   "url": "https://nanomachines.net/business-markets/the-timeline-of-china-s-four-frontier-ai-models-launch-in-record-time",
   "title": "The Timeline Of China\u2019s Four Frontier AI Models Launch In Record Time - NanoMachines",
   "published_at": "2026-07-15T06:19:58",
   "organization": "Z.ai",
   "model": "GLM-5.2",
   "release_date": "2026-06-13",
   "access_type": "open-weight",
   "license": "MIT",
   "significance": "753B MoE, MIT, top open-weight on Artificial Analysis index",
   "organization_evidence": [
    "Chinese organizations such as Z.ai, Moonshot, and Alibaba have each introduced models targeting different aspects of AI performance, including cost efficiency, long-term stability, and accessibility.",
    "### The production line \u2014 spring 2026"
   ],
   "model_evidence": [
    "### The production line \u2014 spring 2026"
   ],
   "release_date_evidence": [
    "Starting with DeepSeek V4 on April 24, followed by MiniMax M3 on June 1, and then Kimi K2.7-Code and GLM-5.2 within days of each other in mid-June, Chinese laboratories have exhibited a rapid development cycle.",
    "### The production line \u2014 spring 2026"
   ],
   "access_type_evidence": [
    "In just eight weeks, Chinese AI labs launched four frontier-class open-weight models, demonstrating an accelerated development pace.",
    "### The production line \u2014 spring 2026"
   ],
   "license_evidence": [
    "### The production line \u2014 spring 2026"
   ],
   "significance_evidence": [
    "### The production line \u2014 spring 2026"
   ]
  },
  {
   "url": "https://forkast.news/thinking-machines-ships-inkling-small-open-weights-competition-now-has-a-us-entrant",
   "title": "Thinking Machines Ships Inkling-Small: Open-Weights Competition Now Has a US Entrant",
   "published_at": "2026-08-01T13:42:24",
   "organization": "Thinking Machines",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "A 276B-parameter open-weights Mixture-of-Experts model that matches its larger sibling on agentic benchmarks while competing on price.",
   "organization_evidence": [
    "Analysis",
    "# Thinking Machines Ships Inkling-Small: Open-Weights Competition Now Has a US Entrant"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Analysis",
    "# Thinking Machines Ships Inkling-Small: Open-Weights Competition Now Has a US Entrant"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Analysis",
    "# Thinking Machines Ships Inkling-Small: Open-Weights Competition Now Has a US Entrant",
    "Mira Murati's lab releases a 276B-parameter open-weights model that matches its larger sibling on agentic benchmarks \u2014 and enters a pricing war dominated by Chinese labs.",
    "The model is a Mixture-of-Experts architecture with 276 billion total parameters but only 12 billion active at inference, a design that keeps compute costs low while preserving the reasoning depth of its larger sibling, Inkling (975B total, 41B active)."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/glm-5-1-non-reasoning-vs-claude-sonnet-4-6",
   "title": "GLM-5.1 (Non-reasoning) vs Claude Sonnet 4.6 (Non-reasoning, High Effort): Model Comparison",
   "published_at": null,
   "organization": "Z AI",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": null,
   "organization_evidence": [
    "GLM-5.1 (Non-reasoning) vs. Claude Sonnet 4.6 (Non-reasoning, High Effort)",
    "| Creator | ![Z AI](/img/logos/zai_small.svg)[](https://z.ai/) | ![Anthropic](/img/logos/anthropic_small.svg)[](https://www.anthropic.com/) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "GLM-5.1 (Non-reasoning) vs. Claude Sonnet 4.6 (Non-reasoning, High Effort)",
    "| Release Date | April, 2026 | February, 2026 | GLM-5.1 (Non-reasoning) has a more recent release date than Claude Sonnet 4.6 (Non-reasoning, High Effort) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": null
  },
  {
   "url": "https://renovateqr.com/blog/chinese-ai-models-guide",
   "title": "Chinese AI Models in 2026: DeepSeek V4, Kimi K3, GLM-5.2, Qwen 3.8, and the Full US-China Gap Analysis",
   "published_at": "2026-08-03T00:00:00",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-27",
   "access_type": null,
   "license": null,
   "significance": "largest open weight model ever shipped, beating Claude Opus 4.8 and GPT-5.5 on several coding and agent benchmarks",
   "organization_evidence": [
    "Chinese AI Models in 2026: The Complete Guide",
    "Moonshot's Kimi K3, a 2.8 trillion parameter model released July 27, then took the title as the largest open weight model ever shipped, beating Claude Opus 4.8 and GPT-5.5 on several coding and agent benchmarks.",
    "Moonshot AI's Kimi K3 is the single most consequential release covered in this guide, not because it is unambiguously the best Chinese model, but because of how directly it challenged assumptions that had held through most of the year."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Chinese AI Models in 2026: The Complete Guide",
    "Moonshot's Kimi K3, a 2.8 trillion parameter model released July 27, then took the title as the largest open weight model ever shipped, beating Claude Opus 4.8 and GPT-5.5 on several coding and agent benchmarks.",
    "The company announced the model July 17 and released downloadable weights July 27, allowing any developer to run it on their own infrastructure at no licensing cost."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Chinese AI Models in 2026: The Complete Guide",
    "Moonshot's Kimi K3, a 2.8 trillion parameter model released July 27, then took the title as the largest open weight model ever shipped, beating Claude Opus 4.8 and GPT-5.5 on several coding and agent benchmarks."
   ]
  },
  {
   "url": "https://renovateqr.com/blog/chinese-ai-models-guide",
   "title": "Chinese AI Models in 2026: DeepSeek V4, Kimi K3, GLM-5.2, Qwen 3.8, and the Full US-China Gap Analysis",
   "published_at": "2026-08-03T00:00:00",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "1.6 trillion total parameters with 49 billion active per token, Hybrid Attention design cutting long-context compute by roughly 73%",
   "organization_evidence": [
    "Chinese AI Models in 2026: The Complete Guide",
    "DeepSeek V4 Pro launched in April at $1.74 per million input tokens, then dropped to a permanent $0.435 input / $0.87 output rate on May 22.",
    "DeepSeek V4 Pro launched April 24, 2026, the same day OpenAI shipped GPT-5.5, at 1.6 trillion total parameters with 49 billion active per token."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Chinese AI Models in 2026: The Complete Guide",
    "DeepSeek V4 Pro launched April 24, 2026, the same day OpenAI shipped GPT-5.5, at 1.6 trillion total parameters with 49 billion active per token."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Chinese AI Models in 2026: The Complete Guide",
    "DeepSeek V4 Pro launched April 24, 2026, the same day OpenAI shipped GPT-5.5, at 1.6 trillion total parameters with 49 billion active per token.",
    "DeepSeek V4 uses a Hybrid Attention design combining Compressed Sparse Attention and Heavily Compressed Attention to cut long-context compute by roughly 73%."
   ]
  },
  {
   "url": "https://renovateqr.com/blog/chinese-ai-models-guide",
   "title": "Chinese AI Models in 2026: DeepSeek V4, Kimi K3, GLM-5.2, Qwen 3.8, and the Full US-China Gap Analysis",
   "published_at": "2026-08-03T00:00:00",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "284 billion total with 13 billion active, running as one of the cheapest capable coding models available anywhere",
   "organization_evidence": [
    "Chinese AI Models in 2026: The Complete Guide",
    "V4 Flash runs at $0.14/$0.28, making it one of the cheapest capable coding models available anywhere."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Chinese AI Models in 2026: The Complete Guide",
    "DeepSeek V4 Pro launched April 24, 2026, the same day OpenAI shipped GPT-5.5, at 1.6 trillion total parameters with 49 billion active per token."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Chinese AI Models in 2026: The Complete Guide",
    "V4 Flash runs at $0.14/$0.28, making it one of the cheapest capable coding models available anywhere."
   ]
  },
  {
   "url": "https://renovateqr.com/blog/chinese-ai-models-guide",
   "title": "Chinese AI Models in 2026: DeepSeek V4, Kimi K3, GLM-5.2, Qwen 3.8, and the Full US-China Gap Analysis",
   "published_at": "2026-08-03T00:00:00",
   "organization": "Z.ai (Zhipu AI)",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "753 billion parameter Mixture-of-Experts model with 40 billion active parameters per token and IndexShare sparse attention mechanism",
   "organization_evidence": [
    "Chinese AI Models in 2026: The Complete Guide",
    "GLM-5.2 from Z.ai led open weight coding benchmarks after its June 13 launch.",
    "Zhipu AI, now operating under the Z.ai brand, released GLM-5.2 on June 13, 2026, first through a Coding Plan subscription and then via standalone API on June 16."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Chinese AI Models in 2026: The Complete Guide",
    "GLM-5.2 from Z.ai led open weight coding benchmarks after its June 13 launch.",
    "Zhipu AI, now operating under the Z.ai brand, released GLM-5.2 on June 13, 2026, first through a Coding Plan subscription and then via standalone API on June 16."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Chinese AI Models in 2026: The Complete Guide",
    "GLM-5.2: Z.ai's Long-Horizon Coding Specialist",
    "The architecture is a 753 billion parameter Mixture-of-Experts model with 40 billion active parameters per token, a 1 million token context window, and MIT licensing with weights published on Hugging Face."
   ]
  },
  {
   "url": "https://rohan-paul.com/p/mira-muratis-thinking-machines-lab",
   "title": "\ud83d\uddde\ufe0f Mira Murati's Thinking Machines Lab drops massive open-weight AI model, Inkling, Apache 2.0 license, no restrictions.",
   "published_at": "2026-07-16T22:12:27",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-16",
   "access_type": null,
   "license": null,
   "significance": "A 975B-parameter (41B active) open-weights model with multimodal reasoning, adjustable effort, and a 1M-token context window",
   "organization_evidence": [
    "# \ud83d\uddde\ufe0f Mira Murati's Thinking Machines Lab drops massive open-weight AI model, Inkling, Apache 2.0 license, no restrictions.",
    "### \ud83d\uddde\ufe0f Mira Murati's Thinking Machines Lab released Inkling, Apache 2.0 license, no restrictions."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "### **\u26a1In today\u2019s Edition (16-July-2026):**",
    "# \ud83d\uddde\ufe0f Mira Murati's Thinking Machines Lab drops massive open-weight AI model, Inkling, Apache 2.0 license, no restrictions."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "A 975B-parameter (41B active) open-weights model with multimodal reasoning and adjustable effort.",
    "# \ud83d\uddde\ufe0f Mira Murati's Thinking Machines Lab drops massive open-weight AI model, Inkling, Apache 2.0 license, no restrictions."
   ]
  },
  {
   "url": "https://buildfastwithai.com/blogs/longcat-2-0-review-meituan-open-source-coding-model-2026",
   "title": "LongCat-2.0 Review: Meituan's Open-Source Coding Model Tested (2026)",
   "published_at": "2026-07-11T12:27:25",
   "organization": "Meituan",
   "model": null,
   "release_date": "2026-06-30",
   "access_type": null,
   "license": null,
   "significance": "1.6 trillion total parameters in a Mixture-of-Experts architecture with a native 1-million-token context window",
   "organization_evidence": [
    "LongCat-2.0 Review: Meituan's Open-Source Coding Model Tested (2026)",
    "# LongCat-2.0 Review: Meituan's Open-Source Coding Model Tested (2026)",
    "On June 30, 2026, Meituan stepped forward to claim it."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "LongCat-2.0 Review: Meituan's Open-Source Coding Model Tested (2026)",
    "# LongCat-2.0 Review: Meituan's Open-Source Coding Model Tested (2026)",
    "On June 30, 2026, Meituan stepped forward to claim it."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "LongCat-2.0 Review: Meituan's Open-Source Coding Model Tested (2026)",
    "# LongCat-2.0 Review: Meituan's Open-Source Coding Model Tested (2026)",
    "The model's specifications match what two months of production usage had already demonstrated: 1.6 trillion total parameters in a Mixture-of-Experts architecture, approximately 48 billion active per token, a native 1-million-token context window built on a custom attention mechanism called LongCat Sparse Attention, and an MIT license that makes it one of the most permissive trillion-parameter models ever released."
   ]
  },
  {
   "url": "https://ventureatlas.org/news/2026-07-27-moonshot-kimi-k3-open-weights-release",
   "title": "Moonshot releases Kimi K3 open weights, the largest ever",
   "published_at": "2026-07-27T00:00:00",
   "organization": "Moonshot AI",
   "model": "Kimi K3",
   "release_date": "2026-07-27",
   "access_type": "open weights",
   "license": "modified MIT license",
   "significance": "the roughly 2.8-trillion-parameter mixture-of-experts system the largest open-weight model ever published",
   "organization_evidence": [
    "# Moonshot releases Kimi K3 open weights, the largest ever",
    "Chinese AI lab Moonshot AI released the full open weights of its Kimi K3 model on July 27, making the roughly 2.8-trillion-parameter mixture-of-experts system the largest open-weight model ever published, under a modified MIT license on Hugging Face."
   ],
   "model_evidence": [
    "# Moonshot releases Kimi K3 open weights, the largest ever",
    "Chinese AI lab Moonshot AI released the full open weights of its Kimi K3 model on July 27, making the roughly 2.8-trillion-parameter mixture-of-experts system the largest open-weight model ever published, under a modified MIT license on Hugging Face."
   ],
   "release_date_evidence": [
    "Monday, July 27, 2026",
    "# Moonshot releases Kimi K3 open weights, the largest ever",
    "Chinese AI lab Moonshot AI released the full open weights of its Kimi K3 model on July 27, making the roughly 2.8-trillion-parameter mixture-of-experts system the largest open-weight model ever published, under a modified MIT license on Hugging Face."
   ],
   "access_type_evidence": [
    "# Moonshot releases Kimi K3 open weights, the largest ever",
    "Chinese AI lab Moonshot AI released the full open weights of its Kimi K3 model on July 27, making the roughly 2.8-trillion-parameter mixture-of-experts system the largest open-weight model ever published, under a modified MIT license on Hugging Face."
   ],
   "license_evidence": [
    "# Moonshot releases Kimi K3 open weights, the largest ever",
    "Chinese AI lab Moonshot AI released the full open weights of its Kimi K3 model on July 27, making the roughly 2.8-trillion-parameter mixture-of-experts system the largest open-weight model ever published, under a modified MIT license on Hugging Face."
   ],
   "significance_evidence": [
    "# Moonshot releases Kimi K3 open weights, the largest ever",
    "Chinese AI lab Moonshot AI released the full open weights of its Kimi K3 model on July 27, making the roughly 2.8-trillion-parameter mixture-of-experts system the largest open-weight model ever published, under a modified MIT license on Hugging Face."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/gpt-5-6-sol-medium-vs-gemma-4-31b",
   "title": "GPT-5.6 Sol (medium) vs Gemma 4 31B (Reasoning): Model Comparison | Artificial Analysis",
   "published_at": null,
   "organization": "Google",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "Reasoning and image input support",
   "organization_evidence": [
    "# GPT-5.6 Sol (medium) vs. Gemma 4 31B (Reasoning)",
    "| | ![OpenAI logo](/img/logos/openai_small.svg)OpenAI GPT-5.6 Sol (medium) | ![Google logo](/img/logos/google_small.svg)Google Gemma 4 31B (Reasoning) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# GPT-5.6 Sol (medium) vs. Gemma 4 31B (Reasoning)",
    "| Release Date | July 2026 | April 2026 | GPT-5.6 Sol (medium) has a more recent release date than Gemma 4 31B (Reasoning) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# GPT-5.6 Sol (medium) vs. Gemma 4 31B (Reasoning)",
    "| Reasoning | Yes | Yes | Both GPT-5.6 Sol (medium) and Gemma 4 31B (Reasoning) have reasoning |",
    "| Image Input Support | Yes | Yes | Both GPT-5.6 Sol (medium) and Gemma 4 31B (Reasoning) have image input support |"
   ]
  },
  {
   "url": "https://creativeainews.com/articles/thinking-machines-inkling-open-weights-multimodal-2026",
   "title": "Inkling: Thinking Machines Open Multimodal Model",
   "published_at": "2026-07-17T15:11:20",
   "organization": "Thinking Machines",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "975B-parameter Mixture-of-Experts model that reads text, images, and audio",
   "organization_evidence": [
    "Deep Dive",
    "# Inkling: Thinking Machines' Open Multimodal AI",
    "Thinking Machines released Inkling, a 975B-parameter open-weights Mixture-of-Experts model that reads text, images, and audio."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Deep Dive",
    "# Inkling: Thinking Machines' Open Multimodal AI",
    "[Inkling](https://thinkingmachines.ai/inkling/?ref=creativeainews.com) is a 975-billion-parameter Mixture-of-Experts model that reads text, images, and audio, and it shipped on July 15, 2026 with open weights on Hugging Face and a one-click path to fine-tuning."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Deep Dive",
    "# Inkling: Thinking Machines' Open Multimodal AI",
    "Thinking Machines released Inkling, a 975B-parameter open-weights Mixture-of-Experts model that reads text, images, and audio."
   ]
  },
  {
   "url": "https://ailearningguides.com/kimi-k3-2026-open-source-model",
   "title": "Kimi K3 2026: The 2.8-Trillion Open-Source AI Giant",
   "published_at": "2026-07-17T18:14:38",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-16",
   "access_type": null,
   "license": null,
   "significance": "A 2.8-trillion-parameter Mixture-of-Experts model with frontier-tier benchmark scores, 256K native context, and Kimi Delta Attention.",
   "organization_evidence": [
    "# Kimi K3 2026: The 2.8-Trillion Open-Source AI Giant",
    "Moonshot AI just moved the open-weight goalposts again."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Kimi K3 2026: The 2.8-Trillion Open-Source AI Giant",
    "On July 16, 2026, the Beijing lab released **Kimi K3**, a 2.8-trillion-parameter Mixture-of-Experts model you can download, self-host, and run without paying a per-token toll to anyone."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Kimi K3 2026: The 2.8-Trillion Open-Source AI Giant",
    "On July 16, 2026, the Beijing lab released **Kimi K3**, a 2.8-trillion-parameter Mixture-of-Experts model you can download, self-host, and run without paying a per-token toll to anyone.",
    "It is the largest open-weight model shipped this year, it posts frontier-tier benchmark scores, and it runs on a new attention design \u2014 Kimi Delta Attention \u2014 that makes serving a model this size far less painful than the parameter count suggests."
   ]
  },
  {
   "url": "https://murakumo.cloud/blog/2026-07-llm-landscape-frontier-open-weight-edge",
   "title": "State of LLMs, July 2026: frontier, open-weight, and what actually fits on your own hardware \u2014 murakumo.cloud",
   "published_at": null,
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-16",
   "access_type": null,
   "license": null,
   "significance": "Largest open-weight model to date; #1 on Arena's blind Frontend Code eval, ahead of Fable 5",
   "organization_evidence": [
    "# State of LLMs, July 2026: frontier, open-weight, and what actually fits on your own hardware",
    "Open-weight (self-hostable, if you have the hardware)",
    "| Model | Vendor | Size | License | Notable |",
    "| Kimi K3 | Moonshot AI | 2.8T total | Modified MIT | Announced 2026-07-16, largest open-weight model to date; full weights promised 2026-07-27; #1 on Arena's blind Frontend Code eval, ahead of Fable 5 |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# State of LLMs, July 2026: frontier, open-weight, and what actually fits on your own hardware",
    "Open-weight (self-hostable, if you have the hardware)",
    "| Model | Vendor | Size | License | Notable |",
    "| Kimi K3 | Moonshot AI | 2.8T total | Modified MIT | Announced 2026-07-16, largest open-weight model to date; full weights promised 2026-07-27; #1 on Arena's blind Frontend Code eval, ahead of Fable 5 |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# State of LLMs, July 2026: frontier, open-weight, and what actually fits on your own hardware",
    "Open-weight (self-hostable, if you have the hardware)",
    "| Model | Vendor | Size | License | Notable |",
    "| Kimi K3 | Moonshot AI | 2.8T total | Modified MIT | Announced 2026-07-16, largest open-weight model to date; full weights promised 2026-07-27; #1 on Arena's blind Frontend Code eval, ahead of Fable 5 |"
   ]
  },
  {
   "url": "https://mlq.ai/news/moonshot-ai-releases-kimi-k3-a-28-trillion-parameter-open-weight-model-rivaling-top-us-systems",
   "title": "Moonshot AI Releases Kimi K3, a 2.8-Trillion-Parameter Open-Weight Model Rivaling Top U.S. Systems",
   "published_at": "2026-07-17T16:29:26",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-16",
   "access_type": null,
   "license": null,
   "significance": "A 2.8-trillion-parameter sparse mixture-of-experts model that is the largest open-weight AI system ever built, rivaling top U.S. systems.",
   "organization_evidence": [
    "# Moonshot AI Releases Kimi K3, a 2.8-Trillion-Parameter Open-Weight Model Rivaling Top U.S. Systems",
    "Chinese AI startup Moonshot AI released Kimi K3 on July 16, a 2.8-trillion-parameter sparse mixture-of-experts model that the company says is the largest open-weight AI system ever built."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Moonshot AI Releases Kimi K3, a 2.8-Trillion-Parameter Open-Weight Model Rivaling Top U.S. Systems",
    "Jul 17, 2026 \u00b7 12:29 PM \u00b7 by MLQ Agent \u00b7 5 min read",
    "Chinese AI startup Moonshot AI released Kimi K3 on July 16, a 2.8-trillion-parameter sparse mixture-of-experts model that the company says is the largest open-weight AI system ever built."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Moonshot AI Releases Kimi K3, a 2.8-Trillion-Parameter Open-Weight Model Rivaling Top U.S. Systems",
    "Chinese AI startup Moonshot AI released Kimi K3 on July 16, a 2.8-trillion-parameter sparse mixture-of-experts model that the company says is the largest open-weight AI system ever built."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/glm-5-1-non-reasoning-vs-qwen3-5-27b-non-reasoning",
   "title": "GLM-5.1 (Non-reasoning) vs Qwen3.5 27B (Non-reasoning): Model Comparison",
   "published_at": null,
   "organization": "Z",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "744B, 40B active at inference time",
   "organization_evidence": [
    "| Creator | ![Z AI](/img/logos/zai_small.svg)[](https://z.ai/) | ![Alibaba](/img/logos/alibaba_small.svg)[](https://www.alibabacloud.com/) | |",
    "# GLM-5.1 (Non-reasoning) vs. Qwen3.5 27B (Non-reasoning)",
    "## Model Comparison",
    "| Metric | ![Z AI logo](/img/logos/zai_small.svg)GLM-5.1 (Non-reasoning) | ![Alibaba logo](/img/logos/alibaba_small.svg)Qwen3.5 27B (Non-reasoning) | Analysis |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "| Release Date | April, 2026 | February, 2026 | GLM-5.1 (Non-reasoning) has a more recent release date than Qwen3.5 27B (Non-reasoning) |",
    "# GLM-5.1 (Non-reasoning) vs. Qwen3.5 27B (Non-reasoning)",
    "## Model Comparison",
    "| Metric | ![Z AI logo](/img/logos/zai_small.svg)GLM-5.1 (Non-reasoning) | ![Alibaba logo](/img/logos/alibaba_small.svg)Qwen3.5 27B (Non-reasoning) | Analysis |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "| Parameters | 744B, 40B active at inference time | 27.8B | GLM-5.1 (Non-reasoning) is larger than Qwen3.5 27B (Non-reasoning) |",
    "# GLM-5.1 (Non-reasoning) vs. Qwen3.5 27B (Non-reasoning)",
    "## Model Comparison",
    "| Metric | ![Z AI logo](/img/logos/zai_small.svg)GLM-5.1 (Non-reasoning) | ![Alibaba logo](/img/logos/alibaba_small.svg)Qwen3.5 27B (Non-reasoning) | Analysis |"
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/glm-5-1-non-reasoning-vs-qwen3-5-27b-non-reasoning",
   "title": "GLM-5.1 (Non-reasoning) vs Qwen3.5 27B (Non-reasoning): Model Comparison",
   "published_at": null,
   "organization": "Z",
   "model": "GLM-5.1 (Non-reasoning)",
   "release_date": "April, 2026",
   "access_type": "Yes",
   "license": "Mit",
   "significance": "744B, 40B active at inference time",
   "organization_evidence": [
    "| Creator | ![Z AI](/img/logos/zai_small.svg)[](https://z.ai/) | ![Alibaba](/img/logos/alibaba_small.svg)[](https://www.alibabacloud.com/) | |",
    "# GLM-5.1 (Non-reasoning) vs. Qwen3.5 27B (Non-reasoning)",
    "## Model Comparison",
    "| Metric | ![Z AI logo](/img/logos/zai_small.svg)GLM-5.1 (Non-reasoning) | ![Alibaba logo](/img/logos/alibaba_small.svg)Qwen3.5 27B (Non-reasoning) | Analysis |"
   ],
   "model_evidence": [
    "# GLM-5.1 (Non-reasoning) vs. Qwen3.5 27B (Non-reasoning)",
    "## Model Comparison",
    "| Metric | ![Z AI logo](/img/logos/zai_small.svg)GLM-5.1 (Non-reasoning) | ![Alibaba logo](/img/logos/alibaba_small.svg)Qwen3.5 27B (Non-reasoning) | Analysis |"
   ],
   "release_date_evidence": [
    "| Release Date | April, 2026 | February, 2026 | GLM-5.1 (Non-reasoning) has a more recent release date than Qwen3.5 27B (Non-reasoning) |",
    "# GLM-5.1 (Non-reasoning) vs. Qwen3.5 27B (Non-reasoning)",
    "## Model Comparison",
    "| Metric | ![Z AI logo](/img/logos/zai_small.svg)GLM-5.1 (Non-reasoning) | ![Alibaba logo](/img/logos/alibaba_small.svg)Qwen3.5 27B (Non-reasoning) | Analysis |"
   ],
   "access_type_evidence": [
    "| Open Source (Weights) | Yes[](https://huggingface.co/zai-org/GLM-5.1) | Yes[](https://huggingface.co/Qwen/Qwen3.5-27B) | Both GLM-5.1 (Non-reasoning) and Qwen3.5 27B (Non-reasoning) are open source |",
    "# GLM-5.1 (Non-reasoning) vs. Qwen3.5 27B (Non-reasoning)",
    "## Model Comparison",
    "| Metric | ![Z AI logo](/img/logos/zai_small.svg)GLM-5.1 (Non-reasoning) | ![Alibaba logo](/img/logos/alibaba_small.svg)Qwen3.5 27B (Non-reasoning) | Analysis |"
   ],
   "license_evidence": [
    "| License | Mit[](https://opensource.org/license/mit) | Apache 2.0[](https://huggingface.co/Qwen/Qwen3.5-122B-A10B/blob/main/LICENSE) | |",
    "# GLM-5.1 (Non-reasoning) vs. Qwen3.5 27B (Non-reasoning)",
    "## Model Comparison",
    "| Metric | ![Z AI logo](/img/logos/zai_small.svg)GLM-5.1 (Non-reasoning) | ![Alibaba logo](/img/logos/alibaba_small.svg)Qwen3.5 27B (Non-reasoning) | Analysis |"
   ],
   "significance_evidence": [
    "| Parameters | 744B, 40B active at inference time | 27.8B | GLM-5.1 (Non-reasoning) is larger than Qwen3.5 27B (Non-reasoning) |",
    "# GLM-5.1 (Non-reasoning) vs. Qwen3.5 27B (Non-reasoning)",
    "## Model Comparison",
    "| Metric | ![Z AI logo](/img/logos/zai_small.svg)GLM-5.1 (Non-reasoning) | ![Alibaba logo](/img/logos/alibaba_small.svg)Qwen3.5 27B (Non-reasoning) | Analysis |"
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/glm-5-1-non-reasoning-vs-qwen3-5-27b-non-reasoning",
   "title": "GLM-5.1 (Non-reasoning) vs Qwen3.5 27B (Non-reasoning): Model Comparison",
   "published_at": null,
   "organization": "Alibaba",
   "model": "Qwen3.5 27B (Non-reasoning)",
   "release_date": "February, 2026",
   "access_type": "Yes",
   "license": "Apache 2.0",
   "significance": "27.8B",
   "organization_evidence": [
    "| Creator | ![Z AI](/img/logos/zai_small.svg)[](https://z.ai/) | ![Alibaba](/img/logos/alibaba_small.svg)[](https://www.alibabacloud.com/) | |",
    "# GLM-5.1 (Non-reasoning) vs. Qwen3.5 27B (Non-reasoning)",
    "## Model Comparison",
    "| Metric | ![Z AI logo](/img/logos/zai_small.svg)GLM-5.1 (Non-reasoning) | ![Alibaba logo](/img/logos/alibaba_small.svg)Qwen3.5 27B (Non-reasoning) | Analysis |"
   ],
   "model_evidence": [
    "# GLM-5.1 (Non-reasoning) vs. Qwen3.5 27B (Non-reasoning)",
    "## Model Comparison",
    "| Metric | ![Z AI logo](/img/logos/zai_small.svg)GLM-5.1 (Non-reasoning) | ![Alibaba logo](/img/logos/alibaba_small.svg)Qwen3.5 27B (Non-reasoning) | Analysis |"
   ],
   "release_date_evidence": [
    "| Release Date | April, 2026 | February, 2026 | GLM-5.1 (Non-reasoning) has a more recent release date than Qwen3.5 27B (Non-reasoning) |",
    "# GLM-5.1 (Non-reasoning) vs. Qwen3.5 27B (Non-reasoning)",
    "## Model Comparison",
    "| Metric | ![Z AI logo](/img/logos/zai_small.svg)GLM-5.1 (Non-reasoning) | ![Alibaba logo](/img/logos/alibaba_small.svg)Qwen3.5 27B (Non-reasoning) | Analysis |"
   ],
   "access_type_evidence": [
    "| Open Source (Weights) | Yes[](https://huggingface.co/zai-org/GLM-5.1) | Yes[](https://huggingface.co/Qwen/Qwen3.5-27B) | Both GLM-5.1 (Non-reasoning) and Qwen3.5 27B (Non-reasoning) are open source |",
    "# GLM-5.1 (Non-reasoning) vs. Qwen3.5 27B (Non-reasoning)",
    "## Model Comparison",
    "| Metric | ![Z AI logo](/img/logos/zai_small.svg)GLM-5.1 (Non-reasoning) | ![Alibaba logo](/img/logos/alibaba_small.svg)Qwen3.5 27B (Non-reasoning) | Analysis |"
   ],
   "license_evidence": [
    "| License | Mit[](https://opensource.org/license/mit) | Apache 2.0[](https://huggingface.co/Qwen/Qwen3.5-122B-A10B/blob/main/LICENSE) | |",
    "# GLM-5.1 (Non-reasoning) vs. Qwen3.5 27B (Non-reasoning)",
    "## Model Comparison",
    "| Metric | ![Z AI logo](/img/logos/zai_small.svg)GLM-5.1 (Non-reasoning) | ![Alibaba logo](/img/logos/alibaba_small.svg)Qwen3.5 27B (Non-reasoning) | Analysis |"
   ],
   "significance_evidence": [
    "| Parameters | 744B, 40B active at inference time | 27.8B | GLM-5.1 (Non-reasoning) is larger than Qwen3.5 27B (Non-reasoning) |",
    "# GLM-5.1 (Non-reasoning) vs. Qwen3.5 27B (Non-reasoning)",
    "## Model Comparison",
    "| Metric | ![Z AI logo](/img/logos/zai_small.svg)GLM-5.1 (Non-reasoning) | ![Alibaba logo](/img/logos/alibaba_small.svg)Qwen3.5 27B (Non-reasoning) | Analysis |"
   ]
  },
  {
   "url": "https://alphamatch.ai/blog/open-source-llm-comparison-blog-2026",
   "title": "The Open-Source LLM Revolution 2026: How Chinese Models Are Redefining AI Supremacy",
   "published_at": null,
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-02",
   "access_type": null,
   "license": null,
   "significance": "Established leadership across 19 benchmarks with performance comparable to leading models such as GPT-5.2-Thinking, Claude-Opus-4.5, and Gemini 3.",
   "organization_evidence": [
    "The Open-Source LLM Revolution 2026: How Chinese Models Are Redefining AI Supremacy",
    "## 2. Qwen 3: The Multilingual Master",
    "Alibaba's Qwen series continues to excel with the latest Qwen3 generation, which demonstrates performance comparable to leading models such as GPT-5.2-Thinking, Claude-Opus-4.5, and Gemini 3."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The Open-Source LLM Revolution 2026: How Chinese Models Are Redefining AI Supremacy",
    "## 2. Qwen 3: The Multilingual Master",
    "**Latest Model:** Qwen3-Max and Qwen3.5-Medium (Released February 2026)"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The Open-Source LLM Revolution 2026: How Chinese Models Are Redefining AI Supremacy",
    "## 2. Qwen 3: The Multilingual Master",
    "Alibaba's Qwen series continues to excel with the latest Qwen3 generation, which demonstrates performance comparable to leading models such as GPT-5.2-Thinking, Claude-Opus-4.5, and Gemini 3.",
    "- **Performance:** Established leadership across 19 benchmarks"
   ]
  },
  {
   "url": "https://alphamatch.ai/blog/open-source-llm-comparison-blog-2026",
   "title": "The Open-Source LLM Revolution 2026: How Chinese Models Are Redefining AI Supremacy",
   "published_at": null,
   "organization": "Zhipu AI",
   "model": null,
   "release_date": "2026",
   "access_type": null,
   "license": null,
   "significance": "Achieved 77.8 on SWE-bench Verified, outperforming Gemini 3 Pro (76.2) and approaching Claude Opus 4.6 (80.9).",
   "organization_evidence": [
    "The Open-Source LLM Revolution 2026: How Chinese Models Are Redefining AI Supremacy",
    "## 3. GLM-5: The Thinking Model",
    "Zhipu AI's GLM-5 represents a massive leap forward, scaling from 355B parameters (32B active) to 744B parameters (40B active), with pre-training data increased from 23T to a substantially larger dataset."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The Open-Source LLM Revolution 2026: How Chinese Models Are Redefining AI Supremacy",
    "## 3. GLM-5: The Thinking Model",
    "**Latest Model:** GLM-5 (745B MoE) (Released 2026)"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The Open-Source LLM Revolution 2026: How Chinese Models Are Redefining AI Supremacy",
    "## 3. GLM-5: The Thinking Model",
    "- **Performance:** Achieved 77.8 on SWE-bench Verified, outperforming Gemini 3 Pro (76.2) and approaching Claude Opus 4.6 (80.9)"
   ]
  },
  {
   "url": "https://alphamatch.ai/blog/open-source-llm-comparison-blog-2026",
   "title": "The Open-Source LLM Revolution 2026: How Chinese Models Are Redefining AI Supremacy",
   "published_at": null,
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026",
   "access_type": null,
   "license": null,
   "significance": "Scores 42 on the Artificial Analysis Intelligence Index, placing it well above average among comparable models, and is now considered on par with Opus 4.6.",
   "organization_evidence": [
    "The Open-Source LLM Revolution 2026: How Chinese Models Are Redefining AI Supremacy",
    "## 4. MiniMax M2.5: The Efficiency Champion",
    "MiniMax has focused on creating highly efficient models that deliver exceptional performance per parameter."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The Open-Source LLM Revolution 2026: How Chinese Models Are Redefining AI Supremacy",
    "## 4. MiniMax M2.5: The Efficiency Champion",
    "**Latest Model:** MiniMax-M2.5 (Released 2026)"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The Open-Source LLM Revolution 2026: How Chinese Models Are Redefining AI Supremacy",
    "## 4. MiniMax M2.5: The Efficiency Champion",
    "MiniMax-M2.5 scores 42 on the Artificial Analysis Intelligence Index, placing it well above average among comparable models, and is now considered on par with Opus 4.6."
   ]
  },
  {
   "url": "https://alphamatch.ai/blog/open-source-llm-comparison-blog-2026",
   "title": "The Open-Source LLM Revolution 2026: How Chinese Models Are Redefining AI Supremacy",
   "published_at": null,
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026",
   "access_type": null,
   "license": null,
   "significance": "Best-in-class for document analysis and summarization with new records for context window handling.",
   "organization_evidence": [
    "The Open-Source LLM Revolution 2026: How Chinese Models Are Redefining AI Supremacy",
    "## 6. Kimi K2.5: The Context King",
    "Moonshot AI's Kimi series has set new records for context window handling, supporting extended context lengths while maintaining high performance across various benchmarks."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The Open-Source LLM Revolution 2026: How Chinese Models Are Redefining AI Supremacy",
    "## 6. Kimi K2.5: The Context King",
    "**Latest Model:** Kimi K2.5 (2026)"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The Open-Source LLM Revolution 2026: How Chinese Models Are Redefining AI Supremacy",
    "## 6. Kimi K2.5: The Context King",
    "Moonshot AI's Kimi series has set new records for context window handling, supporting extended context lengths while maintaining high performance across various benchmarks.",
    "- **Unique Capability:** Best-in-class for document analysis and summarization"
   ]
  },
  {
   "url": "https://fxcrypto24.com/news/thinking-machines-lab-unveils-inkling-open-source-multimodal-ai-model",
   "title": "Thinking Machines Lab Unveils Inkling, a Fully Open-Source Multimodal AI Model Built From the Ground Up",
   "published_at": "2026-07-16T18:27:00",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "975-billion-parameter multimodal AI model trained entirely from scratch",
   "organization_evidence": [
    "# Thinking Machines Lab Unveils Inkling, a Fully Open-Source Multimodal AI Model Built From the Ground Up",
    "Thinking Machines Lab, the AI company founded by former OpenAI CTO Mira Murati, released its debut model on July 15."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Thinking Machines Lab Unveils Inkling, a Fully Open-Source Multimodal AI Model Built From the Ground Up",
    "Thinking Machines Lab, the AI company founded by former OpenAI CTO Mira Murati, released its debut model on July 15."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Thinking Machines Lab Unveils Inkling, a Fully Open-Source Multimodal AI Model Built From the Ground Up",
    "Called Inkling, it is a 975-billion-parameter multimodal AI model trained entirely from scratch, with full weights freely available on Hugging Face under an Apache 2.0 license."
   ]
  },
  {
   "url": "https://fazm.ai/t/new-open-source-llm-release-april-2026",
   "title": "Fazm - AI Computer Agent for macOS | Voice-Controlled Desktop Automation",
   "published_at": "2026-04-20T11:47:31",
   "organization": "Google",
   "model": null,
   "release_date": "2026-04-02",
   "access_type": null,
   "license": null,
   "significance": "Multimodal open weights in 2B/9B/27B sizes",
   "organization_evidence": [
    "What new open source LLMs shipped in April 2026?",
    "The most significant April 2026 open-source releases: Google Gemma 4 (Apache 2.0, April 2), Zhipu GLM-5.1 (MIT, 754B MoE with 37B active), Alibaba Qwen 3.6-Plus (1M-token context, Tongyi license), DeepSeek V3.2 (DeepSeek license, April 11), Mistral Small 3.2 (Apache 2.0, early April), and PrismML Bonsai 8B (Apache 2.0)."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "What new open source LLMs shipped in April 2026?",
    "The most significant April 2026 open-source releases: Google Gemma 4 (Apache 2.0, April 2), Zhipu GLM-5.1 (MIT, 754B MoE with 37B active), Alibaba Qwen 3.6-Plus (1M-token context, Tongyi license), DeepSeek V3.2 (DeepSeek license, April 11), Mistral Small 3.2 (Apache 2.0, early April), and PrismML Bonsai 8B (Apache 2.0)."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "### Google Gemma 4 (Apache 2.0, April 2)",
    "Multimodal open weights in 2B/9B/27B sizes."
   ]
  },
  {
   "url": "https://fazm.ai/t/new-open-source-llm-release-april-2026",
   "title": "Fazm - AI Computer Agent for macOS | Voice-Controlled Desktop Automation",
   "published_at": "2026-04-20T11:47:31",
   "organization": "Zhipu",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "754B MoE with 37B active",
   "organization_evidence": [
    "What new open source LLMs shipped in April 2026?",
    "The most significant April 2026 open-source releases: Google Gemma 4 (Apache 2.0, April 2), Zhipu GLM-5.1 (MIT, 754B MoE with 37B active), Alibaba Qwen 3.6-Plus (1M-token context, Tongyi license), DeepSeek V3.2 (DeepSeek license, April 11), Mistral Small 3.2 (Apache 2.0, early April), and PrismML Bonsai 8B (Apache 2.0)."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "What new open source LLMs shipped in April 2026?",
    "The most significant April 2026 open-source releases: Google Gemma 4 (Apache 2.0, April 2), Zhipu GLM-5.1 (MIT, 754B MoE with 37B active), Alibaba Qwen 3.6-Plus (1M-token context, Tongyi license), DeepSeek V3.2 (DeepSeek license, April 11), Mistral Small 3.2 (Apache 2.0, early April), and PrismML Bonsai 8B (Apache 2.0)."
   ]
  },
  {
   "url": "https://fazm.ai/t/new-open-source-llm-release-april-2026",
   "title": "Fazm - AI Computer Agent for macOS | Voice-Controlled Desktop Automation",
   "published_at": "2026-04-20T11:47:31",
   "organization": "Alibaba",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "1M-token context",
   "organization_evidence": [
    "What new open source LLMs shipped in April 2026?",
    "The most significant April 2026 open-source releases: Google Gemma 4 (Apache 2.0, April 2), Zhipu GLM-5.1 (MIT, 754B MoE with 37B active), Alibaba Qwen 3.6-Plus (1M-token context, Tongyi license), DeepSeek V3.2 (DeepSeek license, April 11), Mistral Small 3.2 (Apache 2.0, early April), and PrismML Bonsai 8B (Apache 2.0)."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "What new open source LLMs shipped in April 2026?",
    "The most significant April 2026 open-source releases: Google Gemma 4 (Apache 2.0, April 2), Zhipu GLM-5.1 (MIT, 754B MoE with 37B active), Alibaba Qwen 3.6-Plus (1M-token context, Tongyi license), DeepSeek V3.2 (DeepSeek license, April 11), Mistral Small 3.2 (Apache 2.0, early April), and PrismML Bonsai 8B (Apache 2.0)."
   ]
  },
  {
   "url": "https://fazm.ai/t/new-open-source-llm-release-april-2026",
   "title": "Fazm - AI Computer Agent for macOS | Voice-Controlled Desktop Automation",
   "published_at": "2026-04-20T11:47:31",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-11",
   "access_type": null,
   "license": null,
   "significance": "Chinese-lab flagship",
   "organization_evidence": [
    "What new open source LLMs shipped in April 2026?",
    "The most significant April 2026 open-source releases: Google Gemma 4 (Apache 2.0, April 2), Zhipu GLM-5.1 (MIT, 754B MoE with 37B active), Alibaba Qwen 3.6-Plus (1M-token context, Tongyi license), DeepSeek V3.2 (DeepSeek license, April 11), Mistral Small 3.2 (Apache 2.0, early April), and PrismML Bonsai 8B (Apache 2.0)."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "What new open source LLMs shipped in April 2026?",
    "The most significant April 2026 open-source releases: Google Gemma 4 (Apache 2.0, April 2), Zhipu GLM-5.1 (MIT, 754B MoE with 37B active), Alibaba Qwen 3.6-Plus (1M-token context, Tongyi license), DeepSeek V3.2 (DeepSeek license, April 11), Mistral Small 3.2 (Apache 2.0, early April), and PrismML Bonsai 8B (Apache 2.0)."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "### DeepSeek V3.2 (April 11)",
    "Chinese-lab flagship, DeepSeek license."
   ]
  },
  {
   "url": "https://fazm.ai/t/new-open-source-llm-release-april-2026",
   "title": "Fazm - AI Computer Agent for macOS | Voice-Controlled Desktop Automation",
   "published_at": "2026-04-20T11:47:31",
   "organization": "Mistral",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Smaller Apache 2.0 release",
   "organization_evidence": [
    "What new open source LLMs shipped in April 2026?",
    "The most significant April 2026 open-source releases: Google Gemma 4 (Apache 2.0, April 2), Zhipu GLM-5.1 (MIT, 754B MoE with 37B active), Alibaba Qwen 3.6-Plus (1M-token context, Tongyi license), DeepSeek V3.2 (DeepSeek license, April 11), Mistral Small 3.2 (Apache 2.0, early April), and PrismML Bonsai 8B (Apache 2.0)."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "### Mistral Small 3.2 + PrismML Bonsai 8B + Llama 4 fine-tunes",
    "Three smaller Apache 2.0 releases."
   ]
  },
  {
   "url": "https://fazm.ai/t/new-open-source-llm-release-april-2026",
   "title": "Fazm - AI Computer Agent for macOS | Voice-Controlled Desktop Automation",
   "published_at": "2026-04-20T11:47:31",
   "organization": "PrismML",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Smaller Apache 2.0 release",
   "organization_evidence": [
    "What new open source LLMs shipped in April 2026?",
    "The most significant April 2026 open-source releases: Google Gemma 4 (Apache 2.0, April 2), Zhipu GLM-5.1 (MIT, 754B MoE with 37B active), Alibaba Qwen 3.6-Plus (1M-token context, Tongyi license), DeepSeek V3.2 (DeepSeek license, April 11), Mistral Small 3.2 (Apache 2.0, early April), and PrismML Bonsai 8B (Apache 2.0)."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "### Mistral Small 3.2 + PrismML Bonsai 8B + Llama 4 fine-tunes",
    "Three smaller Apache 2.0 releases."
   ]
  },
  {
   "url": "https://synthszr.com/en/posts/mira-murati-releases-new-us-open-weight-model-better-than-nvidia-but-weaker-than-chinese-frontier-models",
   "title": "Mira Murati Releases New US Open-Weight Model \u2014 Better Than Nvidia but Weaker Than Chinese Frontier Models",
   "published_at": "2026-07-16T05:45:58",
   "organization": "Thinking Machines",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "a natively multimodal Mixture-of-Experts system with 975 billion total parameters (41 billion active) and processes text, images, and audio",
   "organization_evidence": [
    "Mira Murati Releases New US Open-Weight Model \u2014 Better Than Nvidia but Weaker Than Chinese Frontier Models",
    "Thinking Machines, the well-funded US startup from former OpenAI CTO Mira Murati, has released its first major language model, Inkling, under the Apache 2.0 license."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Mira Murati Releases New US Open-Weight Model \u2014 Better Than Nvidia but Weaker Than Chinese Frontier Models",
    "Thinking Machines, the well-funded US startup from former OpenAI CTO Mira Murati, has released its first major language model, Inkling, under the Apache 2.0 license.",
    "The model is a natively multimodal Mixture-of-Experts system with 975 billion total parameters (41 billion active) and processes text, images, and audio."
   ]
  },
  {
   "url": "https://techieus.com/general/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - TechieUS",
   "published_at": "2026-07-15T12:26:21",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "1.6T total / 49B active MoE, 1M context, MIT",
   "organization_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "The Chinese models now occupy prominent positions in the open-weight AI landscape, with four of the five most capable models originating from Chinese labs, including DeepSeek, Z.ai, Moonshot, and Alibaba.",
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "Between late April and mid-June 2026, Chinese labs launched four major open-weight AI models: DeepSeek V4 on April 24, MiniMax M3 on June 1, and Kimi K2.7-Code along with GLM-5.2 in mid-June.",
    "APR 24",
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor",
    "DeepSeek V4 emphasizes affordability with 1.6 trillion parameters but activates only 49 billion per pass"
   ]
  },
  {
   "url": "https://techieus.com/general/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - TechieUS",
   "published_at": "2026-07-15T12:26:21",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "cheap 1M-token context, native multimodal, modified-MIT",
   "organization_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "Between late April and mid-June 2026, Chinese labs launched four major open-weight AI models: DeepSeek V4 on April 24, MiniMax M3 on June 1, and Kimi K2.7-Code along with GLM-5.2 in mid-June.",
    "JUN 01"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks"
   ]
  },
  {
   "url": "https://techieus.com/general/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - TechieUS",
   "published_at": "2026-07-15T12:26:21",
   "organization": "Moonshot",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "agent-run specialist, ~30% fewer thinking tokens than K2.6",
   "organization_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "The Chinese models now occupy prominent positions in the open-weight AI landscape, with four of the five most capable models originating from Chinese labs, including DeepSeek, Z.ai, Moonshot, and Alibaba."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "Between late April and mid-June 2026, Chinese labs launched four major open-weight AI models: DeepSeek V4 on April 24, MiniMax M3 on June 1, and Kimi K2.7-Code along with GLM-5.2 in mid-June.",
    "JUN 13"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks"
   ]
  },
  {
   "url": "https://techieus.com/general/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - TechieUS",
   "published_at": "2026-07-15T12:26:21",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "753B MoE, MIT, top open-weight on Artificial Analysis index",
   "organization_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "The Chinese models now occupy prominent positions in the open-weight AI landscape, with four of the five most capable models originating from Chinese labs, including DeepSeek, Z.ai, Moonshot, and Alibaba."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "Between late April and mid-June 2026, Chinese labs launched four major open-weight AI models: DeepSeek V4 on April 24, MiniMax M3 on June 1, and Kimi K2.7-Code along with GLM-5.2 in mid-June.",
    "JUN 13\u201316"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks"
   ]
  },
  {
   "url": "https://developersdigest.tech/blog/inkling-small-open-weights-2026",
   "title": "Inkling-Small: Thinking Machines Ships a 12B-Active Open Model That Beats Its Big Sibling on Agent Work",
   "published_at": "2026-07-31T00:00:00",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-30",
   "access_type": null,
   "license": null,
   "significance": "A 276B-parameter MoE with 12B active per token that beats the 975B Inkling on SWEBench Verified and HLE at a quarter of the size",
   "organization_evidence": [
    "TL;DR",
    "On July 30, Thinking Machines Lab released Inkling-Small, an open-[weights](/glossary#weights) Mixture-of-Experts model with 276B total parameters and only 12B active per [token](/glossary#token)."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "TL;DR",
    "On July 30, Thinking Machines Lab released Inkling-Small, an open-[weights](/glossary#weights) Mixture-of-Experts model with 276B total parameters and only 12B active per [token](/glossary#token)."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "TL;DR",
    "Inkling-Small is a 276B-parameter MoE with 12B active per token, Apache 2.0, and open weights. It beats the 975B Inkling on SWEBench Verified (80.2), HLE (31.6), and tool use at a quarter of the size and a third of the output price."
   ]
  },
  {
   "url": "https://atomic.chat/blog/llm-updates/best-open-source-llm",
   "title": "Best Open Source LLM in 2026: 10 Models Ranked - Atomic Chat",
   "published_at": null,
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "One of the most popular LLMs for local deployment, running on consumer hardware",
   "organization_evidence": [
    "What are the best open source LLMs in 2026?",
    "| Qwen3.6 | Alibaba | 27B dense | 262K | Apache 2.0 | Running yourself |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "### 1. Qwen3.6",
    "Released April 2026, Qwen3.6 is an open weight model from Alibaba and one of the most popular LLMs for local deployment, as it runs on consumer hardware."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "### 1. Qwen3.6",
    "Released April 2026, Qwen3.6 is an open weight model from Alibaba and one of the most popular LLMs for local deployment, as it runs on consumer hardware."
   ]
  },
  {
   "url": "https://atomic.chat/blog/llm-updates/best-open-source-llm",
   "title": "Best Open Source LLM in 2026: 10 Models Ranked - Atomic Chat",
   "published_at": null,
   "organization": "Google DeepMind",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "Most-downloaded open family of 2026, supporting multi-modal inputs natively including audio",
   "organization_evidence": [
    "What are the best open source LLMs in 2026?",
    "| Gemma 4 | Google | 2B\u201331B | 256K | Apache 2.0 | Small models and laptops |",
    "### 2. Gemma 4",
    "Gemma 4 is another small open weights model you can run locally."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "### 2. Gemma 4",
    "It was developed by Google DeepMind and released in April 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "### 2. Gemma 4",
    "It's no surprise, then, that Gemma 4 became the most-downloaded open family of 2026.",
    "Notably this family has 2B and 4B variants that support multi-modal inputs natively, even including audio processing for up to 30 seconds of speech, as well as video and images."
   ]
  },
  {
   "url": "https://atomic.chat/blog/llm-updates/best-open-source-llm",
   "title": "Best Open Source LLM in 2026: 10 Models Ranked - Atomic Chat",
   "published_at": null,
   "organization": "Tencent",
   "model": null,
   "release_date": "2026-07-06",
   "access_type": null,
   "license": null,
   "significance": "Positioned to compete with trillion-parameter class models while fitting under 300GB in FP8",
   "organization_evidence": [
    "What are the best open source LLMs in 2026?",
    "| Hunyuan Hy3 | Tencent | 295B (21B active) | 256K | Apache 2.0 | Capability per gigabyte |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "### 4. Hunyuan Hy3",
    "Hunyuan Hy3 is an open model from Tencent released on July 6, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "### 4. Hunyuan Hy3",
    "It's a 295B parameter model, but it's positioned to compete with trillion-parameter class models while fitting under 300GB in FP8, and under 100GB in the official 4-bit quantization, and you can run it locally if you have more than 128GB of VRAM or unified memory."
   ]
  },
  {
   "url": "https://atomic.chat/blog/llm-updates/best-open-source-llm",
   "title": "Best Open Source LLM in 2026: 10 Models Ranked - Atomic Chat",
   "published_at": null,
   "organization": "Cohere",
   "model": null,
   "release_date": "2026-05",
   "access_type": null,
   "license": null,
   "significance": "Native citation generation built directly into the model weights",
   "organization_evidence": [
    "What are the best open source LLMs in 2026?",
    "| Command A+ | Cohere | 218B (25B active) | 128K | Apache 2.0 | Enterprise RAG |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "### 5. Command A+",
    "Command A+ is an open weights model released by Cohere in May 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "### 5. Command A+",
    "Its differentiating feature is native citation generation that is built directly into the model weights."
   ]
  },
  {
   "url": "https://atomic.chat/blog/llm-updates/best-open-source-llm",
   "title": "Best Open Source LLM in 2026: 10 Models Ranked - Atomic Chat",
   "published_at": null,
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-06",
   "access_type": null,
   "license": null,
   "significance": "The first open release that can replace Claude Opus-level LLMs in agentic workflows",
   "organization_evidence": [
    "What are the best open source LLMs in 2026?",
    "| GLM-5.2 | Z.ai | 753B (40B active) | 1M | MIT | Best overall |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "### 6. GLM-5.2",
    "GLM-5.2 is a 753B-parameter mixture-of-experts model from Z.ai, released in June 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "### 6. GLM-5.2",
    "Interconnects called it \"the step change for open models\" \u2014 the first open release that can replace Claude Opus-level LLMs in agentic workflows."
   ]
  },
  {
   "url": "https://atomic.chat/blog/llm-updates/best-open-source-llm",
   "title": "Best Open Source LLM in 2026: 10 Models Ranked - Atomic Chat",
   "published_at": null,
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "Most-deployed open LLM in production within a month of release because it made long context cheap to serve",
   "organization_evidence": [
    "What are the best open source LLMs in 2026?",
    "| DeepSeek-V4-Pro | DeepSeek | 1.6T (49B active) | 1M | MIT | Long context at scale |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "### 7. DeepSeek-V4-Pro",
    "DeepSeek-V4-Pro, released in April 2026, is the flagship model from DeepSeek."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "### 7. DeepSeek-V4-Pro",
    "V4-Pro became the most-deployed open LLM in production within a month of release because it made long context cheap to serve: processing a 1M-token context takes about 4\u00d7 less compute and 10\u00d7 less memory than on DeepSeek's previous generation"
   ]
  },
  {
   "url": "https://atomic.chat/blog/llm-updates/best-open-source-llm",
   "title": "Best Open Source LLM in 2026: 10 Models Ranked - Atomic Chat",
   "published_at": null,
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-16",
   "access_type": null,
   "license": null,
   "significance": "Most performant open weight model",
   "organization_evidence": [
    "What are the best open source LLMs in 2026?",
    "| Kimi K3 | Moonshot AI | 2.8T | 1M | Modified MIT | Frontier agentic work |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "### 8. Kimi K3",
    "Kimi K3 is Moonshot AI's 2.8-trillion-parameter flagship, launched on July 16, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "### 8. Kimi K3",
    "As of this writing it's the most performant open weight model, occupying the #3 position on the Artificial Analysis intelligence index, behind only Claude Fable 5 and GPT-5.6 Sol."
   ]
  },
  {
   "url": "https://atomic.chat/blog/llm-updates/best-open-source-llm",
   "title": "Best Open Source LLM in 2026: 10 Models Ranked - Atomic Chat",
   "published_at": null,
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-06",
   "access_type": null,
   "license": null,
   "significance": "Multimodal LLM that supports text, images, and video input, features a 1M-token context window",
   "organization_evidence": [
    "What are the best open source LLMs in 2026?",
    "| MiniMax-M3 | MiniMax | 428B (23B active) | 1M | Community | Multimodal + 1M context |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "### 9. MiniMax-M3",
    "MiniMax-M3 is the flagship model created by Shanghai-based MiniMax.",
    "Released | June 2026 |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "### 9. MiniMax-M3",
    "MiniMax-M3 is an open source multimodal LLM that supports text, images, and video input, features a 1M-token context window, and 23B active parameters."
   ]
  },
  {
   "url": "https://atomic.chat/blog/llm-updates/best-open-source-llm",
   "title": "Best Open Source LLM in 2026: 10 Models Ranked - Atomic Chat",
   "published_at": null,
   "organization": "NVIDIA",
   "model": null,
   "release_date": "2026-06",
   "access_type": null,
   "license": null,
   "significance": "Most advanced open weights model made outside of China, publicly releasing training datasets and post-training recipes",
   "organization_evidence": [
    "What are the best open source LLMs in 2026?",
    "| Nemotron 3 Ultra | NVIDIA | 550B (55B active) | 1M | OpenMDW-1.1 | Most complete open release |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "### 10. Nemotron 3 Ultra",
    "Released | June 2026 |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "### 10. Nemotron 3 Ultra",
    "Nemotron 3 Ultra is NVIDIA's flagship model.",
    "This is the most advanced open weights model made outside of China.",
    "It's also the most truly open source flagship-level model, as NVIDIA publicly released not only the weights but also:\n\n- the training datasets (over 3T tokens of them)\n- the post-training recipes and RL environments\n- the evaluation harness used to produce its scores"
   ]
  },
  {
   "url": "https://aimagazine.blog/kimi-k3-open-weights-are-live-what-the-worlds-largest-open-weight-model-actually-means",
   "title": "Download, Specs & What's New",
   "published_at": "2026-07-27T10:49:56",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-27",
   "access_type": null,
   "license": null,
   "significance": "2.8 trillion parameter open weights, making it the largest open-weight model publicly available with a 1 million token context window",
   "organization_evidence": [
    "# Kimi K3 Open Weights Are Now Live \u2014 Everything Confirmed About the World\u2019s Largest Open-Weight Model",
    "Moonshot AI has released its 2.8-trillion-parameter model for free download \u2014 confirmed today, July 27, 2026 \u2014 making it the largest open-weight model publicly available."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Kimi K3 Open Weights Are Now Live \u2014 Everything Confirmed About the World\u2019s Largest Open-Weight Model",
    "Moonshot AI has released its 2.8-trillion-parameter model for free download \u2014 confirmed today, July 27, 2026 \u2014 making it the largest open-weight model publicly available."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Kimi K3 Open Weights Are Now Live \u2014 Everything Confirmed About the World\u2019s Largest Open-Weight Model",
    "Moonshot AI has released its 2.8-trillion-parameter model for free download \u2014 confirmed today, July 27, 2026 \u2014 making it the largest open-weight model publicly available.",
    "The context window is 1 million tokens \u2014 meaning users can feed entire large codebases or lengthy documents into a single query."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/glm-5-2-vs-nvidia-nemotron-3-ultra-550b-a55b",
   "title": "GLM-5.2 (max) vs Nemotron 3 Ultra 550B A55B (Reasoning): Model Comparison | Artificial Analysis",
   "published_at": null,
   "organization": "Z AI",
   "model": null,
   "release_date": "2026-06",
   "access_type": null,
   "license": null,
   "significance": "753B parameters, 40B active at inference time, 1000k context window, Intelligence Index 53",
   "organization_evidence": [
    "Comparison between GLM-5.2 (max) and Nemotron 3 Ultra 550B A55B (Reasoning) across intelligence, price, speed, context window and more.",
    "| | ![Z AI logo](/img/logos/zai_small.svg)Z AI GLM-5.2 (max) | ![NVIDIA logo](/img/logos/nvidia_small.svg)NVIDIA Nemotron 3 Ultra 550B A55B (Reasoning) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Comparison between GLM-5.2 (max) and Nemotron 3 Ultra 550B A55B (Reasoning) across intelligence, price, speed, context window and more.",
    "| Release Date | June 2026 | June 2026 | GLM-5.2 (max) has a more recent release date than Nemotron 3 Ultra 550B A55B (Reasoning) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Comparison between GLM-5.2 (max) and Nemotron 3 Ultra 550B A55B (Reasoning) across intelligence, price, speed, context window and more.",
    "| Intelligence Index | 53 | 38 | GLM-5.2 (max) is more intelligent than Nemotron 3 Ultra 550B A55B (Reasoning) |",
    "| Context Window | 1000k tokens~1,500 A4 pages of size 12 Arial font | 262k tokens~393 A4 pages of size 12 Arial font | GLM-5.2 (max) has a larger context window than Nemotron 3 Ultra 550B A55B (Reasoning) |",
    "| Parameters | 753B, 40B active at inference time | 550B, 55B active at inference time | GLM-5.2 (max) has more parameters than Nemotron 3 Ultra 550B A55B (Reasoning) |"
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/glm-5-2-vs-nvidia-nemotron-3-ultra-550b-a55b",
   "title": "GLM-5.2 (max) vs Nemotron 3 Ultra 550B A55B (Reasoning): Model Comparison | Artificial Analysis",
   "published_at": null,
   "organization": "NVIDIA",
   "model": null,
   "release_date": "2026-06",
   "access_type": null,
   "license": null,
   "significance": "550B parameters, 55B active at inference time, 262k context window, Intelligence Index 38",
   "organization_evidence": [
    "Comparison between GLM-5.2 (max) and Nemotron 3 Ultra 550B A55B (Reasoning) across intelligence, price, speed, context window and more.",
    "| | ![Z AI logo](/img/logos/zai_small.svg)Z AI GLM-5.2 (max) | ![NVIDIA logo](/img/logos/nvidia_small.svg)NVIDIA Nemotron 3 Ultra 550B A55B (Reasoning) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Comparison between GLM-5.2 (max) and Nemotron 3 Ultra 550B A55B (Reasoning) across intelligence, price, speed, context window and more.",
    "| Release Date | June 2026 | June 2026 | GLM-5.2 (max) has a more recent release date than Nemotron 3 Ultra 550B A55B (Reasoning) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Comparison between GLM-5.2 (max) and Nemotron 3 Ultra 550B A55B (Reasoning) across intelligence, price, speed, context window and more.",
    "| Intelligence Index | 53 | 38 | GLM-5.2 (max) is more intelligent than Nemotron 3 Ultra 550B A55B (Reasoning) |",
    "| Context Window | 1000k tokens~1,500 A4 pages of size 12 Arial font | 262k tokens~393 A4 pages of size 12 Arial font | GLM-5.2 (max) has a larger context window than Nemotron 3 Ultra 550B A55B (Reasoning) |",
    "| Parameters | 753B, 40B active at inference time | 550B, 55B active at inference time | GLM-5.2 (max) has more parameters than Nemotron 3 Ultra 550B A55B (Reasoning) |"
   ]
  },
  {
   "url": "https://lumichats.com/blog/longcat-2-meituan-owl-alpha-chinese-open-source-ai-2026",
   "title": "A Mystery AI Topped the Charts for Two Months. It's Chinese, Free, and Built Without Nvidia.",
   "published_at": "2026-06-23T00:00:00",
   "organization": "Meituan",
   "model": null,
   "release_date": "2026-06-30",
   "access_type": null,
   "license": null,
   "significance": "1.6-trillion-parameter open-source coding model with a 1-million-token context window",
   "organization_evidence": [
    "Then, on June 30, 2026, the mystery ended: Owl Alpha was LongCat-2.0, an open-source model from Meituan \u2014 yes, the Chinese food-delivery and local-services giant."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Then, on June 30, 2026, the mystery ended: Owl Alpha was LongCat-2.0, an open-source model from Meituan \u2014 yes, the Chinese food-delivery and local-services giant."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Quick summary: LongCat-2.0 is a 1.6-trillion-parameter open-source (MIT) coding model from Meituan, with about 48 billion parameters active per token and a native 1-million-token context window."
   ]
  },
  {
   "url": "https://imisofts.com/blog/thinking-machines-inkling-open-weights-news-july-20-2026",
   "title": "Thinking Machines Ships Inkling: The Top US Open Weights Model Is a Fine Tuning Play",
   "published_at": "2026-07-20T00:00:00",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "A 975 billion parameter Mixture of Experts model featuring native text, image, and audio input, up to 1 million tokens of context, and leading US open weights performance on intelligence benchmarks.",
   "organization_evidence": [
    "Thinking Machines Ships Inkling: The Top US Open Weights Model Is a Fine Tuning Play",
    "Thinking Machines Lab, the startup founded by former OpenAI CTO Mira Murati, released its first production model on July 15."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Thinking Machines Ships Inkling: The Top US Open Weights Model Is a Fine Tuning Play",
    "Thinking Machines Lab, the startup founded by former OpenAI CTO Mira Murati, released its first production model on July 15."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Thinking Machines Ships Inkling: The Top US Open Weights Model Is a Fine Tuning Play"
   ]
  },
  {
   "url": "https://tea4tech.com/artificial-intelligence/moonshots-kimi-k3-becomes-largest-open-weight-ai-model-ever",
   "title": "Kimi K3 Becomes Largest Open-Weight AI Model Ever",
   "published_at": "2026-07-28T05:56:20",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-16",
   "access_type": null,
   "license": null,
   "significance": "a 2.8 trillion-parameter model that instantly becomes the largest open-weight system ever shipped",
   "organization_evidence": [
    "# Moonshot\u2019s Kimi K3 Becomes Largest Open-Weight AI Model Ever",
    "**BEIJING**: Moonshot AI releases Kimi K3, a 2.8 trillion-parameter model that instantly becomes the largest open-weight system ever shipped."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Moonshot\u2019s Kimi K3 Becomes Largest Open-Weight AI Model Ever",
    "The July 16 launch puts K3 live on Kimi\u2019s apps and API immediately."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Moonshot\u2019s Kimi K3 Becomes Largest Open-Weight AI Model Ever",
    "**BEIJING**: Moonshot AI releases Kimi K3, a 2.8 trillion-parameter model that instantly becomes the largest open-weight system ever shipped."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/gemma-4-26b-a4b-vs-claude-opus-4-6",
   "title": "Gemma 4 26B A4B (Reasoning) vs Claude Opus 4.6 (Non-reasoning, High Effort): Model Comparison",
   "published_at": null,
   "organization": "Google",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "Reasoning model with 256k context window and image input support",
   "organization_evidence": [
    "# Gemma 4 26B A4B (Reasoning) vs. Claude Opus 4.6 (Non-reasoning, High Effort)",
    "| Creator | ![Google](/img/logos/google_small.svg)[](https://deepmind.google/technologies/gemini/) | ![Anthropic](/img/logos/anthropic_small.svg)[](https://www.anthropic.com/) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Gemma 4 26B A4B (Reasoning) vs. Claude Opus 4.6 (Non-reasoning, High Effort)",
    "| Release Date | April, 2026 | February, 2026 | Gemma 4 26B A4B (Reasoning) has a more recent release date than Claude Opus 4.6 (Non-reasoning, High Effort) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Gemma 4 26B A4B (Reasoning) vs. Claude Opus 4.6 (Non-reasoning, High Effort)",
    "| Context Window | 256k tokens (~384 A4 pages of size 12 Arial font) | 1000k tokens (~1500 A4 pages of size 12 Arial font) | Gemma 4 26B A4B (Reasoning) is smaller than Claude Opus 4.6 (Non-reasoning, High Effort) |",
    "| Image Input Support | Yes | Yes | Both Gemma 4 26B A4B (Reasoning) and Claude Opus 4.6 (Non-reasoning, High Effort) have image input support |"
   ]
  },
  {
   "url": "https://coderfacts.com/market/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - Coder Facts",
   "published_at": "2026-07-15T12:08:34",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor",
   "organization_evidence": [
    "The Chinese open-weight landscape has expanded from a single lab two years ago to four prominent families \u2014 DeepSeek, Z.ai, Moonshot, Alibaba \u2014 each with distinct strategic focuses, such as cost-efficiency, long-horizon stability, or broad self-hosting capabilities.",
    "From late April to mid-June 2026, Chinese research institutions delivered four major open models: DeepSeek V4 on April 24, MiniMax M3 on June 1, and Kimi K2.7-Code and GLM-5.2 within days of each other in mid-June."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "From late April to mid-June 2026, Chinese research institutions delivered four major open models: DeepSeek V4 on April 24, MiniMax M3 on June 1, and Kimi K2.7-Code and GLM-5.2 within days of each other in mid-June.",
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor"
   ]
  },
  {
   "url": "https://coderfacts.com/market/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - Coder Facts",
   "published_at": "2026-07-15T12:08:34",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "cheap 1M-token context, native multimodal, modified-MIT",
   "organization_evidence": [
    "Chinese labs have released four frontier-class open-weight language models in just eight weeks, including DeepSeek V4, MiniMax M3, Kimi K2.7-Code, and GLM-5.2."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "From late April to mid-June 2026, Chinese research institutions delivered four major open models: DeepSeek V4 on April 24, MiniMax M3 on June 1, and Kimi K2.7-Code and GLM-5.2 within days of each other in mid-June."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": null
  },
  {
   "url": "https://coderfacts.com/market/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - Coder Facts",
   "published_at": "2026-07-15T12:08:34",
   "organization": "Moonshot",
   "model": null,
   "release_date": "2026-06",
   "access_type": null,
   "license": null,
   "significance": "agent-run specialist, ~30% fewer thinking tokens than K2.6",
   "organization_evidence": [
    "The Chinese open-weight landscape has expanded from a single lab two years ago to four prominent families \u2014 DeepSeek, Z.ai, Moonshot, Alibaba \u2014 each with distinct strategic focuses, such as cost-efficiency, long-horizon stability, or broad self-hosting capabilities.",
    "Kimi K2.7-Code (Moonshot)"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "From late April to mid-June 2026, Chinese research institutions delivered four major open models: DeepSeek V4 on April 24, MiniMax M3 on June 1, and Kimi K2.7-Code and GLM-5.2 within days of each other in mid-June."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": null
  },
  {
   "url": "https://coderfacts.com/market/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - Coder Facts",
   "published_at": "2026-07-15T12:08:34",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-06",
   "access_type": null,
   "license": null,
   "significance": "753B MoE, MIT, top open-weight on Artificial Analysis index",
   "organization_evidence": [
    "The Chinese open-weight landscape has expanded from a single lab two years ago to four prominent families \u2014 DeepSeek, Z.ai, Moonshot, Alibaba \u2014 each with distinct strategic focuses, such as cost-efficiency, long-horizon stability, or broad self-hosting capabilities.",
    "GLM-5.2 (Z.ai)"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "From late April to mid-June 2026, Chinese research institutions delivered four major open models: DeepSeek V4 on April 24, MiniMax M3 on June 1, and Kimi K2.7-Code and GLM-5.2 within days of each other in mid-June."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": null
  },
  {
   "url": "https://me.aiyu.co.in/blogs/inkling-open-weights-model-975b-moe-with-multimodal-reasoning-and-controllable-thinking",
   "title": "Inkling Open-Weights Model: 975B MoE with Multimodal Reasoning and Controllable Thinking",
   "published_at": "2026-07-16T12:38:48",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-16",
   "access_type": null,
   "license": null,
   "significance": "975B MoE with Multimodal Reasoning and Controllable Thinking",
   "organization_evidence": [
    "# Inkling Open-Weights Model: 975B MoE with Multimodal Reasoning and Controllable Thinking",
    "Thinking Machines Lab, the AI research company founded by former OpenAI researchers, has released **Inkling**, a new open-weights Mixture-of-Experts model that developers can fine-tune and deploy today."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Inkling Open-Weights Model: 975B MoE with Multimodal Reasoning and Controllable Thinking",
    "July 16, 2026\u202211 min read"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Inkling Open-Weights Model: 975B MoE with Multimodal Reasoning and Controllable Thinking",
    "Thinking Machines Lab, the AI research company founded by former OpenAI researchers, has released **Inkling**, a new open-weights Mixture-of-Experts model that developers can fine-tune and deploy today."
   ]
  },
  {
   "url": "https://1ban.news/kimi-k3-2-8-trillion-open-weight-2026",
   "title": "China\u2019s Moonshot AI releases 2.8-trillion-parameter Kimi K3, the largest open-weight model ever",
   "published_at": "2026-07-18T00:44:18",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-16",
   "access_type": null,
   "license": null,
   "significance": "a 2.8-trillion-parameter mixture-of-experts model that the company describes as the world\u2019s first open 3T-class system and the largest open-weight AI model ever released",
   "organization_evidence": [
    "# China\u2019s Moonshot AI releases 2.8-trillion-parameter Kimi K3, the largest open-weight model ever",
    "Moonshot AI has released Kimi K3, a 2.8-trillion-parameter mixture-of-experts model that the company describes as the world\u2019s first open 3T-class system and the largest open-weight AI model ever released."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# China\u2019s Moonshot AI releases 2.8-trillion-parameter Kimi K3, the largest open-weight model ever",
    "The model, which went live on July 16 through the Kimi app, Playground, and API, activates 16 of 896 experts per token using Stable LatentMoE, with an architecture built around the company\u2019s proprietary Kimi Delta Attention mechanism, a hybrid linear-attention design that Moonshot says delivers up to 6.3 times faster decoding in million-token contexts compared to standard attention."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# China\u2019s Moonshot AI releases 2.8-trillion-parameter Kimi K3, the largest open-weight model ever",
    "Moonshot AI has released Kimi K3, a 2.8-trillion-parameter mixture-of-experts model that the company describes as the world\u2019s first open 3T-class system and the largest open-weight AI model ever released."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/grok-4-20-vs-glm-5-1-non-reasoning",
   "title": "Grok 4.20 0309 v2 (Reasoning) vs GLM-5.1 (Non-reasoning): Model Comparison",
   "published_at": null,
   "organization": "Z AI",
   "model": null,
   "release_date": "April, 2026",
   "access_type": null,
   "license": null,
   "significance": "A major non-reasoning language model with open source weights and a 200k token context window",
   "organization_evidence": [
    "## Model Comparison",
    "| Metric | ![xAI logo](/img/logos/xai.svg)Grok 4.20 0309 v2 (Reasoning) | ![Z AI logo](/img/logos/zai_small.svg)GLM-5.1 (Non-reasoning) | Analysis |",
    "| Creator | ![xAI](/img/logos/xai.svg)[](https://x.ai/) | ![Z AI](/img/logos/zai_small.svg)[](https://z.ai/) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## Model Comparison",
    "| Metric | ![xAI logo](/img/logos/xai.svg)Grok 4.20 0309 v2 (Reasoning) | ![Z AI logo](/img/logos/zai_small.svg)GLM-5.1 (Non-reasoning) | Analysis |",
    "| Release Date | April, 2026 | April, 2026 | Both Grok 4.20 0309 v2 (Reasoning) and GLM-5.1 (Non-reasoning) have the same release date |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Grok 4.20 0309 v2 (Reasoning) vs. GLM-5.1 (Non-reasoning)",
    "Comparison between Grok 4.20 0309 v2 (Reasoning) and GLM-5.1 (Non-reasoning) across intelligence, price, speed, context window and more.",
    "## Model Comparison",
    "| Metric | ![xAI logo](/img/logos/xai.svg)Grok 4.20 0309 v2 (Reasoning) | ![Z AI logo](/img/logos/zai_small.svg)GLM-5.1 (Non-reasoning) | Analysis |",
    "| Context Window | 2000k tokens (~3000 A4 pages of size 12 Arial font) | 200k tokens (~300 A4 pages of size 12 Arial font) | Grok 4.20 0309 v2 (Reasoning) is larger than GLM-5.1 (Non-reasoning) |",
    "| Open Source (Weights) | No | Yes[](https://huggingface.co/zai-org/GLM-5.1) | GLM-5.1 (Non-reasoning) is open source while Grok 4.20 0309 v2 (Reasoning) is proprietary |"
   ]
  },
  {
   "url": "https://subagentic.ai/posts/meituan-longcat-2-open-source-1-6t-coding-model",
   "title": "Meituan Open-Sources LongCat-2.0: 1.6T-Parameter Agentic Coding Model Trained on Chinese Chips, MIT License, Leads OpenRouter",
   "published_at": "2026-07-01T15:12:00",
   "organization": "Meituan",
   "model": null,
   "release_date": "2026-06-30",
   "access_type": null,
   "license": null,
   "significance": "1.6 trillion parameter Mixture-of-Experts agentic coding model",
   "organization_evidence": [
    "# Meituan Open-Sources LongCat-2.0: 1.6T-Parameter Agentic Coding Model Trained on Chinese Chips, MIT License, Leads OpenRouter",
    "Chinese food-delivery giant Meituan just dropped one of the most significant open-source AI releases of 2026: LongCat-2.0, a 1.6 trillion parameter Mixture-of-Experts agentic coding model \u2014 trained entirely on domestic Chinese chips, licensed under MIT, and now unmasked as \u201cOwl Alpha,\u201d the anonymous model that spent two months leading OpenRouter\u2019s performance charts."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Meituan Open-Sources LongCat-2.0: 1.6T-Parameter Agentic Coding Model Trained on Chinese Chips, MIT License, Leads OpenRouter",
    "That changed when Meituan officially unveiled LongCat-2.0 on June 30, 2026, releasing it simultaneously on GitHub, Hugging Face, and its native platform at longcatai.org."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Meituan Open-Sources LongCat-2.0: 1.6T-Parameter Agentic Coding Model Trained on Chinese Chips, MIT License, Leads OpenRouter",
    "Chinese food-delivery giant Meituan just dropped one of the most significant open-source AI releases of 2026: LongCat-2.0, a 1.6 trillion parameter Mixture-of-Experts agentic coding model \u2014 trained entirely on domestic Chinese chips, licensed under MIT, and now unmasked as \u201cOwl Alpha,\u201d the anonymous model that spent two months leading OpenRouter\u2019s performance charts."
   ]
  },
  {
   "url": "https://kanerika.com/blogs/llama-3-vs-llama-2",
   "title": "10 Best Open Source LLMs to Evaluate in 2026",
   "published_at": "2026-06-16T13:58:04",
   "organization": "Meta Superintelligence Labs",
   "model": null,
   "release_date": "2026-04-08",
   "access_type": null,
   "license": null,
   "significance": "scoring 58% on Humanity's Last Exam, competitive with GPT Pro and Gemini Deep Think",
   "organization_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "Meta Superintelligence Labs launched [Muse Spark](https://ai.meta.com/blog/introducing-muse-spark-msl/) on April 8, 2026, distinct from the Llama family and available at meta.ai with a private API preview."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "Meta Superintelligence Labs launched [Muse Spark](https://ai.meta.com/blog/introducing-muse-spark-msl/) on April 8, 2026, distinct from the Llama family and available at meta.ai with a private API preview."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "Its Contemplating mode runs specialized agents in parallel, each reasoning independently before converging on a single verified answer, [scoring 58% on Humanity\u2019s Last Exam](https://ai.meta.com/blog/introducing-muse-spark-msl/), competitive with GPT Pro and Gemini Deep Think."
   ]
  },
  {
   "url": "https://kanerika.com/blogs/llama-3-vs-llama-2",
   "title": "10 Best Open Source LLMs to Evaluate in 2026",
   "published_at": "2026-06-16T13:58:04",
   "organization": "Meta",
   "model": null,
   "release_date": "2026-04-05",
   "access_type": null,
   "license": null,
   "significance": "10 million token context window and Maverick scaling to 128 experts and 400B total parameters",
   "organization_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "Meta released [Llama 4 Scout and Maverick](https://ai.meta.com/blog/llama-4-multimodal-intelligence/) on April 5, 2026, the first Llama models built on MoE architecture Meta released [Llama 4 Scout and Maverick](https://ai.meta.com/blog/llama-4-multimodal-intelligence/) on April 5, 2026, the first Llama models built on MoE architecture and trained as natively multimodal systems using early fusion, processing text, images, and video through a unified model."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "Meta released [Llama 4 Scout and Maverick](https://ai.meta.com/blog/llama-4-multimodal-intelligence/) on April 5, 2026, the first Llama models built on MoE architecture Meta released [Llama 4 Scout and Maverick](https://ai.meta.com/blog/llama-4-multimodal-intelligence/) on April 5, 2026, the first Llama models built on MoE architecture and trained as natively multimodal systems using early fusion, processing text, images, and video through a unified model."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "Scout runs on a single H100 GPU with a [10 million token context window](https://www.llama.com/models/llama-4/), currently the largest in any open-weight model."
   ]
  },
  {
   "url": "https://kanerika.com/blogs/llama-3-vs-llama-2",
   "title": "10 Best Open Source LLMs to Evaluate in 2026",
   "published_at": "2026-06-16T13:58:04",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-30",
   "access_type": null,
   "license": null,
   "significance": "most cost-effective frontier-class open model available today with 1M token context window",
   "organization_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "DeepSeek V4 launched in late April 2026 and comes in two sizes: [V4-Pro (1.6T total, 49B active parameters)](https://api-docs.deepseek.com/) for maximum performance, and V4-Flash (284B total, 13B active) as a lighter, cheaper alternative."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "DeepSeek V4 launched in late April 2026 and comes in two sizes: [V4-Pro (1.6T total, 49B active parameters)](https://api-docs.deepseek.com/) for maximum performance, and V4-Flash (284B total, 13B active) as a lighter, cheaper alternative."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "Both expose a 1M token context window and carry MIT licensing for full self-hosting."
   ]
  },
  {
   "url": "https://kanerika.com/blogs/llama-3-vs-llama-2",
   "title": "10 Best Open Source LLMs to Evaluate in 2026",
   "published_at": "2026-06-16T13:58:04",
   "organization": "Alibaba Cloud",
   "model": null,
   "release_date": "2026-02-01",
   "access_type": null,
   "license": null,
   "significance": "MoE model with 397B total and 17B active parameters, 1M token context window, supporting 201 languages",
   "organization_evidence": [
    "Top 10 Open-Source LLMs in 2026"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Top 10 Open-Source LLMs in 2026"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "The Qwen3.6-35B-A3B is a MoE model with 397B total and 17B active parameters, a 1 million token context window, and native multimodality across text, image, and video through early fusion architecture."
   ]
  },
  {
   "url": "https://kanerika.com/blogs/llama-3-vs-llama-2",
   "title": "10 Best Open Source LLMs to Evaluate in 2026",
   "published_at": "2026-06-16T13:58:04",
   "organization": "Google DeepMind",
   "model": null,
   "release_date": "2026-04-02",
   "access_type": null,
   "license": null,
   "significance": "ranks #3 among all open models on Arena AI, scoring 89.2% on AIME 2026",
   "organization_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "Google DeepMind released [Gemma 4](https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/) on April 2, 2026 under Apache 2.0 with full commercial freedom."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "Google DeepMind released [Gemma 4](https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/) on April 2, 2026 under Apache 2.0 with full commercial freedom."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "The 31B ranks [#3 among all open models on Arena AI](https://deepmind.google/models/gemma/gemma-4/), scores 89.2% on AIME 2026, 80.0% on LiveCodeBench v6, and 86.4% on \u03c42-bench agentic tool use."
   ]
  },
  {
   "url": "https://kanerika.com/blogs/llama-3-vs-llama-2",
   "title": "10 Best Open Source LLMs to Evaluate in 2026",
   "published_at": "2026-06-16T13:58:04",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-04-07",
   "access_type": null,
   "license": null,
   "significance": "744B total and 40B active parameters and a 200K context window with strong SWE-bench Pro scores",
   "organization_evidence": [
    "Top 10 Open-Source LLMs in 2026"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Top 10 Open-Source LLMs in 2026"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "With 744B total and 40B active parameters and a 200K context window, it is MIT licensed and has posted some of the strongest SWE-bench Pro scores in the open-weight category."
   ]
  },
  {
   "url": "https://kanerika.com/blogs/llama-3-vs-llama-2",
   "title": "10 Best Open Source LLMs to Evaluate in 2026",
   "published_at": "2026-06-16T13:58:04",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "combines a 1M token context window, native multimodality, and frontier coding ability topping open-weight SWE-Bench Pro at 59.0%",
   "organization_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "[MiniMax M3](https://huggingface.co/MiniMaxAI/MiniMax-M3), released June 1, 2026, is the most capable open-weight model MiniMax has shipped."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "[MiniMax M3](https://huggingface.co/MiniMaxAI/MiniMax-M3), released June 1, 2026, is the most capable open-weight model MiniMax has shipped."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "It combines a 1M token context window, native multimodality, and frontier coding ability in a single model, topping open-weight SWE-Bench Pro at 59.0%."
   ]
  },
  {
   "url": "https://heybeagle.com/blog/inkling-thinking-machines-labs-open-weight-model-for-agent-teams",
   "title": "Inkling: Thinking Machines Lab's Open-Weight Model for Agent Teams",
   "published_at": "2026-07-16T20:01:52",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "A 975B MoE open-weight model supporting one million token context windows with strong performance on reasoning, coding, and agentic workflows.",
   "organization_evidence": [
    "Inkling: Thinking Machines Lab's Open-Weight Model for Agent Teams",
    "Thinking Machines Lab dropped Inkling on July 15 - a 975B MoE open-weight model that scores 77.6% on SWE-bench Verified but refuses to claim the top spot."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Inkling: Thinking Machines Lab's Open-Weight Model for Agent Teams",
    "Thinking Machines Lab dropped Inkling on July 15 - a 975B MoE open-weight model that scores 77.6% on SWE-bench Verified but refuses to claim the top spot."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Inkling: Thinking Machines Lab's Open-Weight Model for Agent Teams",
    "Thinking Machines Lab dropped Inkling on July 15 - a 975B MoE open-weight model that scores 77.6% on SWE-bench Verified but refuses to claim the top spot.",
    "Inkling is a Mixture-of-Experts transformer with 975 billion total parameters and 41 billion active during inference."
   ]
  },
  {
   "url": "https://braindetox.kr/en/posts/longcat_2_open_model_2026.html",
   "title": "LongCat-2.0: A 1.6-Trillion-Parameter Open Model Trained Without Nvidia (2026)",
   "published_at": "2026-07-03T00:00:00",
   "organization": "Meituan",
   "model": "LongCat-2.0",
   "release_date": "2026-06-30",
   "access_type": "open source",
   "license": "MIT",
   "significance": "1.6-trillion-parameter Mixture-of-Experts model trained without Nvidia GPUs",
   "organization_evidence": [
    "# LongCat-2.0: A 1.6-Trillion-Parameter Open Model Trained Without Nvidia",
    "On June 30, 2026, Chinese food-delivery and local-services giant **Meituan** released a large language model called **LongCat-2.0**."
   ],
   "model_evidence": [
    "# LongCat-2.0: A 1.6-Trillion-Parameter Open Model Trained Without Nvidia",
    "On June 30, 2026, Chinese food-delivery and local-services giant **Meituan** released a large language model called **LongCat-2.0**."
   ],
   "release_date_evidence": [
    "# LongCat-2.0: A 1.6-Trillion-Parameter Open Model Trained Without Nvidia",
    "On June 30, 2026, Chinese food-delivery and local-services giant **Meituan** released a large language model called **LongCat-2.0**."
   ],
   "access_type_evidence": [
    "# LongCat-2.0: A 1.6-Trillion-Parameter Open Model Trained Without Nvidia",
    "There's an asterisk here that needs to be stated plainly.",
    "LongCat-2.0 carries an MIT license and is billed as \"open source,\" but as of the announcement, both the GitHub repository and the Hugging Face model card only said weights were **\"coming soon.\"**"
   ],
   "license_evidence": [
    "# LongCat-2.0: A 1.6-Trillion-Parameter Open Model Trained Without Nvidia",
    "There's an asterisk here that needs to be stated plainly.",
    "LongCat-2.0 carries an MIT license and is billed as \"open source,\" but as of the announcement, both the GitHub repository and the Hugging Face model card only said weights were **\"coming soon.\"**"
   ],
   "significance_evidence": [
    "# LongCat-2.0: A 1.6-Trillion-Parameter Open Model Trained Without Nvidia",
    "According to public materials, LongCat-2.0 is a **1.6-trillion-parameter** Mixture-of-Experts (MoE) model, of which only about **33 to 56 billion parameters (roughly 48 billion on average)** actually activate per token."
   ]
  },
  {
   "url": "https://rits.shanghai.nyu.edu/ai/arcee-ai-releases-trinity-large-thinking-a-400b-open-reasoning-agent",
   "title": "Arcee AI Releases Trinity-Large-Thinking: A 400B Open Reasoning Agent",
   "published_at": null,
   "organization": "Arcee AI",
   "model": null,
   "release_date": "2026-04-01",
   "access_type": null,
   "license": null,
   "significance": "A 398B parameter open reasoning agent and sparse Mixture-of-Experts model achieving strong performance on agentic benchmarks.",
   "organization_evidence": [
    "# Arcee AI Releases Trinity-Large-Thinking: A 400B Open Reasoning Agent",
    "**Arcee AI released Trinity-Large-Thinking on April 1, 2026** \u2014 an open-source frontier reasoning model built for complex, long-horizon AI agents."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Arcee AI Releases Trinity-Large-Thinking: A 400B Open Reasoning Agent",
    "**Arcee AI released Trinity-Large-Thinking on April 1, 2026** \u2014 an open-source frontier reasoning model built for complex, long-horizon AI agents."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Arcee AI Releases Trinity-Large-Thinking: A 400B Open Reasoning Agent",
    "**Arcee AI released Trinity-Large-Thinking on April 1, 2026** \u2014 an open-source frontier reasoning model built for complex, long-horizon AI agents.",
    "Weighing in at 398 billion total parameters (with ~13B active per token), it is one of the largest open-source models ever released by a U.S. startup, and it puts a credible challenge to proprietary alternatives at a fraction of the cost."
   ]
  },
  {
   "url": "https://marktechpost.com/2026/07/21/poolside-releases-laguna-s-2-1",
   "title": "Poolside Releases Laguna S 2.1, an Open-Weight Agentic Coding Model Punching Above Its Weight Class on SWE-Bench Multilingual",
   "published_at": "2026-07-22T00:01:32",
   "organization": "Poolside",
   "model": null,
   "release_date": "2026-05-22",
   "access_type": null,
   "license": null,
   "significance": "a 118B-parameter open-weight model built for agentic coding supporting a context window of up to 1M tokens in both thinking and no-thinking modes",
   "organization_evidence": [
    "Poolside Releases Laguna S 2.1, a 118B Open-Weight MoE Coding Model Matching Rivals Many Times Its Size"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Poolside Releases Laguna S 2.1, a 118B Open-Weight MoE Coding Model Matching Rivals Many Times Its Size",
    "Pre-training began on 22 May 2026 on 4,096 NVIDIA H200 GPUs."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Poolside Releases Laguna S 2.1, a 118B Open-Weight MoE Coding Model Matching Rivals Many Times Its Size"
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/mimo-v2-5-pro-vs-glm-5-1",
   "title": "MiMo-V2.5-Pro vs GLM-5.1 (Reasoning): Model Comparison | Artificial Analysis",
   "published_at": null,
   "organization": "Xiaomi",
   "model": "MiMo-V2.5-Pro",
   "release_date": "2026-04",
   "access_type": "open source",
   "license": "Mit",
   "significance": "Intelligence Index 43, 1023B parameters, 1000k context window",
   "organization_evidence": [
    "# MiMo-V2.5-Pro vs. GLM-5.1 (Reasoning)",
    "## Model Comparison",
    "| | ![Xiaomi logo](/img/logos/xiaomi_small.svg)Xiaomi MiMo-V2.5-Pro | ![Z AI logo](/img/logos/zai_small.svg)Z AI GLM-5.1 (Reasoning) | |"
   ],
   "model_evidence": [
    "# MiMo-V2.5-Pro vs. GLM-5.1 (Reasoning)",
    "## Model Comparison",
    "| | ![Xiaomi logo](/img/logos/xiaomi_small.svg)Xiaomi MiMo-V2.5-Pro | ![Z AI logo](/img/logos/zai_small.svg)Z AI GLM-5.1 (Reasoning) | |"
   ],
   "release_date_evidence": [
    "# MiMo-V2.5-Pro vs. GLM-5.1 (Reasoning)",
    "## Model Comparison",
    "| | ![Xiaomi logo](/img/logos/xiaomi_small.svg)Xiaomi MiMo-V2.5-Pro | ![Z AI logo](/img/logos/zai_small.svg)Z AI GLM-5.1 (Reasoning) | |",
    "| Release Date | April 2026 | April 2026 | MiMo-V2.5-Pro has a more recent release date than GLM-5.1 (Reasoning) |"
   ],
   "access_type_evidence": [
    "# MiMo-V2.5-Pro vs. GLM-5.1 (Reasoning)",
    "## Model Comparison",
    "| | ![Xiaomi logo](/img/logos/xiaomi_small.svg)Xiaomi MiMo-V2.5-Pro | ![Z AI logo](/img/logos/zai_small.svg)Z AI GLM-5.1 (Reasoning) | |",
    "| Open Source (Weights) | Yes | Yes | Both MiMo-V2.5-Pro and GLM-5.1 (Reasoning) are open source |"
   ],
   "license_evidence": [
    "# MiMo-V2.5-Pro vs. GLM-5.1 (Reasoning)",
    "## Model Comparison",
    "| | ![Xiaomi logo](/img/logos/xiaomi_small.svg)Xiaomi MiMo-V2.5-Pro | ![Z AI logo](/img/logos/zai_small.svg)Z AI GLM-5.1 (Reasoning) | |",
    "| License | Mit| Mit| |"
   ],
   "significance_evidence": [
    "# MiMo-V2.5-Pro vs. GLM-5.1 (Reasoning)",
    "## Model Comparison",
    "| | ![Xiaomi logo](/img/logos/xiaomi_small.svg)Xiaomi MiMo-V2.5-Pro | ![Z AI logo](/img/logos/zai_small.svg)Z AI GLM-5.1 (Reasoning) | |",
    "| Intelligence Index | 43 | 41 | MiMo-V2.5-Pro is more intelligent than GLM-5.1 (Reasoning) |",
    "| Context Window | 1000k tokens~1,500 A4 pages of size 12 Arial font | 200k tokens~300 A4 pages of size 12 Arial font | MiMo-V2.5-Pro has a larger context window than GLM-5.1 (Reasoning) |",
    "| Parameters | 1023B, 42B active at inference time | 744B, 40B active at inference time | MiMo-V2.5-Pro has more parameters than GLM-5.1 (Reasoning) |"
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/mimo-v2-5-pro-vs-glm-5-1",
   "title": "MiMo-V2.5-Pro vs GLM-5.1 (Reasoning): Model Comparison | Artificial Analysis",
   "published_at": null,
   "organization": "Z AI",
   "model": "GLM-5.1 (Reasoning)",
   "release_date": "2026-04",
   "access_type": "open source",
   "license": "Mit",
   "significance": "Intelligence Index 41, 744B parameters, reasoning capabilities",
   "organization_evidence": [
    "# MiMo-V2.5-Pro vs. GLM-5.1 (Reasoning)",
    "## Model Comparison",
    "| | ![Xiaomi logo](/img/logos/xiaomi_small.svg)Xiaomi MiMo-V2.5-Pro | ![Z AI logo](/img/logos/zai_small.svg)Z AI GLM-5.1 (Reasoning) | |"
   ],
   "model_evidence": [
    "# MiMo-V2.5-Pro vs. GLM-5.1 (Reasoning)",
    "## Model Comparison",
    "| | ![Xiaomi logo](/img/logos/xiaomi_small.svg)Xiaomi MiMo-V2.5-Pro | ![Z AI logo](/img/logos/zai_small.svg)Z AI GLM-5.1 (Reasoning) | |"
   ],
   "release_date_evidence": [
    "# MiMo-V2.5-Pro vs. GLM-5.1 (Reasoning)",
    "## Model Comparison",
    "| | ![Xiaomi logo](/img/logos/xiaomi_small.svg)Xiaomi MiMo-V2.5-Pro | ![Z AI logo](/img/logos/zai_small.svg)Z AI GLM-5.1 (Reasoning) | |",
    "| Release Date | April 2026 | April 2026 | MiMo-V2.5-Pro has a more recent release date than GLM-5.1 (Reasoning) |"
   ],
   "access_type_evidence": [
    "# MiMo-V2.5-Pro vs. GLM-5.1 (Reasoning)",
    "## Model Comparison",
    "| | ![Xiaomi logo](/img/logos/xiaomi_small.svg)Xiaomi MiMo-V2.5-Pro | ![Z AI logo](/img/logos/zai_small.svg)Z AI GLM-5.1 (Reasoning) | |",
    "| Open Source (Weights) | Yes | Yes | Both MiMo-V2.5-Pro and GLM-5.1 (Reasoning) are open source |"
   ],
   "license_evidence": [
    "# MiMo-V2.5-Pro vs. GLM-5.1 (Reasoning)",
    "## Model Comparison",
    "| | ![Xiaomi logo](/img/logos/xiaomi_small.svg)Xiaomi MiMo-V2.5-Pro | ![Z AI logo](/img/logos/zai_small.svg)Z AI GLM-5.1 (Reasoning) | |",
    "| License | Mit| Mit| |"
   ],
   "significance_evidence": [
    "# MiMo-V2.5-Pro vs. GLM-5.1 (Reasoning)",
    "## Model Comparison",
    "| | ![Xiaomi logo](/img/logos/xiaomi_small.svg)Xiaomi MiMo-V2.5-Pro | ![Z AI logo](/img/logos/zai_small.svg)Z AI GLM-5.1 (Reasoning) | |",
    "| Intelligence Index | 43 | 41 | MiMo-V2.5-Pro is more intelligent than GLM-5.1 (Reasoning) |",
    "| Parameters | 1023B, 42B active at inference time | 744B, 40B active at inference time | MiMo-V2.5-Pro has more parameters than GLM-5.1 (Reasoning) |",
    "| Reasoning | Yes | Yes | Both MiMo-V2.5-Pro and GLM-5.1 (Reasoning) have reasoning |"
   ]
  },
  {
   "url": "http://eng.tibet.cn/eng/index/rolling/202603/t20260316_7946977.html",
   "title": "Tibetan Large Language Model DeepZang Launched in Lhasa_News2_China Tibet Online",
   "published_at": "2026-03-16T00:00:00",
   "organization": "Xizang CHOKNOR Digital Industry Co., Ltd.",
   "model": null,
   "release_date": "2026-03-15",
   "access_type": null,
   "license": null,
   "significance": "the first Tibetan large language model in China to have completed national registration for generative artificial intelligence",
   "organization_evidence": [
    "Tibetan Large Language Model DeepZang Launched in Lhasa",
    "DeepZang was independently developed by Xizang CHOKNOR Digital Industry Co., Ltd."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Tibetan Large Language Model DeepZang Launched in Lhasa",
    "On March 15, DeepZang, a Tibetan large language model with its intelligent applications, was launched in Lhasa, Xizang."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Tibetan Large Language Model DeepZang Launched in Lhasa",
    "This is the first Tibetan large language model in China to have completed national registration for generative artificial intelligence (AI)."
   ]
  },
  {
   "url": "https://felloai.com/microsoft-mai-models",
   "title": "Microsoft Just Launched Its Own MAI Models",
   "published_at": "2026-06-03T02:46:00",
   "organization": "Microsoft",
   "model": null,
   "release_date": "2026-06-02",
   "access_type": null,
   "license": null,
   "significance": "A 35 billion active parameter Mixture of Experts reasoning flagship model trained from scratch with clean data",
   "organization_evidence": [
    "Microsoft Just Launched Its Own MAI Models",
    "On June 2, 2026, at its Build developer conference, Microsoft announced a family of seven in-house AI models under the MAI brand, the most concrete step yet in the company\u2019s plan to stop being just the best distributor of someone else\u2019s frontier models."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "On June 2, 2026, at its Build developer conference, Microsoft announced a family of seven in-house AI models under the MAI brand, the most concrete step yet in the company\u2019s plan to stop being just the best distributor of someone else\u2019s frontier models."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "MAI-Thinking-1, the Reasoning Flagship",
    "MAI-Thinking-1 is a sparse Mixture of Experts model with 35 billion active parameters and roughly 1 trillion total parameters.",
    "Microsoft Just Launched Its Own MAI Models",
    "On June 2, 2026, at its Build developer conference, Microsoft announced a family of seven in-house AI models under the MAI brand, the most concrete step yet in the company\u2019s plan to stop being just the best distributor of someone else\u2019s frontier models."
   ]
  },
  {
   "url": "https://jagadishwrites.com/blog/open-source-ai-models-challenging-proprietary-llms-in-2026-the-gap-has-closed",
   "title": "Open-Source AI Models Challenging Proprietary LLMs in 2026: The Gap Has Closed",
   "published_at": "2026-07-08T00:00:00",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-06",
   "access_type": null,
   "license": null,
   "significance": "flagship open-weight model for agentic engineering, software development, and long-horizon reasoning tasks",
   "organization_evidence": [
    "Open-Source AI Models Challenging Proprietary LLMs in 2026: The Gap Has Closed",
    "### GLM-5.2: The New #1 for Agentic Engineering",
    "Released in June 2026, **GLM-5.2** from Z.ai has emerged as the flagship open-weight model for agentic engineering, software development, and long-horizon reasoning tasks."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Open-Source AI Models Challenging Proprietary LLMs in 2026: The Gap Has Closed",
    "### GLM-5.2: The New #1 for Agentic Engineering",
    "Released in June 2026, **GLM-5.2** from Z.ai has emerged as the flagship open-weight model for agentic engineering, software development, and long-horizon reasoning tasks."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Open-Source AI Models Challenging Proprietary LLMs in 2026: The Gap Has Closed",
    "### GLM-5.2: The New #1 for Agentic Engineering",
    "Released in June 2026, **GLM-5.2** from Z.ai has emerged as the flagship open-weight model for agentic engineering, software development, and long-horizon reasoning tasks."
   ]
  },
  {
   "url": "https://jagadishwrites.com/blog/open-source-ai-models-challenging-proprietary-llms-in-2026-the-gap-has-closed",
   "title": "Open-Source AI Models Challenging Proprietary LLMs in 2026: The Gap Has Closed",
   "published_at": "2026-07-08T00:00:00",
   "organization": null,
   "model": null,
   "release_date": "2026-06",
   "access_type": null,
   "license": null,
   "significance": "first open-weight model to combine frontier coding, 1M context, and native multimodality",
   "organization_evidence": null,
   "model_evidence": null,
   "release_date_evidence": [
    "Open-Source AI Models Challenging Proprietary LLMs in 2026: The Gap Has Closed",
    "### MiniMax M3: The First Open-Weight Model to Combine Frontier Capabilities",
    "Released in June 2026, **MiniMax M3** is the first open-weight model to combine frontier coding, 1M context, and native multimodality."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Open-Source AI Models Challenging Proprietary LLMs in 2026: The Gap Has Closed",
    "### MiniMax M3: The First Open-Weight Model to Combine Frontier Capabilities",
    "Released in June 2026, **MiniMax M3** is the first open-weight model to combine frontier coding, 1M context, and native multimodality."
   ]
  },
  {
   "url": "https://jagadishwrites.com/blog/open-source-ai-models-challenging-proprietary-llms-in-2026-the-gap-has-closed",
   "title": "Open-Source AI Models Challenging Proprietary LLMs in 2026: The Gap Has Closed",
   "published_at": "2026-07-08T00:00:00",
   "organization": null,
   "model": null,
   "release_date": "2026-06",
   "access_type": null,
   "license": null,
   "significance": "excels at agent swarms and long autonomous runs",
   "organization_evidence": null,
   "model_evidence": null,
   "release_date_evidence": [
    "Open-Source AI Models Challenging Proprietary LLMs in 2026: The Gap Has Closed",
    "### Kimi K2.7 Code: The Agentic Coding Specialist",
    "**Kimi K2.7 Code**, released in June 2026, excels at agent swarms and long autonomous runs."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Open-Source AI Models Challenging Proprietary LLMs in 2026: The Gap Has Closed",
    "### Kimi K2.7 Code: The Agentic Coding Specialist",
    "**Kimi K2.7 Code**, released in June 2026, excels at agent swarms and long autonomous runs."
   ]
  },
  {
   "url": "https://thequantumdispatch.com/articles/qwen-agentworld-open-weight-language-world-model-ai-agents-june-2026",
   "title": "Qwen-AgentWorld Is an Open Model That Simulates Worlds for AI Agents",
   "published_at": "2026-06-28T04:12:39",
   "organization": "Alibaba",
   "model": "Qwen-AgentWorld",
   "release_date": "2026-06-24",
   "access_type": "open-weight",
   "license": "Apache 2.0 license",
   "significance": "a 35B-parameter mixture-of-experts language world model covering seven distinct agent environments with a 262,144-token context window",
   "organization_evidence": [
    "# Qwen-AgentWorld Is an Open Model That Simulates Worlds for AI Agents",
    "Alibaba's Qwen team open-sourced AgentWorld on June 24, 2026 \u2014 a language world model that simulates digital environments so AI agents can practice and improve.",
    "On **June 24, 2026**, Alibaba's **Qwen team** released **Qwen-AgentWorld**, an **open-weight \"language world model\"** designed to do exactly that \u2014 simulate the digital environments an agent works in so it can learn against them safely and cheaply."
   ],
   "model_evidence": [
    "# Qwen-AgentWorld Is an Open Model That Simulates Worlds for AI Agents",
    "On **June 24, 2026**, Alibaba's **Qwen team** released **Qwen-AgentWorld**, an **open-weight \"language world model\"** designed to do exactly that \u2014 simulate the digital environments an agent works in so it can learn against them safely and cheaply."
   ],
   "release_date_evidence": [
    "# Qwen-AgentWorld Is an Open Model That Simulates Worlds for AI Agents",
    "Alibaba's Qwen team open-sourced AgentWorld on June 24, 2026 \u2014 a language world model that simulates digital environments so AI agents can practice and improve.",
    "On **June 24, 2026**, Alibaba's **Qwen team** released **Qwen-AgentWorld**, an **open-weight \"language world model\"** designed to do exactly that \u2014 simulate the digital environments an agent works in so it can learn against them safely and cheaply."
   ],
   "access_type_evidence": [
    "# Qwen-AgentWorld Is an Open Model That Simulates Worlds for AI Agents",
    "Worlds for AI Agents\n\nAlibaba's Qwen team open-sourced AgentWorld on June 24, 2026 \u2014 a language world model that simulates digital environments so AI agents can practice and improve.",
    "On **June 24, 2026**, Alibaba's **Qwen team** released **Qwen-AgentWorld**, an **open-weight \"language world model\"** designed to do exactly that \u2014 simulate the digital environments an agent works in so it can learn against them safely and cheaply.",
    "This is an **open-weight AI model**, which is the part I find most valuable for the broader community."
   ],
   "license_evidence": [
    "# Qwen-AgentWorld Is an Open Model That Simulates Worlds for AI Agents",
    "The Qwen team published a compact **35B-parameter mixture-of-experts release** (with roughly 3B active parameters) under the permissive **Apache 2.0 license**, with weights available on Hugging Face and ModelScope and commercial use permitted."
   ],
   "significance_evidence": [
    "# Qwen-AgentWorld Is an Open Model That Simulates Worlds for AI Agents",
    "Remarkably, the model covers **seven distinct agent environments in one place** \u2014 MCP, search, terminal, software engineering, web, operating system, and Android \u2014 giving developers a single sandbox spanning the surfaces real agents operate on.",
    "The Qwen team published a compact **35B-parameter mixture-of-experts release** (with roughly 3B active parameters) under the permissive **Apache 2.0 license**, with weights available on Hugging Face and ModelScope and commercial use permitted.",
    "It supports a generous context window of **262,144 tokens**."
   ]
  },
  {
   "url": "https://venturebeat.com/technology/mistral-drops-voxtral-transcribe-2-an-open-source-speech-model-that-runs-on",
   "title": "Mistral drops Voxtral Transcribe 2, an open-source speech model that runs on-device for pennies",
   "published_at": "2026-02-04T22:30:56",
   "organization": "Mistral AI",
   "model": null,
   "release_date": "2026-02-26",
   "access_type": null,
   "license": null,
   "significance": "Processes live audio with a latency that can be configured down to 200 milliseconds.",
   "organization_evidence": [
    "[Mistral AI](https://mistral.ai/), the Paris-based startup positioning itself as Europe's answer to OpenAI, released a pair of speech-to-text models on Wednesday that the company says can transcribe audio faster, more accurately, and far more cheaply than anything else on the market \u2014 all while running entirely on a smartphone or laptop."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "[Mistral AI](https://mistral.ai/), the Paris-based startup positioning itself as Europe's answer to OpenAI, released a pair of speech-to-text models on Wednesday that the company says can transcribe audio faster, more accurately, and far more cheaply than anything else on the market \u2014 all while running entirely on a smartphone or laptop.",
    "- [**Voxtral Mini Transcribe V2**](https://docs.mistral.ai/models/voxtral-mini-transcribe-26-02) handles batch transcription, processing pre-recorded audio files in bulk.",
    "The [Realtime model](https://huggingface.co/mistralai/Voxtral-Mini-4B-Realtime-2602) ships under an [Apache 2.0](https://www.apache.org/licenses/LICENSE-2.0) open-source license, meaning developers can download the model weights from [Hugging Face](https://huggingface.co/mistralai), modify them, and deploy them without paying Mistral a licensing fee."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": null
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/gemma-4-26b-a4b-vs-gpt-5-4",
   "title": "Gemma 4 26B A4B (Reasoning) vs GPT-5.4 (xhigh): Model Comparison",
   "published_at": null,
   "organization": "Google",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "A major reasoning model released with 256k tokens context window and open source weights.",
   "organization_evidence": [
    "# Gemma 4 26B A4B (Reasoning) vs. GPT-5.4 (xhigh)",
    "| Creator | ![Google](/img/logos/google_small.svg)[](https://deepmind.google/technologies/gemini/) | ![OpenAI](/img/logos/openai_small.svg)[](https://openai.com/) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Gemma 4 26B A4B (Reasoning) vs. GPT-5.4 (xhigh)",
    "| Release Date | April, 2026 | March, 2026 | Gemma 4 26B A4B (Reasoning) has a more recent release date than GPT-5.4 (xhigh) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Gemma 4 26B A4B (Reasoning) vs. GPT-5.4 (xhigh)",
    "| Context Window | 256k tokens (~384 A4 pages of size 12 Arial font) | 1050k tokens (~1575 A4 pages of size 12 Arial font) | Gemma 4 26B A4B (Reasoning) is smaller than GPT-5.4 (xhigh) |",
    "| Open Source (Weights) | Yes[](https://huggingface.co/google/gemma-4-26B-A4B-it) | No | Gemma 4 26B A4B (Reasoning) is open source while GPT-5.4 (xhigh) is proprietary |"
   ]
  },
  {
   "url": "https://makerstack.co/reviews/inkling-review",
   "title": "Inkling Review (2026): Pricing, Features & Honest Verdict",
   "published_at": "2026-07-26T03:58:23",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "A 975B-parameter multimodal mixture-of-experts release with a 1M token context window and controllable thinking effort.",
   "organization_evidence": [
    "# Inkling Review (2026): Pricing, Features & Honest Verdict",
    "Inkling is Thinking Machines Lab\u2019s first open-weights model, a 975B-parameter multimodal mixture-of-experts release under Apache 2.0."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Inkling Review (2026): Pricing, Features & Honest Verdict",
    "Inkling is a large [multimodal language model](https://makerstack.co/reviews/qwen-35-omni-review/) released on 15 July 2026 by Thinking Machines Lab, the company [Mira](https://makerstack.co/reviews/mira-ai-moderator-review/) Murati founded after leaving OpenAI as CTO."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Inkling Review (2026): Pricing, Features & Honest Verdict",
    "Inkling is Thinking Machines Lab\u2019s first open-weights model, a 975B-parameter multimodal mixture-of-experts release under Apache 2.0."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/grok-4-20-vs-nvidia-nemotron-3-super-120b-a12b",
   "title": "Grok 4.20 0309 v2 (Reasoning) vs NVIDIA Nemotron 3 Super 120B A12B (Reasoning): Model Comparison",
   "published_at": null,
   "organization": "NVIDIA",
   "model": null,
   "release_date": "2026-03",
   "access_type": null,
   "license": null,
   "significance": "120B A12B reasoning model with open source weights and 1000k token context window",
   "organization_evidence": [
    "Comparison between Grok 4.20 0309 v2 (Reasoning) and NVIDIA Nemotron 3 Super 120B A12B (Reasoning) across intelligence, price, speed, context window and more.",
    "| Metric | ![xAI logo](/img/logos/xai.svg)Grok 4.20 0309 v2 (Reasoning) | ![NVIDIA logo](/img/logos/nvidia_small.svg)NVIDIA Nemotron 3 Super 120B A12B (Reasoning) | Analysis |",
    "| Creator | ![xAI](/img/logos/xai.svg)[](https://x.ai/) | ![NVIDIA](/img/logos/nvidia_small.svg)[](https://www.nvidia.com/) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Comparison between Grok 4.20 0309 v2 (Reasoning) and NVIDIA Nemotron 3 Super 120B A12B (Reasoning) across intelligence, price, speed, context window and more.",
    "| Metric | ![xAI logo](/img/logos/xai.svg)Grok 4.20 0309 v2 (Reasoning) | ![NVIDIA logo](/img/logos/nvidia_small.svg)NVIDIA Nemotron 3 Super 120B A12B (Reasoning) | Analysis |",
    "| Release Date | April, 2026 | March, 2026 | Grok 4.20 0309 v2 (Reasoning) has a more recent release date than NVIDIA Nemotron 3 Super 120B A12B (Reasoning) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Comparison between Grok 4.20 0309 v2 (Reasoning) and NVIDIA Nemotron 3 Super 120B A12B (Reasoning) across intelligence, price, speed, context window and more.",
    "| Metric | ![xAI logo](/img/logos/xai.svg)Grok 4.20 0309 v2 (Reasoning) | ![NVIDIA logo](/img/logos/nvidia_small.svg)NVIDIA Nemotron 3 Super 120B A12B (Reasoning) | Analysis |",
    "| Context Window | 2000k tokens (~3000 A4 pages of size 12 Arial font) | 1000k tokens (~1500 A4 pages of size 12 Arial font) | Grok 4.20 0309 v2 (Reasoning) is larger than NVIDIA Nemotron 3 Super 120B A12B (Reasoning) |",
    "| Open Source (Weights) | No | Yes[](https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16) | NVIDIA Nemotron 3 Super 120B A12B (Reasoning) is open source while Grok 4.20 0309 v2 (Reasoning) is proprietary |"
   ]
  },
  {
   "url": "https://clauxel.com/llm",
   "title": "Latest LLM | clauxel",
   "published_at": "2026-08-01T00:00:00",
   "organization": "Huawei / Ascend Tribe",
   "model": null,
   "release_date": "2026-07-01",
   "access_type": null,
   "license": null,
   "significance": "Ascend-native long-context reasoning, code, and enterprise agent work",
   "organization_evidence": [
    "Ascend-native long-context reasoning, code, and enterprise agent work",
    "## openPangu-2.0-Pro",
    "**Huawei / Ascend Tribe** - Ascend-native long-context reasoning, code, and enterprise agent work"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "July 2026. 512K context. Ascend Tribe model repository is live.",
    "## openPangu-2.0-Pro"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Ascend-native long-context reasoning, code, and enterprise agent work",
    "## openPangu-2.0-Pro",
    "**Huawei / Ascend Tribe** - Ascend-native long-context reasoning, code, and enterprise agent work"
   ]
  },
  {
   "url": "https://clauxel.com/llm",
   "title": "Latest LLM | clauxel",
   "published_at": "2026-08-01T00:00:00",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-01",
   "access_type": null,
   "license": null,
   "significance": "Long-horizon coding, end-to-end knowledge work, and native vision",
   "organization_evidence": [
    "Long-horizon coding, end-to-end knowledge work, and native vision",
    "## Kimi K3",
    "**Moonshot AI** - Long-horizon coding, end-to-end knowledge work, and native vision"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "July 2026. 1M context. Open-weight Kimi K3 repository is public.",
    "## Kimi K3"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Long-horizon coding, end-to-end knowledge work, and native vision",
    "## Kimi K3",
    "**Moonshot AI** - Long-horizon coding, end-to-end knowledge work, and native vision"
   ]
  },
  {
   "url": "https://saassentinel.com/2026/06/30/meituan-open-sources-longcat-2-0-a-1-6t-coding-model-trained-entirely-on-chinese-chips",
   "title": "Meituan Open Sources LongCat-2.0, a 1.6T Coding Model Trained Entirely on Chinese Chips",
   "published_at": "2026-06-30T09:09:14",
   "organization": "Meituan",
   "model": null,
   "release_date": "2026-06-30",
   "access_type": null,
   "license": null,
   "significance": "A 1.6 trillion parameter Mixture-of-Experts coding model trained entirely on Chinese chips that outscores GPT-5.5 on SWE-bench Pro.",
   "organization_evidence": [
    "# Meituan Open Sources LongCat-2.0, a 1.6T Coding Model Trained Entirely on Chinese Chips",
    "Meituan has open sourced LongCat-2.0, a 1.6 trillion parameter Mixture-of-Experts coding model that spent two months topping global developer charts under a secret alias."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The company published the model on GitHub and Hugging Face on June 30, 2026.",
    "# Meituan Open Sources LongCat-2.0, a 1.6T Coding Model Trained Entirely on Chinese Chips"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "- LongCat-2.0 has 1.6 trillion total parameters and outscores GPT-5.5 on SWE-bench Pro with a 59.5 versus 58.6 rating.",
    "Meituan has open sourced LongCat-2.0, a 1.6 trillion parameter Mixture-of-Experts coding model that spent two months topping global developer charts under a secret alias.",
    "# Meituan Open Sources LongCat-2.0, a 1.6T Coding Model Trained Entirely on Chinese Chips"
   ]
  },
  {
   "url": "https://fungies.io/top-open-source-llms-local-inference-2026",
   "title": "10 Best Open Source LLMs for Local Inference in 2026: Complete Benchmark Comparison",
   "published_at": "2026-07-01T22:58:00",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-04-07",
   "access_type": null,
   "license": null,
   "significance": "first open-source model to top SWE-Bench Pro with a score of 58.4%",
   "organization_evidence": [
    "10 Best Open Source LLMs for Local Inference in 2026: Complete Benchmark Comparison",
    "1. GLM-5.1 \u2014 Best for Coding (Z.ai)",
    "GLM-5.1 from Z.ai (formerly Zhipu AI) is the current king of open-source coding models."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "10 Best Open Source LLMs for Local Inference in 2026: Complete Benchmark Comparison",
    "1. GLM-5.1 \u2014 Best for Coding (Z.ai)",
    "Released April 7, 2026, it scored 58.4% on SWE-Bench Pro \u2014 the first open model to surpass Claude Opus 4.6 and GPT-5.4 on this benchmark."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "10 Best Open Source LLMs for Local Inference in 2026: Complete Benchmark Comparison",
    "1. GLM-5.1 \u2014 Best for Coding (Z.ai)",
    "On April 7, 2026, Z.ai\u2019s GLM-5.1 became the first open-source model to top SWE-Bench Pro with a score of 58.4% \u2014 beating Claude Opus 4.6 (57.3%) and GPT-5.4 (57.7%)."
   ]
  },
  {
   "url": "https://fungies.io/top-open-source-llms-local-inference-2026",
   "title": "10 Best Open Source LLMs for Local Inference in 2026: Complete Benchmark Comparison",
   "published_at": "2026-07-01T22:58:00",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "1.6 trillion parameter MoE model achieving 80.6% on SWE-Bench Verified",
   "organization_evidence": [
    "10 Best Open Source LLMs for Local Inference in 2026: Complete Benchmark Comparison",
    "2. DeepSeek V4 Pro \u2014 Best for Reasoning (DeepSeek)",
    "DeepSeek V4 Pro is a 1.6 trillion parameter MoE model with only 49B active parameters per token."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "10 Best Open Source LLMs for Local Inference in 2026: Complete Benchmark Comparison",
    "2. DeepSeek V4 Pro \u2014 Best for Reasoning (DeepSeek)",
    "Released April 24, 2026, it achieves 80.6% on SWE-Bench Verified \u2014 the highest open-weights score, tied with Gemini 3.1 Pro."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "10 Best Open Source LLMs for Local Inference in 2026: Complete Benchmark Comparison",
    "2. DeepSeek V4 Pro \u2014 Best for Reasoning (DeepSeek)",
    "DeepSeek V4 Pro is a 1.6 trillion parameter MoE model with only 49B active parameters per token."
   ]
  },
  {
   "url": "https://fungies.io/top-open-source-llms-local-inference-2026",
   "title": "10 Best Open Source LLMs for Local Inference in 2026: Complete Benchmark Comparison",
   "published_at": "2026-07-01T22:58:00",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-06",
   "access_type": null,
   "license": null,
   "significance": "strongest open-source agentic model with vendor-reported SWE-Bench Pro of 58.6%",
   "organization_evidence": [
    "10 Best Open Source LLMs for Local Inference in 2026: Complete Benchmark Comparison",
    "3. Kimi K2.7 Code \u2014 Best for Agentic Workflows (Moonshot AI)",
    "Kimi K2.7 Code is Moonshot AI\u2019s open-weight flagship released June 2026."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "10 Best Open Source LLMs for Local Inference in 2026: Complete Benchmark Comparison",
    "3. Kimi K2.7 Code \u2014 Best for Agentic Workflows (Moonshot AI)",
    "Kimi K2.7 Code is Moonshot AI\u2019s open-weight flagship released June 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "10 Best Open Source LLMs for Local Inference in 2026: Complete Benchmark Comparison",
    "3. Kimi K2.7 Code \u2014 Best for Agentic Workflows (Moonshot AI)",
    "It\u2019s currently the strongest open-source agentic model on public benchmarks, with vendor-reported SWE-Bench Pro of 58.6%."
   ]
  },
  {
   "url": "https://fungies.io/top-open-source-llms-local-inference-2026",
   "title": "10 Best Open Source LLMs for Local Inference in 2026: Complete Benchmark Comparison",
   "published_at": "2026-07-01T22:58:00",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "first open-weights model to combine frontier coding, a 1-million-token context window, and native multimodality",
   "organization_evidence": [
    "10 Best Open Source LLMs for Local Inference in 2026: Complete Benchmark Comparison",
    "4. MiniMax M3 \u2014 Best Multimodal Model (MiniMax)",
    "MiniMax M3 launched June 1, 2026 as the first open-weights model to combine frontier coding, a 1-million-token context window, and native multimodality."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "10 Best Open Source LLMs for Local Inference in 2026: Complete Benchmark Comparison",
    "4. MiniMax M3 \u2014 Best Multimodal Model (MiniMax)",
    "MiniMax M3 launched June 1, 2026 as the first open-weights model to combine frontier coding, a 1-million-token context window, and native multimodality."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "10 Best Open Source LLMs for Local Inference in 2026: Complete Benchmark Comparison",
    "4. MiniMax M3 \u2014 Best Multimodal Model (MiniMax)",
    "MiniMax M3 launched June 1, 2026 as the first open-weights model to combine frontier coding, a 1-million-token context window, and native multimodality."
   ]
  },
  {
   "url": "https://fungies.io/top-open-source-llms-local-inference-2026",
   "title": "10 Best Open Source LLMs for Local Inference in 2026: Complete Benchmark Comparison",
   "published_at": "2026-07-01T22:58:00",
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-02",
   "access_type": null,
   "license": null,
   "significance": "235B parameter MoE model excelling at multilingual tasks",
   "organization_evidence": [
    "10 Best Open Source LLMs for Local Inference in 2026: Complete Benchmark Comparison",
    "6. Qwen 3.5 235B-A22B \u2014 Best for Multilingual (Alibaba)",
    "Qwen 3.5 from Alibaba is a 235B parameter MoE model with 22B active parameters."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "10 Best Open Source LLMs for Local Inference in 2026: Complete Benchmark Comparison",
    "6. Qwen 3.5 235B-A22B \u2014 Best for Multilingual (Alibaba)",
    "Released February 2026, it excels at multilingual tasks and offers competitive coding performance."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "10 Best Open Source LLMs for Local Inference in 2026: Complete Benchmark Comparison",
    "6. Qwen 3.5 235B-A22B \u2014 Best for Multilingual (Alibaba)",
    "Qwen 3.5 from Alibaba is a 235B parameter MoE model with 22B active parameters."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/glm-5-1-non-reasoning-vs-mistral-small-4",
   "title": "GLM-5.1 (Non-reasoning) vs Mistral Small 4 (Reasoning): Model Comparison",
   "published_at": null,
   "organization": "Z AI",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "744B, 40B active at inference time",
   "organization_evidence": [
    "# GLM-5.1 (Non-reasoning) vs. Mistral Small 4 (Reasoning)",
    "## Model Comparison",
    "| Creator | ![Z AI](/img/logos/zai_small.svg)[](https://z.ai/) | ![Mistral](/img/logos/mistral_small.png)[](https://mistral.ai/) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# GLM-5.1 (Non-reasoning) vs. Mistral Small 4 (Reasoning)",
    "## Model Comparison",
    "| Release Date | April, 2026 | March, 2026 | GLM-5.1 (Non-reasoning) has a more recent release date than Mistral Small 4 (Reasoning) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# GLM-5.1 (Non-reasoning) vs. Mistral Small 4 (Reasoning)",
    "## Model Comparison",
    "| Parameters | 744B, 40B active at inference time | 119B, 6.5B active at inference time | GLM-5.1 (Non-reasoning) is larger than Mistral Small 4 (Reasoning) |"
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/glm-5-1-non-reasoning-vs-mistral-small-4",
   "title": "GLM-5.1 (Non-reasoning) vs Mistral Small 4 (Reasoning): Model Comparison",
   "published_at": null,
   "organization": "Mistral",
   "model": null,
   "release_date": "2026-03",
   "access_type": null,
   "license": null,
   "significance": "119B, 6.5B active at inference time",
   "organization_evidence": [
    "# GLM-5.1 (Non-reasoning) vs. Mistral Small 4 (Reasoning)",
    "## Model Comparison",
    "| Creator | ![Z AI](/img/logos/zai_small.svg)[](https://z.ai/) | ![Mistral](/img/logos/mistral_small.png)[](https://mistral.ai/) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# GLM-5.1 (Non-reasoning) vs. Mistral Small 4 (Reasoning)",
    "## Model Comparison",
    "| Release Date | April, 2026 | March, 2026 | GLM-5.1 (Non-reasoning) has a more recent release date than Mistral Small 4 (Reasoning) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# GLM-5.1 (Non-reasoning) vs. Mistral Small 4 (Reasoning)",
    "## Model Comparison",
    "| Parameters | 744B, 40B active at inference time | 119B, 6.5B active at inference time | GLM-5.1 (Non-reasoning) is larger than Mistral Small 4 (Reasoning) |"
   ]
  },
  {
   "url": "https://theearlierstuff.com/general/how-china-s-rapid-ai-releases-are-reshaping-the-industry-four-frontier-class-ope",
   "title": "How China's Rapid AI Releases Are Reshaping The Industry: Four Frontier-Class Open Models In Eight Weeks - The Earlier Stuff",
   "published_at": "2026-07-14T14:24:12",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor",
   "organization_evidence": [
    "TL;DR",
    "Four Chinese laboratories released frontier-class open-weight AI models between April 24 and mid-June 2026, according to a July 13 market report from Thorsten Meyer AI.",
    "The sequence began with **DeepSeek V4** on April 24, followed by **MiniMax M3** on June 1."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "TL;DR",
    "Four Chinese laboratories released frontier-class open-weight AI models between April 24 and mid-June 2026, according to a July 13 market report from Thorsten Meyer AI.",
    "The sequence began with **DeepSeek V4** on April 24, followed by **MiniMax M3** on June 1."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "TL;DR",
    "Four Chinese laboratories released frontier-class open-weight AI models between April 24 and mid-June 2026, according to a July 13 market report from Thorsten Meyer AI.",
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor"
   ]
  },
  {
   "url": "https://theearlierstuff.com/general/how-china-s-rapid-ai-releases-are-reshaping-the-industry-four-frontier-class-ope",
   "title": "How China's Rapid AI Releases Are Reshaping The Industry: Four Frontier-Class Open Models In Eight Weeks - The Earlier Stuff",
   "published_at": "2026-07-14T14:24:12",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "cheap 1M-token context, native multimodal, modified-MIT",
   "organization_evidence": [
    "TL;DR",
    "Four Chinese laboratories released frontier-class open-weight AI models between April 24 and mid-June 2026, according to a July 13 market report from Thorsten Meyer AI.",
    "The sequence began with **DeepSeek V4** on April 24, followed by **MiniMax M3** on June 1."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "TL;DR",
    "Four Chinese laboratories released frontier-class open-weight AI models between April 24 and mid-June 2026, according to a July 13 market report from Thorsten Meyer AI.",
    "The sequence began with **DeepSeek V4** on April 24, followed by **MiniMax M3** on June 1."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "TL;DR",
    "Four Chinese laboratories released frontier-class open-weight AI models between April 24 and mid-June 2026, according to a July 13 market report from Thorsten Meyer AI."
   ]
  },
  {
   "url": "https://theearlierstuff.com/general/how-china-s-rapid-ai-releases-are-reshaping-the-industry-four-frontier-class-ope",
   "title": "How China's Rapid AI Releases Are Reshaping The Industry: Four Frontier-Class Open Models In Eight Weeks - The Earlier Stuff",
   "published_at": "2026-07-14T14:24:12",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "agent-run specialist, ~30% fewer thinking tokens than K2.6",
   "organization_evidence": [
    "TL;DR",
    "Four Chinese laboratories released frontier-class open-weight AI models between April 24 and mid-June 2026, according to a July 13 market report from Thorsten Meyer AI.",
    "Moonshot AI\u2019s **Kimi K2.7-Code** and Z.ai\u2019s **GLM-5.2** arrived within days of each other in mid-June, the report said."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "TL;DR",
    "Four Chinese laboratories released frontier-class open-weight AI models between April 24 and mid-June 2026, according to a July 13 market report from Thorsten Meyer AI.",
    "Moonshot AI\u2019s **Kimi K2.7-Code** and Z.ai\u2019s **GLM-5.2** arrived within days of each other in mid-June, the report said.",
    "JUN 13\n\n**Kimi K2.7-Code (Moonshot)**agent-run specialist, ~30% fewer thinking tokens than K2.6"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "TL;DR",
    "Four Chinese laboratories released frontier-class open-weight AI models between April 24 and mid-June 2026, according to a July 13 market report from Thorsten Meyer AI.",
    "JUN 13\n\n**Kimi K2.7-Code (Moonshot)**agent-run specialist, ~30% fewer thinking tokens than K2.6"
   ]
  },
  {
   "url": "https://theearlierstuff.com/general/how-china-s-rapid-ai-releases-are-reshaping-the-industry-four-frontier-class-ope",
   "title": "How China's Rapid AI Releases Are Reshaping The Industry: Four Frontier-Class Open Models In Eight Weeks - The Earlier Stuff",
   "published_at": "2026-07-14T14:24:12",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "753B MoE, MIT, top open-weight on Artificial Analysis index",
   "organization_evidence": [
    "TL;DR",
    "Four Chinese laboratories released frontier-class open-weight AI models between April 24 and mid-June 2026, according to a July 13 market report from Thorsten Meyer AI.",
    "Moonshot AI\u2019s **Kimi K2.7-Code** and Z.ai\u2019s **GLM-5.2** arrived within days of each other in mid-June, the report said."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "TL;DR",
    "Four Chinese laboratories released frontier-class open-weight AI models between April 24 and mid-June 2026, according to a July 13 market report from Thorsten Meyer AI.",
    "Moonshot AI\u2019s **Kimi K2.7-Code** and Z.ai\u2019s **GLM-5.2** arrived within days of each other in mid-June, the report said.",
    "JUN 13\u201316\n\n**GLM-5.2 (Z.ai)**753B MoE, MIT, top open-weight on Artificial Analysis index"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "TL;DR",
    "Four Chinese laboratories released frontier-class open-weight AI models between April 24 and mid-June 2026, according to a July 13 market report from Thorsten Meyer AI.",
    "JUN 13\u201316\n\n**GLM-5.2 (Z.ai)**753B MoE, MIT, top open-weight on Artificial Analysis index"
   ]
  },
  {
   "url": "https://tpsreport.news/news/thinking-machines-inkling-small-276b-moe",
   "title": "Inkling Small: Thinking Machines' New Open-Weight MoE Model",
   "published_at": "2026-07-31T15:20:33",
   "organization": "Thinking Machines Lab",
   "model": "Inkling Small",
   "release_date": "2026-07-30",
   "access_type": "open-weight",
   "license": null,
   "significance": "multimodal mixture-of-experts model with 12B active parameters out of 276B total and a 524K token context window",
   "organization_evidence": [
    "model release",
    "Thinking Machines Lab Releases Inkling Small: 276B MoE Model with 524K Context Window",
    "Thinking Machines Lab has released Inkling Small, an open-weight multimodal mixture-of-experts model with 12B active parameters out of 276B total and a 524K token context window."
   ],
   "model_evidence": [
    "model release",
    "Thinking Machines Lab Releases Inkling Small: 276B MoE Model with 524K Context Window",
    "Thinking Machines Lab has released Inkling Small, an open-weight multimodal mixture-of-experts model with 12B active parameters out of 276B total and a 524K token context window."
   ],
   "release_date_evidence": [
    "model release",
    "Thinking Machines Lab Releases Inkling Small: 276B MoE Model with 524K Context Window",
    "The model's release date is listed as July 30, 2026."
   ],
   "access_type_evidence": [
    "model release",
    "Thinking Machines Lab Releases Inkling Small: 276B MoE Model with 524K Context Window",
    "Thinking Machines Lab has released Inkling Small, an open-weight multimodal mixture-of-experts model with 12B active parameters out of 276B total and a 524K token context window."
   ],
   "license_evidence": null,
   "significance_evidence": [
    "model release",
    "Thinking Machines Lab Releases Inkling Small: 276B MoE Model with 524K Context Window",
    "Thinking Machines Lab has released Inkling Small, an open-weight multimodal mixture-of-experts model with 12B active parameters out of 276B total and a 524K token context window."
   ]
  },
  {
   "url": "https://ainchina.com/blog/kimi-k3-moonshot-2-8-trillion-open-weight-model-2026",
   "title": "Kimi K3: The 2.8 Trillion-Parameter Open-Weight Model That Moved the Nasdaq",
   "published_at": "2026-07-26T00:00:00",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-17",
   "access_type": null,
   "license": null,
   "significance": "2.8T parameters, open weights",
   "organization_evidence": [
    "Kimi K3: The 2.8 Trillion-Parameter Open-Weight Model That Moved the Nasdaq",
    "The culprit wasn't a Fed announcement or a geopolitical crisis. It was a press release from a Beijing startup called Moonshot AI."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Artificial Intelligence10 min read",
    "# Kimi K3: The 2.8 Trillion-Parameter Open-Weight Model That Moved the Nasdaq",
    "July 26, 2026\u00b7AI in China",
    "At 12:47 AM on July 17, 2026, the trading floor at the Nasdaq was quiet."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "| Milestone | Date | Significance |",
    "| **Kimi K3** | **Jul 17, 2026** | **2.8T parameters, open weights** |",
    "*Table: Moonshot AI key milestones, 2023\u20132026. Source: Company announcements, TechCrunch, Xinhua.*"
   ]
  },
  {
   "url": "https://kanerika.com/blogs/open-source-llms-models",
   "title": "10 Best Open Source LLMs to Evaluate in 2026",
   "published_at": "2026-06-16T13:58:04",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-16",
   "access_type": null,
   "license": null,
   "significance": "Ranked #1 among open-weight models and roughly #4 overall on the Artificial Analysis Index",
   "organization_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "### 1. Kimi K3 (Moonshot AI)",
    "Moonshot AI released Kimi K3 on July 16, 2026, a 2.8 trillion-parameter Mixture-of-Experts model distributed under a Modified MIT license that allows self-hosting and commercial use."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "### 1. Kimi K3 (Moonshot AI)",
    "Moonshot AI released Kimi K3 on July 16, 2026, a 2.8 trillion-parameter Mixture-of-Experts model distributed under a Modified MIT license that allows self-hosting and commercial use."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "### 1. Kimi K3 (Moonshot AI)",
    "On the Artificial Analysis Index, it currently ranks #1 among open-weight models and roughly #4 overall, ahead of several closed models, making it the first Chinese model to break into that frontier tier."
   ]
  },
  {
   "url": "https://kanerika.com/blogs/open-source-llms-models",
   "title": "10 Best Open Source LLMs to Evaluate in 2026",
   "published_at": "2026-06-16T13:58:04",
   "organization": "Meta",
   "model": null,
   "release_date": "2026-04-05",
   "access_type": null,
   "license": null,
   "significance": "Scout features a 10M token context window and Maverick scales to 128 experts and 400B total parameters",
   "organization_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "### 2. Llama 4 Scout and Maverick",
    "Meta released [Llama 4 Scout and Maverick](https://ai.meta.com/blog/llama-4-multimodal-intelligence/) on April 5, 2026, the first Llama models built on MoE architecture Meta released [Llama 4 Scout and Maverick](https://ai.meta.com/blog/llama-4-multimodal-intelligence/) on April 5, 2026, the first Llama models built on MoE architecture and trained as natively multimodal systems using early fusion, processing text, images, and video through a unified model."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "### 2. Llama 4 Scout and Maverick",
    "Meta released [Llama 4 Scout and Maverick](https://ai.meta.com/blog/llama-4-multimodal-intelligence/) on April 5, 2026, the first Llama models built on MoE architecture Meta released [Llama 4 Scout and Maverick](https://ai.meta.com/blog/llama-4-multimodal-intelligence/) on April 5, 2026, the first Llama models built on MoE architecture and trained as natively multimodal systems using early fusion, processing text, images, and video through a unified model."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "### 2. Llama 4 Scout and Maverick",
    "Scout runs on a single H100 GPU with a [10 million token context window](https://www.llama.com/models/llama-4/), currently the largest in any open-weight model."
   ]
  },
  {
   "url": "https://kanerika.com/blogs/open-source-llms-models",
   "title": "10 Best Open Source LLMs to Evaluate in 2026",
   "published_at": "2026-06-16T13:58:04",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-08-12",
   "access_type": null,
   "license": null,
   "significance": "Most cost-effective frontier-class open model with a 1M token context window",
   "organization_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "### 3. DeepSeek V4",
    "DeepSeek V4 comes in two sizes: [V4-Pro (1.6T total, 49B active parameters)](https://api-docs.deepseek.com/) for maximum performance, which reached general availability around August 12-13, 2026, and V4-Flash (304B total, 13B active), released July 31, 2026, as a lighter, cheaper alternative."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "### 3. DeepSeek V4",
    "DeepSeek V4 comes in two sizes: [V4-Pro (1.6T total, 49B active parameters)](https://api-docs.deepseek.com/) for maximum performance, which reached general availability around August 12-13, 2026, and V4-Flash (304B total, 13B active), released July 31, 2026, as a lighter, cheaper alternative."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "### 3. DeepSeek V4",
    "Both expose a 1M token context window and carry MIT licensing for full self-hosting."
   ]
  },
  {
   "url": "https://kanerika.com/blogs/open-source-llms-models",
   "title": "10 Best Open Source LLMs to Evaluate in 2026",
   "published_at": "2026-06-16T13:58:04",
   "organization": "Alibaba Cloud",
   "model": null,
   "release_date": "2026-02-01",
   "access_type": null,
   "license": null,
   "significance": "Supports 201 languages and delivers 8.6x to 19x higher decoding throughput",
   "organization_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "### 4. Qwen3.6-35B-A3B"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "### 4. Qwen3.6-35B-A3B"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "### 4. Qwen3.6-35B-A3B",
    "The Qwen3.6-35B-A3B is a MoE model with 397B total and 17B active parameters, a 1 million token context window, and native multimodality across text, image, and video through early fusion architecture."
   ]
  },
  {
   "url": "https://kanerika.com/blogs/open-source-llms-models",
   "title": "10 Best Open Source LLMs to Evaluate in 2026",
   "published_at": "2026-06-16T13:58:04",
   "organization": "Google DeepMind",
   "model": null,
   "release_date": "2026-04-02",
   "access_type": null,
   "license": null,
   "significance": "Comes in four sizes and ranks #3 among all open models on Arena AI",
   "organization_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "### 5. Gemma 4",
    "Google DeepMind released [Gemma 4](https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/) on April 2, 2026 under Apache 2.0 with full commercial freedom."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "### 5. Gemma 4",
    "Google DeepMind released [Gemma 4](https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/) on April 2, 2026 under Apache 2.0 with full commercial freedom."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "### 5. Gemma 4",
    "Built from Gemini 3 research, it comes in four sizes: E2B and E4B for edge and mobile, 26B MoE, and 31B Dense for server workloads."
   ]
  },
  {
   "url": "https://kanerika.com/blogs/open-source-llms-models",
   "title": "10 Best Open Source LLMs to Evaluate in 2026",
   "published_at": "2026-06-16T13:58:04",
   "organization": "Mistral AI",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Optimized for low latency, delivering 150 tokens per second on consumer-grade hardware",
   "organization_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "### 7. Mistral Small 4",
    "Mistral Small 4\n\n[Mist](https://mistral.ai/news/mistral-small-3-1)[Mistral Small 4](https://mistral.ai/news/mistral-small-3-1) is built for real-time applications where response speed and low hardware requirements take priority."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "### 7. Mistral Small 4",
    "Key capabilities:",
    "- Optimized for low latency, delivering [150 tokens per second](https://mistral.ai/news/mistral-small-3-1)"
   ]
  },
  {
   "url": "https://kanerika.com/blogs/open-source-llms-models",
   "title": "10 Best Open Source LLMs to Evaluate in 2026",
   "published_at": "2026-06-16T13:58:04",
   "organization": "Z.ai (formerly Zhipu AI)",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "744B total, 40B active parameters via MoE, 200K context window for long-horizon agentic engineering tasks",
   "organization_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "### 9. GLM-5.2 (Zhipu AI)"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "### 9. GLM-5.2 (Zhipu AI)"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "### 9. GLM-5.2 (Zhipu AI)",
    "With 744B total and 40B active parameters and a 200K context window, it is MIT licensed and has posted some of the strongest SWE-bench Pro scores in the open-weight category."
   ]
  },
  {
   "url": "https://kanerika.com/blogs/open-source-llms-models",
   "title": "10 Best Open Source LLMs to Evaluate in 2026",
   "published_at": "2026-06-16T13:58:04",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "Combines a 1M token context window, native multimodality, and frontier coding ability",
   "organization_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "### 10. MiniMax-M3",
    "[MiniMax M3](https://huggingface.co/MiniMaxAI/MiniMax-M3), released June 1, 2026, is the most capable open-weight model MiniMax has shipped."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "### 10. MiniMax-M3",
    "[MiniMax M3](https://huggingface.co/MiniMaxAI/MiniMax-M3), released June 1, 2026, is the most capable open-weight model MiniMax has shipped."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Top 10 Open-Source LLMs in 2026",
    "### 10. MiniMax-M3",
    "It combines a 1M token context window, native multimodality, and frontier coding ability in a single model, topping open-weight SWE-Bench Pro at 59.0%."
   ]
  },
  {
   "url": "https://tech-insider.org/nvidia-nemotron-3-ultra-open-weight-2026",
   "title": "Nvidia Nemotron 3 Ultra: 550B Params, Trails Rivals [2026]",
   "published_at": "2026-08-05T10:16:56",
   "organization": "Nvidia",
   "model": "Nvidia Nemotron 3 Ultra",
   "release_date": "2026-06-04",
   "access_type": "open-weight",
   "license": "OpenMDW-1.1",
   "significance": "A 550-billion-parameter hybrid Mamba-Attention Mixture-of-Experts model with a 1-million-token context window aimed at agentic workloads and coding.",
   "organization_evidence": [
    "**Nvidia Nemotron 3 Ultra**, a 550-billion-parameter open-weight language model, is now sitting on Hugging Face for anyone to download, fine-tune, or deploy, no API key required.",
    "## What Is Nemotron 3 Ultra? Nvidia\u2019s Largest Open Model Yet"
   ],
   "model_evidence": [
    "**Nvidia Nemotron 3 Ultra**, a 550-billion-parameter open-weight language model, is now sitting on Hugging Face for anyone to download, fine-tune, or deploy, no API key required.",
    "## What Is Nemotron 3 Ultra? Nvidia\u2019s Largest Open Model Yet"
   ],
   "release_date_evidence": [
    "On June 4, 2026, it made a different kind of statement: it gave away the model too.",
    "The model launched on June 4, 2026 as the flagship of the Nemotron 3 family, which also includes smaller Nano and Super variants.",
    "## What Is Nemotron 3 Ultra? Nvidia\u2019s Largest Open Model Yet"
   ],
   "access_type_evidence": [
    "**Nvidia Nemotron 3 Ultra**, a 550-billion-parameter open-weight language model, is now sitting on Hugging Face for anyone to download, fine-tune, or deploy, no API key required.",
    "## What Is Nemotron 3 Ultra? Nvidia\u2019s Largest Open Model Yet",
    "Nvidia Nemotron 3 Ultra (BF16)"
   ],
   "license_evidence": [
    "Weights are published on Hugging Face in both BF16 and NVFP4 (Nvidia\u2019s 4-bit floating point format) under a permissive open license Nvidia calls OpenMDW-1.1, which secondary coverage describes as following Linux Foundation-style terms rather than a research-only license.",
    "## What Is Nemotron 3 Ultra? Nvidia\u2019s Largest Open Model Yet",
    "The model weights are free to download from Hugging Face under Nvidia\u2019s permissive OpenMDW-1.1 license."
   ],
   "significance_evidence": [
    "**Nvidia Nemotron 3 Ultra**, a 550-billion-parameter open-weight language model, is now sitting on Hugging Face for anyone to download, fine-tune, or deploy, no API key required.",
    "## What Is Nemotron 3 Ultra? Nvidia\u2019s Largest Open Model Yet"
   ]
  },
  {
   "url": "https://tech-insider.org/nvidia-nemotron-3-ultra-open-weight-2026",
   "title": "Nvidia Nemotron 3 Ultra: 550B Params, Trails Rivals [2026]",
   "published_at": "2026-08-05T10:16:56",
   "organization": "MiniMax",
   "model": "MiniMax M3",
   "release_date": "2026-06-01",
   "access_type": "open-weight",
   "license": null,
   "significance": "An open-weight model with a 1-million-token context window and a reported 59.0% score on SWE-bench Pro.",
   "organization_evidence": [
    "MiniMax shipped MiniMax M3 on June 1, an open-weight model with a 1-million-token context window and a reported 59.0% score on SWE-bench Pro.",
    "## A Crowded Month: MiniMax M3, Kimi K2.7 and GLM-5.2 Also Shipped"
   ],
   "model_evidence": [
    "MiniMax shipped MiniMax M3 on June 1, an open-weight model with a 1-million-token context window and a reported 59.0% score on SWE-bench Pro.",
    "## A Crowded Month: MiniMax M3, Kimi K2.7 and GLM-5.2 Also Shipped"
   ],
   "release_date_evidence": [
    "MiniMax shipped MiniMax M3 on June 1, an open-weight model with a 1-million-token context window and a reported 59.0% score on SWE-bench Pro.",
    "## A Crowded Month: MiniMax M3, Kimi K2.7 and GLM-5.2 Also Shipped"
   ],
   "access_type_evidence": [
    "MiniMax shipped MiniMax M3 on June 1, an open-weight model with a 1-million-token context window and a reported 59.0% score on SWE-bench Pro.",
    "## A Crowded Month: MiniMax M3, Kimi K2.7 and GLM-5.2 Also Shipped"
   ],
   "license_evidence": null,
   "significance_evidence": [
    "MiniMax shipped MiniMax M3 on June 1, an open-weight model with a 1-million-token context window and a reported 59.0% score on SWE-bench Pro.",
    "## A Crowded Month: MiniMax M3, Kimi K2.7 and GLM-5.2 Also Shipped"
   ]
  },
  {
   "url": "https://tech-insider.org/nvidia-nemotron-3-ultra-open-weight-2026",
   "title": "Nvidia Nemotron 3 Ultra: 550B Params, Trails Rivals [2026]",
   "published_at": "2026-08-05T10:16:56",
   "organization": "Moonshot AI",
   "model": "Kimi K2.7 Code",
   "release_date": "2026-06-12",
   "access_type": "open-weight",
   "license": null,
   "significance": "A 1-trillion-parameter MoE model built specifically for coding workloads.",
   "organization_evidence": [
    "Moonshot AI followed on June 12 with Kimi K2.7 Code, a 1-trillion-parameter MoE model built specifically for coding workloads and reportedly using 30% fewer \u201cthinking tokens\u201d than its predecessor to reach similar answers.",
    "## A Crowded Month: MiniMax M3, Kimi K2.7 and GLM-5.2 Also Shipped"
   ],
   "model_evidence": [
    "Moonshot AI followed on June 12 with Kimi K2.7 Code, a 1-trillion-parameter MoE model built specifically for coding workloads and reportedly using 30% fewer \u201cthinking tokens\u201d than its predecessor to reach similar answers.",
    "## A Crowded Month: MiniMax M3, Kimi K2.7 and GLM-5.2 Also Shipped"
   ],
   "release_date_evidence": [
    "Moonshot AI followed on June 12 with Kimi K2.7 Code, a 1-trillion-parameter MoE model built specifically for coding workloads and reportedly using 30% fewer \u201cthinking tokens\u201d than its predecessor to reach similar answers.",
    "## A Crowded Month: MiniMax M3, Kimi K2.7 and GLM-5.2 Also Shipped"
   ],
   "access_type_evidence": [
    "Moonshot AI followed on June 12 with Kimi K2.7 Code, a 1-trillion-parameter MoE model built specifically for coding workloads and reportedly using 30% fewer \u201cthinking tokens\u201d than its predecessor to reach similar answers.",
    "## A Crowded Month: MiniMax M3, Kimi K2.7 and GLM-5.2 Also Shipped"
   ],
   "license_evidence": null,
   "significance_evidence": [
    "Moonshot AI followed on June 12 with Kimi K2.7 Code, a 1-trillion-parameter MoE model built specifically for coding workloads and reportedly using 30% fewer \u201cthinking tokens\u201d than its predecessor to reach similar answers.",
    "## A Crowded Month: MiniMax M3, Kimi K2.7 and GLM-5.2 Also Shipped"
   ]
  },
  {
   "url": "https://tech-insider.org/nvidia-nemotron-3-ultra-open-weight-2026",
   "title": "Nvidia Nemotron 3 Ultra: 550B Params, Trails Rivals [2026]",
   "published_at": "2026-08-05T10:16:56",
   "organization": "Zhipu AI",
   "model": "GLM-5.2",
   "release_date": "2026-06-13",
   "access_type": "open-weight",
   "license": null,
   "significance": "A model with a 1-million-token context window and tiered \u201cHigh\u201d and \u201cMax\u201d effort reasoning modes.",
   "organization_evidence": [
    "Zhipu AI\u2019s GLM-5.2 arrived June 13 with its own 1-million-token context window and tiered \u201cHigh\u201d and \u201cMax\u201d effort reasoning modes.",
    "## A Crowded Month: MiniMax M3, Kimi K2.7 and GLM-5.2 Also Shipped"
   ],
   "model_evidence": [
    "Zhipu AI\u2019s GLM-5.2 arrived June 13 with its own 1-million-token context window and tiered \u201cHigh\u201d and \u201cMax\u201d effort reasoning modes.",
    "## A Crowded Month: MiniMax M3, Kimi K2.7 and GLM-5.2 Also Shipped"
   ],
   "release_date_evidence": [
    "Zhipu AI\u2019s GLM-5.2 arrived June 13 with its own 1-million-token context window and tiered \u201cHigh\u201d and \u201cMax\u201d effort reasoning modes.",
    "## A Crowded Month: MiniMax M3, Kimi K2.7 and GLM-5.2 Also Shipped"
   ],
   "access_type_evidence": [
    "Zhipu AI\u2019s GLM-5.2 arrived June 13 with its own 1-million-token context window and tiered \u201cHigh\u201d and \u201cMax\u201d effort reasoning modes.",
    "## A Crowded Month: MiniMax M3, Kimi K2.7 and GLM-5.2 Also Shipped"
   ],
   "license_evidence": null,
   "significance_evidence": [
    "Zhipu AI\u2019s GLM-5.2 arrived June 13 with its own 1-million-token context window and tiered \u201cHigh\u201d and \u201cMax\u201d effort reasoning modes.",
    "## A Crowded Month: MiniMax M3, Kimi K2.7 and GLM-5.2 Also Shipped"
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/qwen3-6-plus-vs-nvidia-nemotron-3-super-120b-a12b",
   "title": "Qwen3.6 Plus vs Nemotron 3 Super 120B A12B (Reasoning): Model Comparison",
   "published_at": null,
   "organization": "NVIDIA",
   "model": null,
   "release_date": "2026-03",
   "access_type": null,
   "license": null,
   "significance": "A 120B reasoning model with 1M context window and open weights",
   "organization_evidence": [
    "# Qwen3.6 Plus vs. Nemotron 3 Super 120B A12B (Reasoning)",
    "Comparison between Qwen3.6 Plus and Nemotron 3 Super 120B A12B (Reasoning) across intelligence, price, speed, context window and more.",
    "## Model Comparison",
    "| | ![Alibaba logo](/img/logos/alibaba_small.svg)Alibaba Qwen3.6 Plus | ![NVIDIA logo](/img/logos/nvidia_small.svg)NVIDIA Nemotron 3 Super 120B A12B (Reasoning) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Qwen3.6 Plus vs. Nemotron 3 Super 120B A12B (Reasoning)",
    "Comparison between Qwen3.6 Plus and Nemotron 3 Super 120B A12B (Reasoning) across intelligence, price, speed, context window and more.",
    "## Model Comparison",
    "| Release Date | April 2026 | March 2026 | Qwen3.6 Plus has a more recent release date than Nemotron 3 Super 120B A12B (Reasoning) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Qwen3.6 Plus vs. Nemotron 3 Super 120B A12B (Reasoning)",
    "Comparison between Qwen3.6 Plus and Nemotron 3 Super 120B A12B (Reasoning) across intelligence, price, speed, context window and more.",
    "## Model Comparison",
    "| Context Window | 1000k tokens~1,500 A4 pages of size 12 Arial font | 1000k tokens~1,500 A4 pages of size 12 Arial font | Both Qwen3.6 Plus and Nemotron 3 Super 120B A12B (Reasoning) have the same sized context window |",
    "| Reasoning | Yes | Yes | Both Qwen3.6 Plus and Nemotron 3 Super 120B A12B (Reasoning) have reasoning |"
   ]
  },
  {
   "url": "https://thorstenmeyerai.com/insights/signal-four-frontier-class-open-models-in-eight-weeks-chinas-release-cadence-is-the-story",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China's Release Cadence Is the Story - Thorsten Meyer AI",
   "published_at": "2026-07-14T14:00:00",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "resets the price floor with 1.6T total / 49B active MoE and 1M context",
   "organization_evidence": [
    "Between late April and mid-June 2026, Chinese labs shipped **four frontier-class open-weight models in roughly eight weeks**: DeepSeek V4 on April 24, MiniMax M3 on June 1, and Kimi K2.7-Code and GLM-5.2 within days of each other in mid-June.",
    "### The production line \u2014 spring 2026",
    "DeepSeek"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Between late April and mid-June 2026, Chinese labs shipped **four frontier-class open-weight models in roughly eight weeks**: DeepSeek V4 on April 24, MiniMax M3 on June 1, and Kimi K2.7-Code and GLM-5.2 within days of each other in mid-June.",
    "### The production line \u2014 spring 2026"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor",
    "### The production line \u2014 spring 2026"
   ]
  },
  {
   "url": "https://thorstenmeyerai.com/insights/signal-four-frontier-class-open-models-in-eight-weeks-chinas-release-cadence-is-the-story",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China's Release Cadence Is the Story - Thorsten Meyer AI",
   "published_at": "2026-07-14T14:00:00",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "cheap 1M-token context, native multimodal, modified-MIT",
   "organization_evidence": [
    "Between late April and mid-June 2026, Chinese labs shipped **four frontier-class open-weight models in roughly eight weeks**: DeepSeek V4 on April 24, MiniMax M3 on June 1, and Kimi K2.7-Code and GLM-5.2 within days of each other in mid-June.",
    "### The production line \u2014 spring 2026"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Between late April and mid-June 2026, Chinese labs shipped **four frontier-class open-weight models in roughly eight weeks**: DeepSeek V4 on April 24, MiniMax M3 on June 1, and Kimi K2.7-Code and GLM-5.2 within days of each other in mid-June.",
    "### The production line \u2014 spring 2026"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "**MiniMax M3**cheap 1M-token context, native multimodal, modified-MIT",
    "### The production line \u2014 spring 2026"
   ]
  },
  {
   "url": "https://thorstenmeyerai.com/insights/signal-four-frontier-class-open-models-in-eight-weeks-chinas-release-cadence-is-the-story",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China's Release Cadence Is the Story - Thorsten Meyer AI",
   "published_at": "2026-07-14T14:00:00",
   "organization": "Moonshot",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "agent-run specialist, ~30% fewer thinking tokens than K2.6",
   "organization_evidence": [
    "**Moonshot\u2019s Kimi** line is tuned for long-horizon agent stability \u2014 K2.7-Code cuts thinking tokens roughly 30% versus its predecessor, directly attacking the cost of long agent runs.",
    "Two years ago the Chinese open field was one lab deep. Today it\u2019s four \u2014 DeepSeek, Z.ai, Moonshot, Alibaba \u2014 each with a distinct bet:",
    "### The production line \u2014 spring 2026"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Between late April and mid-June 2026, Chinese labs shipped **four frontier-class open-weight models in roughly eight weeks**: DeepSeek V4 on April 24, MiniMax M3 on June 1, and Kimi K2.7-Code and GLM-5.2 within days of each other in mid-June.",
    "### The production line \u2014 spring 2026"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "**Kimi K2.7-Code (Moonshot)**agent-run specialist, ~30% fewer thinking tokens than K2.6",
    "### The production line \u2014 spring 2026"
   ]
  },
  {
   "url": "https://thorstenmeyerai.com/insights/signal-four-frontier-class-open-models-in-eight-weeks-chinas-release-cadence-is-the-story",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China's Release Cadence Is the Story - Thorsten Meyer AI",
   "published_at": "2026-07-14T14:00:00",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "753B MoE, MIT, top open-weight on Artificial Analysis index",
   "organization_evidence": [
    "Two years ago the Chinese open field was one lab deep. Today it\u2019s four \u2014 DeepSeek, Z.ai, Moonshot, Alibaba \u2014 each with a distinct bet:",
    "**Z.ai\u2019s GLM-5.2** holds the open-weight intelligence crown on Artificial Analysis\u2019s independent index.",
    "### The production line \u2014 spring 2026"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Between late April and mid-June 2026, Chinese labs shipped **four frontier-class open-weight models in roughly eight weeks**: DeepSeek V4 on April 24, MiniMax M3 on June 1, and Kimi K2.7-Code and GLM-5.2 within days of each other in mid-June.",
    "### The production line \u2014 spring 2026"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "**GLM-5.2 (Z.ai)**753B MoE, MIT, top open-weight on Artificial Analysis index",
    "### The production line \u2014 spring 2026"
   ]
  },
  {
   "url": "https://fireworks.ai/blog/best-open-source-llms-may-2026",
   "title": "Best Open Source LLMs of May 2026: We Reviewed 7 Models",
   "published_at": "2026-05-19T00:00:00",
   "organization": "MiniMax AI",
   "model": null,
   "release_date": "2026-02-12",
   "access_type": null,
   "license": null,
   "significance": "Real-world productivity and software engineering workflows",
   "organization_evidence": [
    "This roundup compares the top open source LLMs available today, breaking down their trade-offs and ideal applications so you can make an informed decision without running your own eval suite.",
    "## The best open source LLMs at a glance",
    "| MiniMax-M2.5 | February 12, 2026 | 229B total (10B active) | 196,608 tokens | Real-world productivity and software engineering workflows | [Try in playground](https://app.fireworks.ai/playground?model=accounts/fireworks/models/minimax-m2p5&_gl=1*145r14b*_gcl_au*MTY5MjA2MTk2OC4xNzcwMjYzODcy) |",
    "## MiniMax-M2.5",
    "### What is MiniMax-M2.5?",
    "\u2022**Corporate Sponsor:** MiniMax AI"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "This roundup compares the top open source LLMs available today, breaking down their trade-offs and ideal applications so you can make an informed decision without running your own eval suite.",
    "## The best open source LLMs at a glance",
    "| MiniMax-M2.5 | February 12, 2026 | 229B total (10B active) | 196,608 tokens | Real-world productivity and software engineering workflows | [Try in playground](https://app.fireworks.ai/playground?model=accounts/fireworks/models/minimax-m2p5&_gl=1*145r14b*_gcl_au*MTY5MjA2MTk2OC4xNzcwMjYzODcy) |",
    "## MiniMax-M2.5",
    "### What is MiniMax-M2.5?",
    "\u2022**Release Date:** February 12, 2026"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "This roundup compares the top open source LLMs available today, breaking down their trade-offs and ideal applications so you can make an informed decision without running your own eval suite.",
    "## The best open source LLMs at a glance",
    "| MiniMax-M2.5 | February 12, 2026 | 229B total (10B active) | 196,608 tokens | Real-world productivity and software engineering workflows | [Try in playground](https://app.fireworks.ai/playground?model=accounts/fireworks/models/minimax-m2p5&_gl=1*145r14b*_gcl_au*MTY5MjA2MTk2OC4xNzcwMjYzODcy) |"
   ]
  },
  {
   "url": "https://fireworks.ai/blog/best-open-source-llms-may-2026",
   "title": "Best Open Source LLMs of May 2026: We Reviewed 7 Models",
   "published_at": "2026-05-19T00:00:00",
   "organization": "Zhipu AI",
   "model": null,
   "release_date": "2026-02-12",
   "access_type": null,
   "license": null,
   "significance": "Complex systems engineering and long-horizon agentic tasks",
   "organization_evidence": [
    "This roundup compares the top open source LLMs available today, breaking down their trade-offs and ideal applications so you can make an informed decision without running your own eval suite.",
    "## The best open source LLMs at a glance",
    "| GLM-5 | February 12, 2026 | 744B total (40B active) | 200K input tokens | Complex systems engineering and long-horizon agentic tasks | [Try in playground](https://app.fireworks.ai/playground?model=accounts/fireworks/models/glm-5) |",
    "## GLM-5",
    "### What is GLM-5?",
    "\u2022**Corporate Sponsor:** Zhipu AI ([Z.ai](http://Z.ai))"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "This roundup compares the top open source LLMs available today, breaking down their trade-offs and ideal applications so you can make an informed decision without running your own eval suite.",
    "## The best open source LLMs at a glance",
    "| GLM-5 | February 12, 2026 | 744B total (40B active) | 200K input tokens | Complex systems engineering and long-horizon agentic tasks | [Try in playground](https://app.fireworks.ai/playground?model=accounts/fireworks/models/glm-5) |",
    "## GLM-5",
    "### What is GLM-5?",
    "\u2022**Release Date:** February 12, 2026"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "This roundup compares the top open source LLMs available today, breaking down their trade-offs and ideal applications so you can make an informed decision without running your own eval suite.",
    "## The best open source LLMs at a glance",
    "| GLM-5 | February 12, 2026 | 744B total (40B active) | 200K input tokens | Complex systems engineering and long-horizon agentic tasks | [Try in playground](https://app.fireworks.ai/playground?model=accounts/fireworks/models/glm-5) |"
   ]
  },
  {
   "url": "https://fireworks.ai/blog/best-open-source-llms-may-2026",
   "title": "Best Open Source LLMs of May 2026: We Reviewed 7 Models",
   "published_at": "2026-05-19T00:00:00",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-01-27",
   "access_type": null,
   "license": null,
   "significance": "Front-end visual coding and distributed market research",
   "organization_evidence": [
    "This roundup compares the top open source LLMs available today, breaking down their trade-offs and ideal applications so you can make an informed decision without running your own eval suite.",
    "## The best open source LLMs at a glance",
    "| Kimi K2.5 | January 27, 2026 | 1 Trillion total (32B active) | 256K tokens | Front-end visual coding and distributed market research | [Try in playground](https://app.fireworks.ai/playground?model=accounts/fireworks/models/kimi-k2p5) |",
    "## Kimi K2.5",
    "### What is Kimi K2.5?",
    "\u2022**Corporate Sponsor:** Moonshot AI"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "This roundup compares the top open source LLMs available today, breaking down their trade-offs and ideal applications so you can make an informed decision without running your own eval suite.",
    "## The best open source LLMs at a glance",
    "| Kimi K2.5 | January 27, 2026 | 1 Trillion total (32B active) | 256K tokens | Front-end visual coding and distributed market research | [Try in playground](https://app.fireworks.ai/playground?model=accounts/fireworks/models/kimi-k2p5) |",
    "## Kimi K2.5",
    "### What is Kimi K2.5?",
    "\u2022**Release Date:** January 27, 2026"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "This roundup compares the top open source LLMs available today, breaking down their trade-offs and ideal applications so you can make an informed decision without running your own eval suite.",
    "## The best open source LLMs at a glance",
    "| Kimi K2.5 | January 27, 2026 | 1 Trillion total (32B active) | 256K tokens | Front-end visual coding and distributed market research | [Try in playground](https://app.fireworks.ai/playground?model=accounts/fireworks/models/kimi-k2p5) |"
   ]
  },
  {
   "url": "https://americangovernancetoday.com/article/928322887-poolside-releases-laguna-s-2-1-the-west-s-most-capable-open-weight-model",
   "title": "Poolside releases Laguna S 2.1, the West\u2019s most capable open-weight model",
   "published_at": null,
   "organization": "Poolside",
   "model": null,
   "release_date": "2026-07-21",
   "access_type": null,
   "license": null,
   "significance": "118-billion-parameter foundation model built for agentic coding that matches or exceeds models several times its size",
   "organization_evidence": [
    "# Poolside releases Laguna S 2.1, the West\u2019s most capable open-weight model",
    "SAN FRANCISCO, July 21, 2026 (GLOBE NEWSWIRE) -- [Poolside today released **Laguna S 2.1**](https://www.globenewswire.com/Tracker?data=_qXFO0siMuP0VpmUSf1ESZv-4mfFJrh7XDdEB0XAraXy_b0m6LmOrEMkD_avRl3jWp7VHZdLi2Ob3fK2YnsiAIHDPsNdweqMtxwEaPG1uOUy5zsRlI7EvyUYlzlsZB_4OTGkph95ZxSE9MFyGBj8eg==), a 118-billion-parameter open-weight foundation model built for agentic coding."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "SAN FRANCISCO, July 21, 2026 (GLOBE NEWSWIRE) -- [Poolside today released **Laguna S 2.1**](https://www.globenewswire.com/Tracker?data=_qXFO0siMuP0VpmUSf1ESZv-4mfFJrh7XDdEB0XAraXy_b0m6LmOrEMkD_avRl3jWp7VHZdLi2Ob3fK2YnsiAIHDPsNdweqMtxwEaPG1uOUy5zsRlI7EvyUYlzlsZB_4OTGkph95ZxSE9MFyGBj8eg==), a 118-billion-parameter open-weight foundation model built for agentic coding."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Poolside releases Laguna S 2.1, the West\u2019s most capable open-weight model",
    "The 118-billion-parameter model matches models several times its size on agentic coding and is small enough to run on a single desktop.",
    "SAN FRANCISCO, July 21, 2026 (GLOBE NEWSWIRE) -- [Poolside today released **Laguna S 2.1**](https://www.globenewswire.com/Tracker?data=_qXFO0siMuP0VpmUSf1ESZv-4mfFJrh7XDdEB0XAraXy_b0m6LmOrEMkD_avRl3jWp7VHZdLi2Ob3fK2YnsiAIHDPsNdweqMtxwEaPG1uOUy5zsRlI7EvyUYlzlsZB_4OTGkph95ZxSE9MFyGBj8eg==), a 118-billion-parameter open-weight foundation model built for agentic coding."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/glm-5-1-non-reasoning-vs-trinity-large-thinking",
   "title": "GLM-5.1 (Non-reasoning) vs Trinity Large Thinking: Model Comparison",
   "published_at": null,
   "organization": "Z AI",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "744B parameters, 40B active at inference time, 200k context window",
   "organization_evidence": [
    "# GLM-5.1 (Non-reasoning) vs. Trinity Large Thinking",
    "## Model Comparison",
    "| Creator | ![Z AI](/img/logos/zai_small.svg)[](https://z.ai/) | ![Arcee AI](/img/logos/arcee_small.svg)[](https://www.arcee.ai/) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# GLM-5.1 (Non-reasoning) vs. Trinity Large Thinking",
    "## Model Comparison",
    "| Release Date | April, 2026 | April, 2026 | GLM-5.1 (Non-reasoning) has a more recent release date than Trinity Large Thinking |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# GLM-5.1 (Non-reasoning) vs. Trinity Large Thinking",
    "## Model Comparison",
    "| Context Window | 200k tokens (~300 A4 pages of size 12 Arial font) | 512k tokens (~768 A4 pages of size 12 Arial font) | GLM-5.1 (Non-reasoning) is smaller than Trinity Large Thinking |",
    "| Parameters | 744B, 40B active at inference time | 399B, 13B active at inference time | GLM-5.1 (Non-reasoning) is larger than Trinity Large Thinking |"
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/glm-5-1-non-reasoning-vs-trinity-large-thinking",
   "title": "GLM-5.1 (Non-reasoning) vs Trinity Large Thinking: Model Comparison",
   "published_at": null,
   "organization": "Arcee AI",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "399B parameters, 13B active at inference time, 512k context window",
   "organization_evidence": [
    "# GLM-5.1 (Non-reasoning) vs. Trinity Large Thinking",
    "## Model Comparison",
    "| Creator | ![Z AI](/img/logos/zai_small.svg)[](https://z.ai/) | ![Arcee AI](/img/logos/arcee_small.svg)[](https://www.arcee.ai/) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# GLM-5.1 (Non-reasoning) vs. Trinity Large Thinking",
    "## Model Comparison",
    "| Release Date | April, 2026 | April, 2026 | GLM-5.1 (Non-reasoning) has a more recent release date than Trinity Large Thinking |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# GLM-5.1 (Non-reasoning) vs. Trinity Large Thinking",
    "## Model Comparison",
    "| Context Window | 200k tokens (~300 A4 pages of size 12 Arial font) | 512k tokens (~768 A4 pages of size 12 Arial font) | GLM-5.1 (Non-reasoning) is smaller than Trinity Large Thinking |",
    "| Parameters | 744B, 40B active at inference time | 399B, 13B active at inference time | GLM-5.1 (Non-reasoning) is larger than Trinity Large Thinking |"
   ]
  },
  {
   "url": "https://noze.it/en/insights/inkling-thinking-machines-open-weights",
   "title": "Inkling: Thinking Machines' open-weights model, built for fine-tuning",
   "published_at": "2026-07-15T00:00:00",
   "organization": "Thinking Machines",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "975-billion-parameter Mixture-of-Experts with 1-million-token context window and native multimodal reasoning",
   "organization_evidence": [
    "# Inkling: Thinking Machines' open-weights model, built for fine-tuning",
    "Thinking Machines, Mira Murati's lab, releases Inkling: its first model, a 975-billion-parameter Mixture-of-Experts (41 active) with open weights, a 1-million-token context window and native multimodal reasoning."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "**Thinking Machines Lab**, the lab founded by **Mira Murati** (former CTO of OpenAI), released its first model on 15 July 2026: **Inkling**."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Thinking Machines, Mira Murati's lab, releases Inkling: its first model, a 975-billion-parameter Mixture-of-Experts (41 active) with open weights, a 1-million-token context window and native multimodal reasoning."
   ]
  },
  {
   "url": "https://fazm.ai/t/open-source-ai-projects-tools-updates-may-2026",
   "title": "Fazm - AI Computer Agent for macOS | Voice-Controlled Desktop Automation",
   "published_at": "2026-05-27T03:17:25",
   "organization": "Cohere",
   "model": null,
   "release_date": "2026-05-20",
   "access_type": null,
   "license": null,
   "significance": "218B sparse MoE (25B active per token), 128K context, 48 languages",
   "organization_evidence": [
    "# Open-source AI projects, tools, and updates in May 2026: the agent-layer month",
    "Open-weight LLMs: **Cohere Command A+** (May 20, Apache 2.0, 218B sparse MoE) and **OpenBMB MiniCPM-V 4.6** (May 11, Apache 2.0, 1.3B multimodal).",
    "### Cohere Command A+"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Open-source AI projects, tools, and updates in May 2026: the agent-layer month",
    "Two new permissive releases landed on Hugging Face (Cohere Command A+ on May 20, OpenBMB MiniCPM-V 4.6 on May 11) and the proprietary side took most of the airtime (OpenAI GPT-5.5 Instant on May 5, xAI Grok 4.3 on May 6, Google's Gemini 3.5 Flash at I/O on May 19, Alibaba's closed-weight Qwen3.7-Max on May 20).",
    "Released May 20. 218B sparse MoE (25B active per token), 128K context, 48 languages, Apache 2.0."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Open-source AI projects, tools, and updates in May 2026: the agent-layer month",
    "Released May 20. 218B sparse MoE (25B active per token), 128K context, 48 languages, Apache 2.0."
   ]
  },
  {
   "url": "https://fazm.ai/t/open-source-ai-projects-tools-updates-may-2026",
   "title": "Fazm - AI Computer Agent for macOS | Voice-Controlled Desktop Automation",
   "published_at": "2026-05-27T03:17:25",
   "organization": "OpenBMB",
   "model": null,
   "release_date": "2026-05-11",
   "access_type": null,
   "license": null,
   "significance": "1.3B parameter multimodal vision-language model, 262K context, accepts text, image, multi-image, and video",
   "organization_evidence": [
    "# Open-source AI projects, tools, and updates in May 2026: the agent-layer month",
    "Open-weight LLMs: **Cohere Command A+** (May 20, Apache 2.0, 218B sparse MoE) and **OpenBMB MiniCPM-V 4.6** (May 11, Apache 2.0, 1.3B multimodal).",
    "### OpenBMB MiniCPM-V 4.6"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Open-source AI projects, tools, and updates in May 2026: the agent-layer month",
    "Two new permissive releases landed on Hugging Face (Cohere Command A+ on May 20, OpenBMB MiniCPM-V 4.6 on May 11) and the proprietary side took most of the airtime (OpenAI GPT-5.5 Instant on May 5, xAI Grok 4.3 on May 6, Google's Gemini 3.5 Flash at I/O on May 19, Alibaba's closed-weight Qwen3.7-Max on May 20).",
    "Released May 11. 1.3B parameter multimodal vision-language model (SigLIP2-400M plus Qwen3.5-0.8B), 262K context, accepts text, image, multi-image, and video."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Open-source AI projects, tools, and updates in May 2026: the agent-layer month",
    "Released May 11. 1.3B parameter multimodal vision-language model (SigLIP2-400M plus Qwen3.5-0.8B), 262K context, accepts text, image, multi-image, and video."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/glm-5-1-non-reasoning-vs-claude-opus-4-6",
   "title": "GLM-5.1 (Non-reasoning) vs Claude Opus 4.6 (Non-reasoning, High Effort): Model Comparison",
   "published_at": null,
   "organization": "Z AI",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": null,
   "organization_evidence": [
    "# GLM-5.1 (Non-reasoning) vs. Claude Opus 4.6 (Non-reasoning, High Effort)",
    "| Metric | ![Z AI logo](/img/logos/zai_small.svg)GLM-5.1 (Non-reasoning) | ![Anthropic logo](/img/logos/anthropic_small.svg)Claude Opus 4.6 (Non-reasoning, High Effort) | Analysis |",
    "| Creator | ![Z AI](/img/logos/zai_small.svg)[](https://z.ai/) | ![Anthropic](/img/logos/anthropic_small.svg)[](https://www.anthropic.com/) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# GLM-5.1 (Non-reasoning) vs. Claude Opus 4.6 (Non-reasoning, High Effort)",
    "| Metric | ![Z AI logo](/img/logos/zai_small.svg)GLM-5.1 (Non-reasoning) | ![Anthropic logo](/img/logos/anthropic_small.svg)Claude Opus 4.6 (Non-reasoning, High Effort) | Analysis |",
    "| Release Date | April, 2026 | February, 2026 | GLM-5.1 (Non-reasoning) has a more recent release date than Claude Opus 4.6 (Non-reasoning, High Effort) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": null
  },
  {
   "url": "https://medium.com/@ffguci8/meituan-open-sources-longcat-2-0-a-1-6t-parameter-model-c8e12978003b",
   "title": "Meituan Open-Sources LongCat-2.0, a 1.6T-Parameter Model",
   "published_at": "2026-07-07T19:50:25",
   "organization": "Meituan",
   "model": null,
   "release_date": "2026-06-30",
   "access_type": null,
   "license": null,
   "significance": "1.6-trillion-parameter model with native 1-million-token context window",
   "organization_evidence": [
    "# Meituan Open-Sources LongCat-2.0, a 1.6T-Parameter Model",
    "On June 30, 2026, Meituan became exactly that source: [LongCat-2.0](https://github.com/meituan-longcat/LongCat-2.0), a 1.6-trillion-parameter model, free for commercial use under the MIT license, trained and served entirely on Chinese-made chips."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Meituan Open-Sources LongCat-2.0, a 1.6T-Parameter Model",
    "On June 30, 2026, Meituan became exactly that source: [LongCat-2.0](https://github.com/meituan-longcat/LongCat-2.0), a 1.6-trillion-parameter model, free for commercial use under the MIT license, trained and served entirely on Chinese-made chips."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Meituan Open-Sources LongCat-2.0, a 1.6T-Parameter Model",
    "On June 30, 2026, Meituan became exactly that source: [LongCat-2.0](https://github.com/meituan-longcat/LongCat-2.0), a 1.6-trillion-parameter model, free for commercial use under the MIT license, trained and served entirely on Chinese-made chips.",
    "## Inside the model: scale, sparsity, and a 1M-token context window"
   ]
  },
  {
   "url": "https://boisclub.news/technology/thinking-machines-inkling-open-weights-model-july-2026",
   "title": "Thinking Machines Lab releases Inkling, its first open-weights AI model",
   "published_at": "2026-07-16T15:08:49",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "A Mixture-of-Experts transformer with 975 billion total parameters and 41 billion active on any given token, with a 1 million token context window.",
   "organization_evidence": [
    "Technology",
    "Thinking Machines Lab releases its first model, an open-weights system called Inkling",
    "Thinking Machines Lab released Inkling on July 15, its first model and the first one it has trained from scratch."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Technology",
    "Thinking Machines Lab releases its first model, an open-weights system called Inkling",
    "Janet Torvalds",
    "July 16, 2026",
    "Thinking Machines Lab released Inkling on July 15, its first model and the first one it has trained from scratch."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Technology",
    "Thinking Machines Lab releases its first model, an open-weights system called Inkling",
    "Inkling is a Mixture-of-Experts transformer with 975 billion total parameters and 41 billion active on any given token.",
    "It takes a context window up to 1 million tokens and was pretrained on 45 trillion tokens of text, images, audio, and video."
   ]
  },
  {
   "url": "https://spoonai.me/posts/2026-07-27-moonshot-kimi-k3-open-weights-jul2026-en",
   "title": "2.8 Trillion Parameters Just Went Public \u2014 Kimi K3's Weights Are Free, and Almost Nobody Can Run Them",
   "published_at": "2026-07-26T23:00:00",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-27",
   "access_type": null,
   "license": null,
   "significance": "largest open-weight model ever shipped with 2.8 trillion parameters",
   "organization_evidence": [
    "TOPMoonshot AIKimi K3Open Weights",
    "At 00:00 UTC on July 27, 2026, China's Moonshot AI did exactly what it said it would: it published the complete weights for Kimi K3."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "TOPMoonshot AIKimi K3Open Weights",
    "Moonshot AI released the full Kimi K3 weights on July 27.",
    "At 00:00 UTC on July 27, 2026, China's Moonshot AI did exactly what it said it would: it published the complete weights for Kimi K3."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "TOPMoonshot AIKimi K3Open Weights",
    "Moonshot AI released the full Kimi K3 weights on July 27. It's the largest open-weight model ever shipped, it ranks third in the world on intelligence indexes, and the download is 1.4TB.",
    "Two-point-eight trillion parameters. Larger than any open-weight model released before it, and the first downloadable model competing in the same league as the top closed systems from American frontier labs."
   ]
  },
  {
   "url": "https://theideamagazine.com/science-nature/eight-weeks-to-new-ai-frontiers-china-s-signal-demonstrates-rapid-innovation",
   "title": "Eight Weeks To New AI Frontiers: China\u2019s Signal Demonstrates Rapid Innovation - The Idea Magazine",
   "published_at": "2026-07-15T06:16:37",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor",
   "organization_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "The Chinese open-weight field now comprises four distinct labs\u2014DeepSeek, Z.ai, Moonshot, and Alibaba\u2014each focusing on different strategic strengths, from cost-efficiency to long-horizon stability and self-hosting flexibility.",
    "### The production line \u2014 spring 2026",
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "Between April 24 and June 15, 2026, Chinese labs launched models including DeepSeek V4, MiniMax M3, Kimi K2.7-Code, and GLM-5.2.",
    "### The production line \u2014 spring 2026",
    "APR 24",
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "### The production line \u2014 spring 2026",
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor"
   ]
  },
  {
   "url": "https://theideamagazine.com/science-nature/eight-weeks-to-new-ai-frontiers-china-s-signal-demonstrates-rapid-innovation",
   "title": "Eight Weeks To New AI Frontiers: China\u2019s Signal Demonstrates Rapid Innovation - The Idea Magazine",
   "published_at": "2026-07-15T06:16:37",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "cheap 1M-token context, native multimodal, modified-MIT",
   "organization_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "Between April 24 and June 15, 2026, Chinese labs launched models including DeepSeek V4, MiniMax M3, Kimi K2.7-Code, and GLM-5.2.",
    "### The production line \u2014 spring 2026",
    "JUN 01",
    "**MiniMax M3**cheap 1M-token context, native multimodal, modified-MIT"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "Between April 24 and June 15, 2026, Chinese labs launched models including DeepSeek V4, MiniMax M3, Kimi K2.7-Code, and GLM-5.2.",
    "### The production line \u2014 spring 2026",
    "JUN 01",
    "**MiniMax M3**cheap 1M-token context, native multimodal, modified-MIT"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "### The production line \u2014 spring 2026",
    "**MiniMax M3**cheap 1M-token context, native multimodal, modified-MIT"
   ]
  },
  {
   "url": "https://theideamagazine.com/science-nature/eight-weeks-to-new-ai-frontiers-china-s-signal-demonstrates-rapid-innovation",
   "title": "Eight Weeks To New AI Frontiers: China\u2019s Signal Demonstrates Rapid Innovation - The Idea Magazine",
   "published_at": "2026-07-15T06:16:37",
   "organization": "Moonshot",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "agent-run specialist, ~30% fewer thinking tokens than K2.6",
   "organization_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "The Chinese open-weight field now comprises four distinct labs\u2014DeepSeek, Z.ai, Moonshot, and Alibaba\u2014each focusing on different strategic strengths, from cost-efficiency to long-horizon stability and self-hosting flexibility.",
    "### The production line \u2014 spring 2026",
    "JUN 13",
    "**Kimi K2.7-Code (Moonshot)**agent-run specialist, ~30% fewer thinking tokens than K2.6"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "Between April 24 and June 15, 2026, Chinese labs launched models including DeepSeek V4, MiniMax M3, Kimi K2.7-Code, and GLM-5.2.",
    "### The production line \u2014 spring 2026",
    "JUN 13",
    "**Kimi K2.7-Code (Moonshot)**agent-run specialist, ~30% fewer thinking tokens than K2.6"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "### The production line \u2014 spring 2026",
    "**Kimi K2.7-Code (Moonshot)**agent-run specialist, ~30% fewer thinking tokens than K2.6"
   ]
  },
  {
   "url": "https://theideamagazine.com/science-nature/eight-weeks-to-new-ai-frontiers-china-s-signal-demonstrates-rapid-innovation",
   "title": "Eight Weeks To New AI Frontiers: China\u2019s Signal Demonstrates Rapid Innovation - The Idea Magazine",
   "published_at": "2026-07-15T06:16:37",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "753B MoE, MIT, top open-weight on Artificial Analysis index",
   "organization_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "The Chinese open-weight field now comprises four distinct labs\u2014DeepSeek, Z.ai, Moonshot, and Alibaba\u2014each focusing on different strategic strengths, from cost-efficiency to long-horizon stability and self-hosting flexibility.",
    "### The production line \u2014 spring 2026",
    "JUN 13\u201316",
    "**GLM-5.2 (Z.ai)**753B MoE, MIT, top open-weight on Artificial Analysis index"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "Between April 24 and June 15, 2026, Chinese labs launched models including DeepSeek V4, MiniMax M3, Kimi K2.7-Code, and GLM-5.2.",
    "### The production line \u2014 spring 2026",
    "JUN 13\u201316",
    "**GLM-5.2 (Z.ai)**753B MoE, MIT, top open-weight on Artificial Analysis index"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "### The production line \u2014 spring 2026",
    "**GLM-5.2 (Z.ai)**753B MoE, MIT, top open-weight on Artificial Analysis index"
   ]
  },
  {
   "url": "https://graphify.net/ai-coding/llms/inkling-model",
   "title": "Inkling Model: Specs, Benchmarks and How to Use It",
   "published_at": "2026-07-16T06:39:21",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "975-billion-parameter Mixture-of-Experts system with 41 billion active parameters, native text, image, and audio reasoning, and weights designed for customization",
   "organization_evidence": [
    "Inkling Model specs at a glance",
    "| Release | July 15, 2026 | First model release from Thinking Machines Lab |",
    "The **Inkling model** is Thinking Machines Lab\u2019s first open-weight foundation model: a 975-billion-parameter Mixture-of-Experts system with 41 billion active parameters, native text, image, and audio reasoning, and weights designed for customization."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Inkling Model specs at a glance",
    "| Specification | Published detail | Practical interpretation |",
    "| Release | July 15, 2026 | First model release from Thinking Machines Lab |",
    "It arrived on July 15, 2026, with a deliberately measured pitch."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Inkling Model specs at a glance",
    "The **Inkling model** is Thinking Machines Lab\u2019s first open-weight foundation model: a 975-billion-parameter Mixture-of-Experts system with 41 billion active parameters, native text, image, and audio reasoning, and weights designed for customization."
   ]
  },
  {
   "url": "https://bit.ly/4wMoxLU",
   "title": "Mira Murati\u2019s Thinking Machines drops Inkling, an open-weights model anyone can access",
   "published_at": "2026-07-15T00:00:00",
   "organization": "Thinking Machines",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "a 975 billion parameter mixture-of-experts model trained on 45 trillion tokens",
   "organization_evidence": [
    "Mira Murati\u2019s Thinking Machines drops Inkling, an open-weights model anyone can access",
    "Mira Murati\u2019s [Thinking Machines Lab Inc.](https://thinkingmachines.ai/) today launched its first foundation model with the release of [Inkling](https://thinkingmachines.ai/inkling/), making its full open weights available to developers so they can fine-tune it as they wish."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "UPDATED 19:51 EDT / JULY 15 2026",
    "\u2014 Mira Murati (@miramurati) [July 15, 2026](https://x.com/miramurati/status/2077455974743593100?ref_src=twsrc%5Etfw)"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "In a blog post, Thinking Machines [explained](https://thinkingmachines.ai/news/introducing-inkling/) that Inkling is a mixture-of-experts model that features 975 billion parameters, although for the average prompt it will only draw on a small fraction of that number \u2013 about 41 billion \u2013 in order to process tasks faster and keep costs low.",
    "The company said the model was trained on about 45 trillion tokens of text, image, audio and video and can reason natively across all four inputs."
   ]
  },
  {
   "url": "https://geotoolbox.ai/blog/best-open-source-llms",
   "title": "The 10 Best Open-Source LLMs in 2026 (Ranked)",
   "published_at": "2026-07-22T00:00:00",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "Best overall, agentic coding",
   "organization_evidence": [
    "The 10 Best Open-Source LLMs in 2026 (Ranked)",
    "The best open source LLMs of July 2026, ranked: GLM-5.2, DeepSeek V4, Kimi K3 and more, with licenses, real hardware needs, and every claim dated and verified.",
    "| **GLM-5.2** | Z.ai | 753B / 40B | 1M | MIT | Best overall, agentic coding |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "### 1. GLM-5.2 (Z.ai): Best Overall",
    "**753B total / 40B active MoE, 1M context, MIT license, announced June 13, 2026 with weights on Hugging Face three days later.**"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The 10 Best Open-Source LLMs in 2026 (Ranked)",
    "The short answer: **GLM-5.2** is the best open-source LLM you can download and run today, **DeepSeek V4** is the best value, and **Kimi K3** is the new frontier flagship, its launch benchmarks already ahead of the open field, with downloadable weights scheduled for July 27, 2026.",
    "| **GLM-5.2** | Z.ai | 753B / 40B | 1M | MIT | Best overall, agentic coding |"
   ]
  },
  {
   "url": "https://geotoolbox.ai/blog/best-open-source-llms",
   "title": "The 10 Best Open-Source LLMs in 2026 (Ranked)",
   "published_at": "2026-07-22T00:00:00",
   "organization": "DeepSeek",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Value, world knowledge",
   "organization_evidence": [
    "The 10 Best Open-Source LLMs in 2026 (Ranked)",
    "The best open source LLMs of July 2026, ranked: GLM-5.2, DeepSeek V4, Kimi K3 and more, with licenses, real hardware needs, and every claim dated and verified.",
    "| **DeepSeek V4 Pro / Flash** | DeepSeek | 1.6T / 49B and 284B / 13B | 1M | MIT | Value, world knowledge |"
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The 10 Best Open-Source LLMs in 2026 (Ranked)",
    "The short answer: **GLM-5.2** is the best open-source LLM you can download and run today, **DeepSeek V4** is the best value, and **Kimi K3** is the new frontier flagship, its launch benchmarks already ahead of the open field, with downloadable weights scheduled for July 27, 2026.",
    "| **DeepSeek V4 Pro / Flash** | DeepSeek | 1.6T / 49B and 284B / 13B | 1M | MIT | Value, world knowledge |"
   ]
  },
  {
   "url": "https://geotoolbox.ai/blog/best-open-source-llms",
   "title": "The 10 Best Open-Source LLMs in 2026 (Ranked)",
   "published_at": "2026-07-22T00:00:00",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-16",
   "access_type": null,
   "license": null,
   "significance": "Agent swarms, frontier scale",
   "organization_evidence": [
    "The 10 Best Open-Source LLMs in 2026 (Ranked)",
    "| **Kimi K2.6 / K3** | Moonshot AI | 1T / 32B and 2.8T (K3) | 256K / 1M | Modified MIT (K2.6); K3 TBD Jul 27 | Agent swarms, frontier scale |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "### 3. Kimi K2.6 and K3 (Moonshot AI): Best for Agents, New Frontier Flagship",
    "**K2.6: ~1T / 32B, 256K context. K3: 2.8T parameters, 1M context, launched July 16, 2026.**"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The 10 Best Open-Source LLMs in 2026 (Ranked)",
    "| **Kimi K2.6 / K3** | Moonshot AI | 1T / 32B and 2.8T (K3) | 256K / 1M | Modified MIT (K2.6); K3 TBD Jul 27 | Agent swarms, frontier scale |"
   ]
  },
  {
   "url": "https://geotoolbox.ai/blog/best-open-source-llms",
   "title": "The 10 Best Open-Source LLMs in 2026 (Ranked)",
   "published_at": "2026-07-22T00:00:00",
   "organization": "MiniMax",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Multimodal agents",
   "organization_evidence": [
    "The 10 Best Open-Source LLMs in 2026 (Ranked)",
    "| **MiniMax M3** | MiniMax | 428B / 23B | 1M | MiniMax Community | Multimodal agents |"
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The 10 Best Open-Source LLMs in 2026 (Ranked)",
    "| **MiniMax M3** | MiniMax | 428B / 23B | 1M | MiniMax Community | Multimodal agents |"
   ]
  },
  {
   "url": "https://geotoolbox.ai/blog/best-open-source-llms",
   "title": "The 10 Best Open-Source LLMs in 2026 (Ranked)",
   "published_at": "2026-07-22T00:00:00",
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "License freedom, multilingual",
   "organization_evidence": [
    "The 10 Best Open-Source LLMs in 2026 (Ranked)",
    "| **Qwen3.5 / 3.6 family** | Alibaba | 0.8B to 397B / 17B | 262K+ | Apache 2.0 | License freedom, multilingual |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "### 5. Qwen3.5 and Qwen3.6 (Alibaba): Best License Freedom and Language Coverage",
    "The generations also turn over fast: the April 2026 **Qwen3.6** wave added a 27B dense model and a 35B-total, 3B-active MoE that punch far above their size, and they, not the Qwen3.5 flagship, are the family's current picks for consumer hardware."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The 10 Best Open-Source LLMs in 2026 (Ranked)",
    "| **Qwen3.5 / 3.6 family** | Alibaba | 0.8B to 397B / 17B | 262K+ | Apache 2.0 | License freedom, multilingual |"
   ]
  },
  {
   "url": "https://geotoolbox.ai/blog/best-open-source-llms",
   "title": "The 10 Best Open-Source LLMs in 2026 (Ranked)",
   "published_at": "2026-07-22T00:00:00",
   "organization": "Xiaomi",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Token efficiency",
   "organization_evidence": [
    "The 10 Best Open-Source LLMs in 2026 (Ranked)",
    "| **MiMo-V2.5-Pro** | Xiaomi | 1.02T / 42B | 1M | MIT | Token efficiency |"
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The 10 Best Open-Source LLMs in 2026 (Ranked)",
    "| **MiMo-V2.5-Pro** | Xiaomi | 1.02T / 42B | 1M | MIT | Token efficiency |"
   ]
  },
  {
   "url": "https://geotoolbox.ai/blog/best-open-source-llms",
   "title": "The 10 Best Open-Source LLMs in 2026 (Ranked)",
   "published_at": "2026-07-22T00:00:00",
   "organization": "Meta",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Western default, long context",
   "organization_evidence": [
    "The 10 Best Open-Source LLMs in 2026 (Ranked)",
    "| **Llama 4 Scout / Maverick** | Meta | 109B / 17B and 400B / 17B | up to 10M claimed | Llama 4 Community | Western default, long context |"
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The 10 Best Open-Source LLMs in 2026 (Ranked)",
    "| **Llama 4 Scout / Maverick** | Meta | 109B / 17B and 400B / 17B | up to 10M claimed | Llama 4 Community | Western default, long context |"
   ]
  },
  {
   "url": "https://geotoolbox.ai/blog/best-open-source-llms",
   "title": "The 10 Best Open-Source LLMs in 2026 (Ranked)",
   "published_at": "2026-07-22T00:00:00",
   "organization": "Google",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "Local and edge hardware",
   "organization_evidence": [
    "The 10 Best Open-Source LLMs in 2026 (Ranked)",
    "| **Gemma 4** | Google | E2B to 31B (4 sizes) | 256K (31B) | Apache 2.0 | Local and edge hardware |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "### 8. Gemma 4 (Google): Best for Local and Edge Hardware",
    "| Google | **Gemma 4** (April 2026) | Gemma 3 |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The 10 Best Open-Source LLMs in 2026 (Ranked)",
    "| **Gemma 4** | Google | E2B to 31B (4 sizes) | 256K (31B) | Apache 2.0 | Local and edge hardware |"
   ]
  },
  {
   "url": "https://geotoolbox.ai/blog/best-open-source-llms",
   "title": "The 10 Best Open-Source LLMs in 2026 (Ranked)",
   "published_at": "2026-07-22T00:00:00",
   "organization": "OpenAI",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Permissive Western reasoning",
   "organization_evidence": [
    "The 10 Best Open-Source LLMs in 2026 (Ranked)",
    "| **gpt-oss-120b** | OpenAI | 117B / 5.1B | 128K | Apache 2.0 | Permissive Western reasoning |"
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The 10 Best Open-Source LLMs in 2026 (Ranked)",
    "| **gpt-oss-120b** | OpenAI | 117B / 5.1B | 128K | Apache 2.0 | Permissive Western reasoning |"
   ]
  },
  {
   "url": "https://geotoolbox.ai/blog/best-open-source-llms",
   "title": "The 10 Best Open-Source LLMs in 2026 (Ranked)",
   "published_at": "2026-07-22T00:00:00",
   "organization": "NVIDIA",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Closest to truly open",
   "organization_evidence": [
    "The 10 Best Open-Source LLMs in 2026 (Ranked)",
    "| **Nemotron 3** | NVIDIA | 550B / 55B (Ultra) | 1M | OpenMDW-1.1 | Closest to truly open |"
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The 10 Best Open-Source LLMs in 2026 (Ranked)",
    "| **Nemotron 3** | NVIDIA | 550B / 55B (Ultra) | 1M | OpenMDW-1.1 | Closest to truly open |"
   ]
  },
  {
   "url": "https://letsdatascience.com/news/zai-releases-glm-52-tops-open-weight-rankings-7c27fe58",
   "title": "Chinese Models Narrow Gap With Anthropic and OpenAI",
   "published_at": "2026-06-27T19:49:23",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-06-16",
   "access_type": null,
   "license": null,
   "significance": "top-ranked open-weights model overall and sits fourth on the full leaderboard behind Claude Fable 5, Opus 4.8, and GPT-5.5",
   "organization_evidence": [
    "Chinese Models Narrow Gap With Anthropic and OpenAI",
    "Z.ai (formerly Zhipu AI) released GLM-5.2 as an MIT-licensed, open-weights model on June 16, 2026, following an initial rollout to paying users three days earlier."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Chinese Models Narrow Gap With Anthropic and OpenAI",
    "Z.ai (formerly Zhipu AI) released GLM-5.2 as an MIT-licensed, open-weights model on June 16, 2026, following an initial rollout to paying users three days earlier."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Chinese Models Narrow Gap With Anthropic and OpenAI",
    "According to Artificial Analysis, whose Intelligence Index is the primary independent benchmark tracker for this claim, GLM-5.2 is now the top-ranked open-weights model overall and sits fourth on the full leaderboard behind Claude Fable 5, Opus 4.8, and GPT-5.5; it also placed second on Code Arena's front-end coding leaderboard."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/qwen3-8-27b-vs-gpt-5-6-sol-low",
   "title": "Qwen3.8 27B vs GPT-5.6 Sol (low): Model Comparison | Artificial Analysis",
   "published_at": null,
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-08",
   "access_type": null,
   "license": null,
   "significance": "More intelligent than GPT-5.6 Sol (low) with 27B parameters, scoring 52 on the Artificial Analysis Intelligence Index.",
   "organization_evidence": [
    "Comparison between Qwen3.8 27B and GPT-5.6 Sol (low) across intelligence, price, speed, context window and more.",
    "| | ![Alibaba logo](/img/logos/alibaba_small.svg)Alibaba Qwen3.8 27B | ![OpenAI logo](/img/logos/openai_small.svg)OpenAI GPT-5.6 Sol (low) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Comparison between Qwen3.8 27B and GPT-5.6 Sol (low) across intelligence, price, speed, context window and more.",
    "| Release Date | August 2026 | July 2026 | Qwen3.8 27B has a more recent release date than GPT-5.6 Sol (low) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Comparison between Qwen3.8 27B and GPT-5.6 Sol (low) across intelligence, price, speed, context window and more.",
    "| Intelligence Index | 52 | 51 | Qwen3.8 27B is more intelligent than GPT-5.6 Sol (low) |"
   ]
  },
  {
   "url": "https://geeksalad.org/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - Geek Salad",
   "published_at": "2026-07-15T10:56:40",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor",
   "organization_evidence": [
    "The production line \u2014 spring 2026",
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Between late April and mid-June 2026, Chinese laboratories launched four major open-weight AI models, each representing a distinct strategic focus.",
    "DeepSeek V4, released on April 24, features 1.6 trillion total parameters but activates only 49 billion per pass, with a 1 million token context, and is priced at the low end of the market."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The production line \u2014 spring 2026",
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor"
   ]
  },
  {
   "url": "https://geeksalad.org/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - Geek Salad",
   "published_at": "2026-07-15T10:56:40",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "cheap 1M-token context, native multimodal, modified-MIT",
   "organization_evidence": [
    "The production line \u2014 spring 2026"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Between late April and mid-June 2026, Chinese laboratories launched four major open-weight AI models, each representing a distinct strategic focus.",
    "Following that, the MiniMax M3 was launched on June 1, and within days, Kimi K2.7-Code and GLM-5.2 appeared in mid-June, all available for download and most under MIT-class licenses."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The production line \u2014 spring 2026"
   ]
  },
  {
   "url": "https://geeksalad.org/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - Geek Salad",
   "published_at": "2026-07-15T10:56:40",
   "organization": "Moonshot",
   "model": null,
   "release_date": "2026-06",
   "access_type": null,
   "license": null,
   "significance": "agent-run specialist, ~30% fewer thinking tokens than K2.6",
   "organization_evidence": [
    "The production line \u2014 spring 2026"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Between late April and mid-June 2026, Chinese laboratories launched four major open-weight AI models, each representing a distinct strategic focus.",
    "Following that, the MiniMax M3 was launched on June 1, and within days, Kimi K2.7-Code and GLM-5.2 appeared in mid-June, all available for download and most under MIT-class licenses."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The production line \u2014 spring 2026"
   ]
  },
  {
   "url": "https://geeksalad.org/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - Geek Salad",
   "published_at": "2026-07-15T10:56:40",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-06",
   "access_type": null,
   "license": null,
   "significance": "753B MoE, MIT, top open-weight on Artificial Analysis index",
   "organization_evidence": [
    "The production line \u2014 spring 2026"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Between late April and mid-June 2026, Chinese laboratories launched four major open-weight AI models, each representing a distinct strategic focus.",
    "Following that, the MiniMax M3 was launched on June 1, and within days, Kimi K2.7-Code and GLM-5.2 appeared in mid-June, all available for download and most under MIT-class licenses."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The production line \u2014 spring 2026"
   ]
  },
  {
   "url": "https://medium.com/@mshojaei77/evaluating-inkling-the-open-weight-moe-with-continuous-reasoning-effort-3883cebbfa92",
   "title": "Evaluating Inkling: The Open-Weight MoE with Continuous Reasoning Effort",
   "published_at": "2026-07-16T12:28:49",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "975B-parameter sparse Mixture-of-Experts with 1-million-token context window and native multimodal reasoning",
   "organization_evidence": [
    "Evaluating Inkling: The Open-Weight MoE with Continuous Reasoning Effort",
    "Thinking Machines Lab released Inkling on July 15, 2026."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Evaluating Inkling: The Open-Weight MoE with Continuous Reasoning Effort",
    "Thinking Machines Lab released Inkling on July 15, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Evaluating Inkling: The Open-Weight MoE with Continuous Reasoning Effort",
    "It is a **975B-parameter sparse Mixture-of-Experts (MoE) with 41B active parameters**, featuring a **1-million-token architectural context window**, native text/image/audio processing, and Apache 2.0-licensed weights in both BF16 and NVIDIA NVFP4 formats."
   ]
  },
  {
   "url": "https://llmcheck.net/blog/state-of-open-source-local-llms-june-2026",
   "title": "State of Open-Source Local LLMs \u2014 June 2026: Qwen 4 Goes Live, Llama 5 Scales, Grok Goes Open",
   "published_at": "2026-06-06T00:00:00",
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "NEW #1 open model, score 75, +2pp across every benchmark vs. the Preview.",
   "organization_evidence": [
    "## The 30-Day Recap (TL;DR)",
    "- **Qwen 4 (full release)** Jun 1 \u2014 Apache 2.0, NEW #1 open model, score 75, +2pp across every benchmark vs. the Preview."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## The 30-Day Recap (TL;DR)",
    "- **Qwen 4 (full release)** Jun 1 \u2014 Apache 2.0, NEW #1 open model, score 75, +2pp across every benchmark vs. the Preview."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## The 30-Day Recap (TL;DR)",
    "- **Qwen 4 (full release)** Jun 1 \u2014 Apache 2.0, NEW #1 open model, score 75, +2pp across every benchmark vs. the Preview."
   ]
  },
  {
   "url": "https://llmcheck.net/blog/state-of-open-source-local-llms-june-2026",
   "title": "State of Open-Source Local LLMs \u2014 June 2026: Qwen 4 Goes Live, Llama 5 Scales, Grok Goes Open",
   "published_at": "2026-06-06T00:00:00",
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-06-02",
   "access_type": null,
   "license": null,
   "significance": "82% SWE-Verified, the best open-source Mac coder \u2014 beats Devstral.",
   "organization_evidence": [
    "## The 30-Day Recap (TL;DR)",
    "- **Qwen 4 Coder 32B-A3B** Jun 2 \u2014 Apache 2.0, 82% SWE-Verified, the best open-source Mac coder \u2014 beats Devstral."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## The 30-Day Recap (TL;DR)",
    "- **Qwen 4 Coder 32B-A3B** Jun 2 \u2014 Apache 2.0, 82% SWE-Verified, the best open-source Mac coder \u2014 beats Devstral."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## The 30-Day Recap (TL;DR)",
    "- **Qwen 4 Coder 32B-A3B** Jun 2 \u2014 Apache 2.0, 82% SWE-Verified, the best open-source Mac coder \u2014 beats Devstral."
   ]
  },
  {
   "url": "https://llmcheck.net/blog/state-of-open-source-local-llms-june-2026",
   "title": "State of Open-Source Local LLMs \u2014 June 2026: Qwen 4 Goes Live, Llama 5 Scales, Grok Goes Open",
   "published_at": "2026-06-06T00:00:00",
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-06-03",
   "access_type": null,
   "license": null,
   "significance": "beats Phi-5 Mini in the 8 GB tier at 135 tok/s on M5 Max.",
   "organization_evidence": [
    "## The 30-Day Recap (TL;DR)",
    "- **Qwen 4 4B** Jun 3 \u2014 Apache 2.0, beats Phi-5 Mini in the 8 GB tier at 135 tok/s on M5 Max."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## The 30-Day Recap (TL;DR)",
    "- **Qwen 4 4B** Jun 3 \u2014 Apache 2.0, beats Phi-5 Mini in the 8 GB tier at 135 tok/s on M5 Max."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## The 30-Day Recap (TL;DR)",
    "- **Qwen 4 4B** Jun 3 \u2014 Apache 2.0, beats Phi-5 Mini in the 8 GB tier at 135 tok/s on M5 Max."
   ]
  },
  {
   "url": "https://llmcheck.net/blog/state-of-open-source-local-llms-june-2026",
   "title": "State of Open-Source Local LLMs \u2014 June 2026: Qwen 4 Goes Live, Llama 5 Scales, Grok Goes Open",
   "published_at": "2026-06-06T00:00:00",
   "organization": "Meta",
   "model": null,
   "release_date": "2026-06-04",
   "access_type": null,
   "license": null,
   "significance": "Meta's bigger dense, MMLU 88%, fills the Scout/8B gap.",
   "organization_evidence": [
    "## The 30-Day Recap (TL;DR)",
    "- **Llama 5 70B** Jun 4 \u2014 Meta's bigger dense, MMLU 88%, fills the Scout/8B gap."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## The 30-Day Recap (TL;DR)",
    "- **Llama 5 70B** Jun 4 \u2014 Meta's bigger dense, MMLU 88%, fills the Scout/8B gap."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## The 30-Day Recap (TL;DR)",
    "- **Llama 5 70B** Jun 4 \u2014 Meta's bigger dense, MMLU 88%, fills the Scout/8B gap."
   ]
  },
  {
   "url": "https://llmcheck.net/blog/state-of-open-source-local-llms-june-2026",
   "title": "State of Open-Source Local LLMs \u2014 June 2026: Qwen 4 Goes Live, Llama 5 Scales, Grok Goes Open",
   "published_at": "2026-06-06T00:00:00",
   "organization": "Mistral",
   "model": null,
   "release_date": "2026-06-04",
   "access_type": null,
   "license": null,
   "significance": "Apache 2.0 70B dense, agentic tool use is strong, the new license-friendly 70B.",
   "organization_evidence": [
    "## The 30-Day Recap (TL;DR)",
    "- **Mistral Voyage Pro 70B** Jun 4 \u2014 Apache 2.0 70B dense, agentic tool use is strong, the new license-friendly 70B."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## The 30-Day Recap (TL;DR)",
    "- **Mistral Voyage Pro 70B** Jun 4 \u2014 Apache 2.0 70B dense, agentic tool use is strong, the new license-friendly 70B."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## The 30-Day Recap (TL;DR)",
    "- **Mistral Voyage Pro 70B** Jun 4 \u2014 Apache 2.0 70B dense, agentic tool use is strong, the new license-friendly 70B."
   ]
  },
  {
   "url": "https://llmcheck.net/blog/state-of-open-source-local-llms-june-2026",
   "title": "State of Open-Source Local LLMs \u2014 June 2026: Qwen 4 Goes Live, Llama 5 Scales, Grok Goes Open",
   "published_at": "2026-06-06T00:00:00",
   "organization": "Google",
   "model": null,
   "release_date": "2026-06-02",
   "access_type": null,
   "license": null,
   "significance": "Google refresh, 1M context jump (from 256K), improved multimodal.",
   "organization_evidence": [
    "## The 30-Day Recap (TL;DR)",
    "- **Gemma 4.5 12B** Jun 2 \u2014 Google refresh, 1M context jump (from 256K), improved multimodal."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## The 30-Day Recap (TL;DR)",
    "- **Gemma 4.5 12B** Jun 2 \u2014 Google refresh, 1M context jump (from 256K), improved multimodal."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## The 30-Day Recap (TL;DR)",
    "- **Gemma 4.5 12B** Jun 2 \u2014 Google refresh, 1M context jump (from 256K), improved multimodal."
   ]
  },
  {
   "url": "https://llmcheck.net/blog/state-of-open-source-local-llms-june-2026",
   "title": "State of Open-Source Local LLMs \u2014 June 2026: Qwen 4 Goes Live, Llama 5 Scales, Grok Goes Open",
   "published_at": "2026-06-06T00:00:00",
   "organization": "Microsoft",
   "model": null,
   "release_date": "2026-05-30",
   "access_type": null,
   "license": null,
   "significance": "MIT, MMLU 86%, AIME 75%, tops the 14B tier outright.",
   "organization_evidence": [
    "## The 30-Day Recap (TL;DR)",
    "- **Phi-5 Medium 14B** May 30 \u2014 MIT, MMLU 86%, AIME 75%, tops the 14B tier outright."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## The 30-Day Recap (TL;DR)",
    "- **Phi-5 Medium 14B** May 30 \u2014 MIT, MMLU 86%, AIME 75%, tops the 14B tier outright."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## The 30-Day Recap (TL;DR)",
    "- **Phi-5 Medium 14B** May 30 \u2014 MIT, MMLU 86%, AIME 75%, tops the 14B tier outright."
   ]
  },
  {
   "url": "https://llmcheck.net/blog/state-of-open-source-local-llms-june-2026",
   "title": "State of Open-Source Local LLMs \u2014 June 2026: Qwen 4 Goes Live, Llama 5 Scales, Grok Goes Open",
   "published_at": "2026-06-06T00:00:00",
   "organization": "xAI",
   "model": null,
   "release_date": "2026-06-05",
   "access_type": null,
   "license": null,
   "significance": "xAI's FIRST open weights ever, custom license, ~32 tok/s on M5 Max 64 GB.",
   "organization_evidence": [
    "## The 30-Day Recap (TL;DR)",
    "- **Grok 4 Open 100B-A20B** Jun 5 \u2014 xAI's FIRST open weights ever, custom license, ~32 tok/s on M5 Max 64 GB."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## The 30-Day Recap (TL;DR)",
    "- **Grok 4 Open 100B-A20B** Jun 5 \u2014 xAI's FIRST open weights ever, custom license, ~32 tok/s on M5 Max 64 GB."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## The 30-Day Recap (TL;DR)",
    "- **Grok 4 Open 100B-A20B** Jun 5 \u2014 xAI's FIRST open weights ever, custom license, ~32 tok/s on M5 Max 64 GB."
   ]
  },
  {
   "url": "https://theagenttimes.com/agents/article/poolside-releases-open-weight-coding-model-that-fits-single--e5b31f82",
   "title": "Poolside Releases Open-Weight Coding Model That Fits Single Workstati\u2026",
   "published_at": null,
   "organization": "Poolside",
   "model": null,
   "release_date": "2026-07-22",
   "access_type": null,
   "license": null,
   "significance": "A 118B-parameter open-weight Mixture-of-Experts coding model that scores 70.2% on Terminal-Bench 2.1 and fits on a single NVIDIA DGX Spark.",
   "organization_evidence": [
    "Poolside Releases Open-Weight Coding Model That Fits Single Workstation",
    "Poolside, a San Francisco-based AI startup focused on agentic coding, has released Laguna S 2.1, a 118-billion-parameter open-weight Mixture-of-Experts model that scores over 70% on Terminal-Bench 2.1 and fits on a single NVIDIA DGX Spark \u2014 a release that investor Nathan Benaich of Air Street Capital amplified as evidence that American open-weight models are closing the gap with frontier systems."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Poolside Releases Open-Weight Coding Model That Fits Single Workstation",
    "Andrea Badia ([@Badiaserra on X](https://x.com/Badiaserra)), whose post Benaich reposted, wrote that \u201cit\u2019s not only the Chinese\u201d and that \u201cAmerican open-weight contenders are also catching up to the frontier,\u201d citing Poolside\u2019s new model as the proof point. Benaich, whose firm publishes the annual *State of AI* report, lent the claim his signal by reposting it \u2014 apparently without additional commentary, based on available source text \u2014 to his [X audience](https://x.com/nathanbenaich/status/2079917948609634470) on July 22, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Poolside Releases Open-Weight Coding Model That Fits Single Workstation",
    "Poolside, a San Francisco-based AI startup focused on agentic coding, has released Laguna S 2.1, a 118-billion-parameter open-weight Mixture-of-Experts model that scores over 70% on Terminal-Bench 2.1 and fits on a single NVIDIA DGX Spark \u2014 a release that investor Nathan Benaich of Air Street Capital amplified as evidence that American open-weight models are closing the gap with frontier systems."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/gemma-4-26b-a4b-vs-glm-5",
   "title": "Gemma 4 26B A4B (Reasoning) vs GLM-5 (Reasoning): Model Comparison",
   "published_at": null,
   "organization": "Google",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "25.2B, 3.8B active at inference time",
   "organization_evidence": [
    "# Gemma 4 26B A4B (Reasoning) vs. GLM-5 (Reasoning)",
    "| Creator | ![Google](/img/logos/google_small.svg)[](https://deepmind.google/technologies/gemini/) | ![Z AI](/img/logos/zai_small.svg)[](https://z.ai/) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Gemma 4 26B A4B (Reasoning) vs. GLM-5 (Reasoning)",
    "| Metric | ![Google logo](/img/logos/google_small.svg)Gemma 4 26B A4B (Reasoning) | ![Z AI logo](/img/logos/zai_small.svg)GLM-5 (Reasoning) | Analysis |",
    "| Release Date | April, 2026 | February, 2026 | Gemma 4 26B A4B (Reasoning) has a more recent release date than GLM-5 (Reasoning) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Gemma 4 26B A4B (Reasoning) vs. GLM-5 (Reasoning)",
    "| Metric | ![Google logo](/img/logos/google_small.svg)Gemma 4 26B A4B (Reasoning) | ![Z AI logo](/img/logos/zai_small.svg)GLM-5 (Reasoning) | Analysis |",
    "| Parameters | 25.2B, 3.8B active at inference time | 744B, 40B active at inference time | Gemma 4 26B A4B (Reasoning) is smaller than GLM-5 (Reasoning) |"
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/gemma-4-26b-a4b-vs-glm-5",
   "title": "Gemma 4 26B A4B (Reasoning) vs GLM-5 (Reasoning): Model Comparison",
   "published_at": null,
   "organization": "Z AI",
   "model": null,
   "release_date": "2026-02",
   "access_type": null,
   "license": null,
   "significance": "744B, 40B active at inference time",
   "organization_evidence": [
    "# Gemma 4 26B A4B (Reasoning) vs. GLM-5 (Reasoning)",
    "| Creator | ![Google](/img/logos/google_small.svg)[](https://deepmind.google/technologies/gemini/) | ![Z AI](/img/logos/zai_small.svg)[](https://z.ai/) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Gemma 4 26B A4B (Reasoning) vs. GLM-5 (Reasoning)",
    "| Metric | ![Google logo](/img/logos/google_small.svg)Gemma 4 26B A4B (Reasoning) | ![Z AI logo](/img/logos/zai_small.svg)GLM-5 (Reasoning) | Analysis |",
    "| Release Date | April, 2026 | February, 2026 | Gemma 4 26B A4B (Reasoning) has a more recent release date than GLM-5 (Reasoning) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Gemma 4 26B A4B (Reasoning) vs. GLM-5 (Reasoning)",
    "| Metric | ![Google logo](/img/logos/google_small.svg)Gemma 4 26B A4B (Reasoning) | ![Z AI logo](/img/logos/zai_small.svg)GLM-5 (Reasoning) | Analysis |",
    "| Parameters | 25.2B, 3.8B active at inference time | 744B, 40B active at inference time | Gemma 4 26B A4B (Reasoning) is smaller than GLM-5 (Reasoning) |"
   ]
  },
  {
   "url": "https://elfysworld.com/general/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - ELFY'S WORLD",
   "published_at": "2026-07-15T12:26:22",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor",
   "organization_evidence": [
    "Between April 24 and June 15, 2026, Chinese research institutions launched four major open-weight models: DeepSeek V4, MiniMax M3, Kimi K2.7-Code, and GLM-5.2.",
    "The recent releases demonstrate a clear shift from the previous state, where the Chinese open field was limited to one lab, to a diversified ecosystem with four distinct model families: DeepSeek, Z.ai, Moonshot, and Alibaba.",
    "### The production line \u2014 spring 2026",
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Between April 24 and June 15, 2026, Chinese research institutions launched four major open-weight models: DeepSeek V4, MiniMax M3, Kimi K2.7-Code, and GLM-5.2.",
    "### The production line \u2014 spring 2026",
    "APR 24",
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "### The production line \u2014 spring 2026",
    "APR 24",
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor"
   ]
  },
  {
   "url": "https://elfysworld.com/general/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - ELFY'S WORLD",
   "published_at": "2026-07-15T12:26:22",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "cheap 1M-token context, native multimodal, modified-MIT",
   "organization_evidence": [
    "Between April 24 and June 15, 2026, Chinese research institutions launched four major open-weight models: DeepSeek V4, MiniMax M3, Kimi K2.7-Code, and GLM-5.2.",
    "### The production line \u2014 spring 2026",
    "JUN 01",
    "**MiniMax M3**cheap 1M-token context, native multimodal, modified-MIT"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Between April 24 and June 15, 2026, Chinese research institutions launched four major open-weight models: DeepSeek V4, MiniMax M3, Kimi K2.7-Code, and GLM-5.2.",
    "### The production line \u2014 spring 2026",
    "JUN 01",
    "**MiniMax M3**cheap 1M-token context, native multimodal, modified-MIT"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "### The production line \u2014 spring 2026",
    "JUN 01",
    "**MiniMax M3**cheap 1M-token context, native multimodal, modified-MIT"
   ]
  },
  {
   "url": "https://elfysworld.com/general/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - ELFY'S WORLD",
   "published_at": "2026-07-15T12:26:22",
   "organization": "Moonshot",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "agent-run specialist, ~30% fewer thinking tokens than K2.6",
   "organization_evidence": [
    "Between April 24 and June 15, 2026, Chinese research institutions launched four major open-weight models: DeepSeek V4, MiniMax M3, Kimi K2.7-Code, and GLM-5.2.",
    "The recent releases demonstrate a clear shift from the previous state, where the Chinese open field was limited to one lab, to a diversified ecosystem with four distinct model families: DeepSeek, Z.ai, Moonshot, and Alibaba.",
    "### The production line \u2014 spring 2026",
    "**Kimi K2.7-Code (Moonshot)**agent-run specialist, ~30% fewer thinking tokens than K2.6"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Between April 24 and June 15, 2026, Chinese research institutions launched four major open-weight models: DeepSeek V4, MiniMax M3, Kimi K2.7-Code, and GLM-5.2.",
    "### The production line \u2014 spring 2026",
    "JUN 13",
    "**Kimi K2.7-Code (Moonshot)**agent-run specialist, ~30% fewer thinking tokens than K2.6"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "### The production line \u2014 spring 2026",
    "JUN 13",
    "**Kimi K2.7-Code (Moonshot)**agent-run specialist, ~30% fewer thinking tokens than K2.6"
   ]
  },
  {
   "url": "https://elfysworld.com/general/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - ELFY'S WORLD",
   "published_at": "2026-07-15T12:26:22",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "753B MoE, MIT, top open-weight on Artificial Analysis index",
   "organization_evidence": [
    "Between April 24 and June 15, 2026, Chinese research institutions launched four major open-weight models: DeepSeek V4, MiniMax M3, Kimi K2.7-Code, and GLM-5.2.",
    "The recent releases demonstrate a clear shift from the previous state, where the Chinese open field was limited to one lab, to a diversified ecosystem with four distinct model families: DeepSeek, Z.ai, Moonshot, and Alibaba.",
    "### The production line \u2014 spring 2026",
    "**GLM-5.2 (Z.ai)**753B MoE, MIT, top open-weight on Artificial Analysis index"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Between April 24 and June 15, 2026, Chinese research institutions launched four major open-weight models: DeepSeek V4, MiniMax M3, Kimi K2.7-Code, and GLM-5.2.",
    "### The production line \u2014 spring 2026",
    "JUN 13\u201316",
    "**GLM-5.2 (Z.ai)**753B MoE, MIT, top open-weight on Artificial Analysis index"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "### The production line \u2014 spring 2026",
    "JUN 13\u201316",
    "**GLM-5.2 (Z.ai)**753B MoE, MIT, top open-weight on Artificial Analysis index"
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/minimax-m2-5",
   "title": "MiniMax-M2.5 - Intelligence, Performance & Price Analysis",
   "published_at": null,
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-02-12",
   "access_type": null,
   "license": null,
   "significance": "MiniMax-M2.5 is a Mixture of Experts (MoE) model with 230 billion total parameters, but only 10 billion active parameters are used during inference.",
   "organization_evidence": [
    "MiniMax-M2.5 was created by MiniMax.",
    "MiniMax has launched a newer model, [MiniMax-M2.7](/models/minimax-m2-7)."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "MiniMax-M2.5 was released on February 12, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "MiniMax-M2.5 is a Mixture of Experts (MoE) model with 230 billion total parameters, but only 10 billion active parameters are used during inference."
   ]
  },
  {
   "url": "https://kunalganglani.com/blog/glm-5-2-open-frontier-model-china",
   "title": "GLM 5.2: 744B Open Model After Anthropic Ban [2026]",
   "published_at": "2026-06-14T12:53:37",
   "organization": "ZhipuAI",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "744B-parameter Mixture-of-Experts system with 40B active parameters, a usable 1M-token context window, and MIT-licensed weights",
   "organization_evidence": [
    "AI and Machine Learning",
    "GLM 5.2: China's Open Frontier Model Dropped the Day Anthropic Got Banned [2026]",
    "Hours later, China's ZhipuAI open-sourced GLM 5.2 under MIT license \u2014 with a 1M context window and frontier-grade coding scores."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "AI and Machine Learning",
    "GLM 5.2: China's Open Frontier Model Dropped the Day Anthropic Got Banned [2026]",
    "On June 13, 2026, the US government cracked down on Anthropic's Claude Fable 5."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "AI and Machine Learning",
    "GLM 5.2: China's Open Frontier Model Dropped the Day Anthropic Got Banned [2026]",
    "GLM 5.2 is ZhipuAI's fully open-source [frontier model](/blog/claude-fable-5-benchmark-developer) \u2014 a 744B-parameter Mixture-of-Experts system with 40B active parameters, a usable 1M-token context window, and MIT-licensed weights."
   ]
  },
  {
   "url": "https://blog.imseankim.com/thinking-machines-inkling-open-weight-model-41b-active-customization-tinker-2026",
   "title": "Thinking Machines Inkling Is Here: Why Mira Murati's First Open-Weight Model Bets 41B Active Parameters on Customization, Not Raw Power - Sean Kim \u2014 Arts and Tech",
   "published_at": "2026-07-19T22:19:07",
   "organization": "Thinking Machines",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "Mixture-of-Experts transformer with 975 billion total parameters (41 billion active), 1 million token context window, and controllable thinking effort",
   "organization_evidence": [
    "# Thinking Machines Inkling Is Here: Why Mira Murati\u2019s First Open-Weight Model Bets 41B Active Parameters on Customization, Not Raw Power",
    "That single ratio is the whole argument behind the first model Mira Murati\u2019s lab has trained from scratch and released as [open weights](https://thinkingmachines.ai/news/introducing-inkling/) on July 15, 2026."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Thinking Machines Inkling Is Here: Why Mira Murati\u2019s First Open-Weight Model Bets 41B Active Parameters on Customization, Not Raw Power",
    "That single ratio is the whole argument behind the first model Mira Murati\u2019s lab has trained from scratch and released as [open weights](https://thinkingmachines.ai/news/introducing-inkling/) on July 15, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Thinking Machines Inkling Is Here: Why Mira Murati\u2019s First Open-Weight Model Bets 41B Active Parameters on Customization, Not Raw Power",
    "Inkling is a Mixture-of-Experts (MoE) transformer with 975 billion total parameters, but a routing layer activates only about 41 billion of them per token.",
    "The context window stretches to 1 million tokens, enough to hold an entire codebase or a stack of contracts in a single prompt.",
    "And \u201cthinking effort\u201d is controllable, so you can dial reasoning depth up for a hard problem or down to save latency and cost."
   ]
  },
  {
   "url": "https://felloai.com/longcat-2-0",
   "title": "LongCat-2.0: Meituan\u2019s 1.6 Trillion-Parameter Open-Source Coding Model",
   "published_at": "2026-06-30T18:28:44",
   "organization": "Meituan",
   "model": null,
   "release_date": "2026-06-30",
   "access_type": null,
   "license": null,
   "significance": "1.6 trillion-parameter Mixture-of-Experts model with a native 1-million-token context window",
   "organization_evidence": [
    "# LongCat-2.0: Meituan\u2019s 1.6 Trillion-Parameter Open-Source Coding Model",
    "**Meituan** has open-sourced **LongCat-2.0**, a **1.6 trillion-parameter** Mixture-of-Experts model with a native **1-million-token context window**, released on **June 30, 2026** under a permissive **MIT license**."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# LongCat-2.0: Meituan\u2019s 1.6 Trillion-Parameter Open-Source Coding Model",
    "**Meituan** has open-sourced **LongCat-2.0**, a **1.6 trillion-parameter** Mixture-of-Experts model with a native **1-million-token context window**, released on **June 30, 2026** under a permissive **MIT license**."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# LongCat-2.0: Meituan\u2019s 1.6 Trillion-Parameter Open-Source Coding Model",
    "**Meituan** has open-sourced **LongCat-2.0**, a **1.6 trillion-parameter** Mixture-of-Experts model with a native **1-million-token context window**, released on **June 30, 2026** under a permissive **MIT license**."
   ]
  },
  {
   "url": "https://gadgetfee.com/tech-trends-innovations/think-machines-and-weights-unlocking-the-secrets-of-artificial-intelligence",
   "title": "Think Machines And Weights: Unlocking The Secrets Of Artificial Intelligence - GadgetFee",
   "published_at": "2026-07-17T04:51:38",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "A mixture-of-experts model with 975 billion total parameters and 41 billion active parameters, supporting a 1-million-token context window and pretrained on 45 trillion tokens spanning text, images, audio, and video.",
   "organization_evidence": [
    "Thinking Machines Lab released its first foundation model, Inkling, with full weights under Apache 2.0 and immediate support from major inference frameworks.",
    "Thinking Machines Lab",
    "Thinking Machines Lab released Inkling\u2019s full weights on July 15 before offering a closed API, making open distribution central to its first foundation-model launch."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "**Thinking Machines Lab**, the 17-month-old company founded by former OpenAI chief technology officer **Mira Murati**, released its first foundation model, Inkling, on July 15 with **full weights available immediately** on Hugging Face under an Apache 2.0 license.",
    "Thinking Machines Lab released its first foundation model, Inkling, with full weights under Apache 2.0 and immediate support from major inference frameworks.",
    "**announcement**When: **announced July 15, 2026; benchmark and\u2026**"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "**Inkling** is a mixture-of-experts model with **975 billion total parameters** and 41 billion active parameters, according to specifications published by Thinking Machines.",
    "Thinking Machines Lab released its first foundation model, Inkling, with full weights under Apache 2.0 and immediate support from major inference frameworks."
   ]
  },
  {
   "url": "https://cleverhack.com/the-urgency-of-open-source-ai",
   "title": "The Urgency of Open Source AI",
   "published_at": null,
   "organization": "DeepSeek",
   "model": "DeepSeek-V4-Pro",
   "release_date": "2026-04",
   "access_type": "open source",
   "license": "MIT",
   "significance": "A 1.6T total / 49B active parameter sparse MoE model pre-trained on 33T tokens with advanced attention mechanisms and reasoning effort modes.",
   "organization_evidence": [
    "## `Recent Major Open Source Model Release Timeline`",
    "- DeepSeek-V4-Pro (April 2026) - China \u2014 A 1.6T total / 49B active parameter sparse MoE model pre-trained on 33T tokens. Features a hybrid Compressed Sparse Attention + Heavily Compressed Attention architecture that requires only 27% of single-token inference FLOPs and 10% of KV cache compared to DeepSeek-V3.2 at 1M context, plus Manifold-Constrained Hyper-Connections and a Muon optimizer. Supports three reasoning effort modes (non-think, think high, think max). Scores 80.6% on SWE-bench Verified, 67.9% on Terminal-Bench 2.0, and 89.8% on IMOAnswerBench. Post-training uses a two-stage paradigm: domain-specific expert cultivation via SFT + GRPO, followed by unified consolidation via on-policy distillation. Supports 1M token context. Released under a MIT license.\ud83d\udd17:https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro"
   ],
   "model_evidence": [
    "## `Recent Major Open Source Model Release Timeline`",
    "- DeepSeek-V4-Pro (April 2026) - China \u2014 A 1.6T total / 49B active parameter sparse MoE model pre-trained on 33T tokens. Features a hybrid Compressed Sparse Attention + Heavily Compressed Attention architecture that requires only 27% of single-token inference FLOPs and 10% of KV cache compared to DeepSeek-V3.2 at 1M context, plus Manifold-Constrained Hyper-Connections and a Muon optimizer. Supports three reasoning effort modes (non-think, think high, think max). Scores 80.6% on SWE-bench Verified, 67.9% on Terminal-Bench 2.0, and 89.8% on IMOAnswerBench. Post-training uses a two-stage paradigm: domain-specific expert cultivation via SFT + GRPO, followed by unified consolidation via on-policy distillation. Supports 1M token context. Released under a MIT license.\ud83d\udd17:https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro"
   ],
   "release_date_evidence": [
    "### `April 2026`",
    "- DeepSeek-V4-Pro (April 2026) - China \u2014 A 1.6T total / 49B active parameter sparse MoE model pre-trained on 33T tokens. Features a hybrid Compressed Sparse Attention + Heavily Compressed Attention architecture that requires only 27% of single-token inference FLOPs and 10% of KV cache compared to DeepSeek-V3.2 at 1M context, plus Manifold-Constrained Hyper-Connections and a Muon optimizer. Supports three reasoning effort modes (non-think, think high, think max). Scores 80.6% on SWE-bench Verified, 67.9% on Terminal-Bench 2.0, and 89.8% on IMOAnswerBench. Post-training uses a two-stage paradigm: domain-specific expert cultivation via SFT + GRPO, followed by unified consolidation via on-policy distillation. Supports 1M token context. Released under a MIT license.\ud83d\udd17:https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro"
   ],
   "access_type_evidence": [
    "## `Recent Major Open Source Model Release Timeline`",
    "- DeepSeek-V4-Pro (April 2026) - China \u2014 A 1.6T total / 49B active parameter sparse MoE model pre-trained on 33T tokens. Features a hybrid Compressed Sparse Attention + Heavily Compressed Attention architecture that requires only 27% of single-token inference FLOPs and 10% of KV cache compared to DeepSeek-V3.2 at 1M context, plus Manifold-Constrained Hyper-Connections and a Muon optimizer. Supports three reasoning effort modes (non-think, think high, think max). Scores 80.6% on SWE-bench Verified, 67.9% on Terminal-Bench 2.0, and 89.8% on IMOAnswerBench. Post-training uses a two-stage paradigm: domain-specific expert cultivation via SFT + GRPO, followed by unified consolidation via on-policy distillation. Supports 1M token context. Released under a MIT license.\ud83d\udd17:https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro"
   ],
   "license_evidence": [
    "## `Recent Major Open Source Model Release Timeline`",
    "- DeepSeek-V4-Pro (April 2026) - China \u2014 A 1.6T total / 49B active parameter sparse MoE model pre-trained on 33T tokens. Features a hybrid Compressed Sparse Attention + Heavily Compressed Attention architecture that requires only 27% of single-token inference FLOPs and 10% of KV cache compared to DeepSeek-V3.2 at 1M context, plus Manifold-Constrained Hyper-Connections and a Muon optimizer. Supports three reasoning effort modes (non-think, think high, think max). Scores 80.6% on SWE-bench Verified, 67.9% on Terminal-Bench 2.0, and 89.8% on IMOAnswerBench. Post-training uses a two-stage paradigm: domain-specific expert cultivation via SFT + GRPO, followed by unified consolidation via on-policy distillation. Supports 1M token context. Released under a MIT license.\ud83d\udd17:https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro"
   ],
   "significance_evidence": [
    "## `Recent Major Open Source Model Release Timeline`",
    "- DeepSeek-V4-Pro (April 2026) - China \u2014 A 1.6T total / 49B active parameter sparse MoE model pre-trained on 33T tokens. Features a hybrid Compressed Sparse Attention + Heavily Compressed Attention architecture that requires only 27% of single-token inference FLOPs and 10% of KV cache compared to DeepSeek-V3.2 at 1M context, plus Manifold-Constrained Hyper-Connections and a Muon optimizer. Supports three reasoning effort modes (non-think, think high, think max)."
   ]
  },
  {
   "url": "https://tobiasreithmeier.de/en/blog/kimi-k3-open-weights-ai-market",
   "title": "Kimi K3: The Frontier Just Went Open",
   "published_at": "2026-07-19T00:00:00",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-16",
   "access_type": null,
   "license": null,
   "significance": "2.8 trillion parameters, multimodal, a one-million-token context window, and third to fourth place among all models worldwide with first place in some disciplines.",
   "organization_evidence": [
    "# Kimi K3: The Frontier Just Went Open",
    "On July 16, 2026, the Chinese startup Moonshot AI unveiled its new flagship, Kimi K3: 2.8 trillion parameters, multimodal, a one-million-token context window - and the announcement that the complete model weights will be published freely on Hugging Face by the end of July."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Kimi K3: The Frontier Just Went Open",
    "On July 16, 2026, the Chinese startup Moonshot AI unveiled its new flagship, Kimi K3: 2.8 trillion parameters, multimodal, a one-million-token context window - and the announcement that the complete model weights will be published freely on Hugging Face by the end of July."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Kimi K3: The Frontier Just Went Open",
    "On July 16, 2026, the Chinese startup Moonshot AI unveiled its new flagship, Kimi K3: 2.8 trillion parameters, multimodal, a one-million-token context window - and the announcement that the complete model weights will be published freely on Hugging Face by the end of July."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/nvidia-nemotron-3-super-120b-a12b-vs-gpt-oss-120b",
   "title": "Nemotron 3 Super 120B A12B (Reasoning) vs gpt-oss-120b (high): Model Comparison",
   "published_at": null,
   "organization": "NVIDIA",
   "model": null,
   "release_date": "2026-03",
   "access_type": null,
   "license": null,
   "significance": "120.6B total parameters, 12.7B active at inference time, open source weights, with reasoning and 1M context window",
   "organization_evidence": [
    "# Nemotron 3 Super 120B A12B (Reasoning) vs. gpt-oss-120b (high)",
    "| | ![NVIDIA logo](/img/logos/nvidia_small.svg)NVIDIA Nemotron 3 Super 120B A12B (Reasoning) | ![OpenAI logo](/img/logos/openai_small.svg)OpenAI gpt-oss-120b (high) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Nemotron 3 Super 120B A12B (Reasoning) vs. gpt-oss-120b (high)",
    "| Release Date | March 2026 | August 2025 | Nemotron 3 Super 120B A12B (Reasoning) has a more recent release date than gpt-oss-120b (high) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Nemotron 3 Super 120B A12B (Reasoning) vs. gpt-oss-120b (high)",
    "| Parameters | 120.6B, 12.7B active at inference time | 117B, 5.1B active at inference time | Nemotron 3 Super 120B A12B (Reasoning) has more parameters than gpt-oss-120b (high) |",
    "| Reasoning | Yes | Yes | Both Nemotron 3 Super 120B A12B (Reasoning) and gpt-oss-120b (high) have reasoning |",
    "| Open Source (Weights) | Yes | Yes | Both Nemotron 3 Super 120B A12B (Reasoning) and gpt-oss-120b (high) are open source |"
   ]
  },
  {
   "url": "https://kingy.ai/blog/inkling-ai-model-benchmarks-specs-open-weights",
   "title": "Inkling AI Model: Benchmarks, Specs and the Open-Weight Bet",
   "published_at": "2026-07-15T19:25:03",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "A 975-billion-parameter, natively multimodal mixture-of-experts system released with downloadable Apache 2.0 weights.",
   "organization_evidence": [
    "AI News",
    "# Inkling AI Model: Benchmarks, Specs and the Open-Weight Bet",
    "Thinking Machines Lab\u2019s Inkling AI model is a 975-billion-parameter, natively multimodal mixture-of-experts system released with downloadable Apache 2.0 weights."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "AI News",
    "# Inkling AI Model: Benchmarks, Specs and the Open-Weight Bet",
    "July 15, 2026 Curtis Pyke [Blog](https://kingy.ai/category/blog/)"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "AI News",
    "# Inkling AI Model: Benchmarks, Specs and the Open-Weight Bet",
    "Thinking Machines Lab\u2019s Inkling AI model is a 975-billion-parameter, natively multimodal mixture-of-experts system released with downloadable Apache 2.0 weights."
   ]
  },
  {
   "url": "https://cdn.localaimaster.com/blog/latest-ai-models-october-2025-roundup",
   "title": "Best Open-Source LLMs (2026): Free Models Ranked",
   "published_at": "2026-02-04T00:00:00",
   "organization": "Z.ai",
   "model": "GLM-5.2",
   "release_date": "2026-06-17",
   "access_type": "open weights",
   "license": "MIT",
   "significance": "753B MoE model",
   "organization_evidence": [
    "## New: The June-July 2026 Giant-Model Wave",
    "Six weeks rewrote the top of the open-weights leaderboard.",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **[GLM-5.2](/models/glm-5-2)** (Z.ai) | Jun 17 | 753B MoE / ~40B | MIT | **Big rigs, yes** \u2014 112 ready-made quants for llama.cpp/Ollama/LM Studio; think Mac Studio 512GB or EPYC + KTransformers, ~200-400GB quantized |"
   ],
   "model_evidence": [
    "## New: The June-July 2026 Giant-Model Wave",
    "Six weeks rewrote the top of the open-weights leaderboard.",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **[GLM-5.2](/models/glm-5-2)** (Z.ai) | Jun 17 | 753B MoE / ~40B | MIT | **Big rigs, yes** \u2014 112 ready-made quants for llama.cpp/Ollama/LM Studio; think Mac Studio 512GB or EPYC + KTransformers, ~200-400GB quantized |"
   ],
   "release_date_evidence": [
    "## New: The June-July 2026 Giant-Model Wave",
    "Six weeks rewrote the top of the open-weights leaderboard.",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **[GLM-5.2](/models/glm-5-2)** (Z.ai) | Jun 17 | 753B MoE / ~40B | MIT | **Big rigs, yes** \u2014 112 ready-made quants for llama.cpp/Ollama/LM Studio; think Mac Studio 512GB or EPYC + KTransformers, ~200-400GB quantized |"
   ],
   "access_type_evidence": [
    "## New: The June-July 2026 Giant-Model Wave",
    "Six weeks rewrote the top of the open-weights leaderboard.",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **[GLM-5.2](/models/glm-5-2)** (Z.ai) | Jun 17 | 753B MoE / ~40B | MIT | **Big rigs, yes** \u2014 112 ready-made quants for llama.cpp/Ollama/LM Studio; think Mac Studio 512GB or EPYC + KTransformers, ~200-400GB quantized |"
   ],
   "license_evidence": [
    "## New: The June-July 2026 Giant-Model Wave",
    "Six weeks rewrote the top of the open-weights leaderboard.",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **[GLM-5.2](/models/glm-5-2)** (Z.ai) | Jun 17 | 753B MoE / ~40B | MIT | **Big rigs, yes** \u2014 112 ready-made quants for llama.cpp/Ollama/LM Studio; think Mac Studio 512GB or EPYC + KTransformers, ~200-400GB quantized |"
   ],
   "significance_evidence": [
    "## New: The June-July 2026 Giant-Model Wave",
    "Six weeks rewrote the top of the open-weights leaderboard.",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **[GLM-5.2](/models/glm-5-2)** (Z.ai) | Jun 17 | 753B MoE / ~40B | MIT | **Big rigs, yes** \u2014 112 ready-made quants for llama.cpp/Ollama/LM Studio; think Mac Studio 512GB or EPYC + KTransformers, ~200-400GB quantized |"
   ]
  },
  {
   "url": "https://cdn.localaimaster.com/blog/latest-ai-models-october-2025-roundup",
   "title": "Best Open-Source LLMs (2026): Free Models Ranked",
   "published_at": "2026-02-04T00:00:00",
   "organization": "Nvidia",
   "model": "Nemotron 3 Ultra",
   "release_date": "2026-06-04",
   "access_type": "open weights",
   "license": "OpenMDW",
   "significance": "~550B / 55B MoE model",
   "organization_evidence": [
    "## New: The June-July 2026 Giant-Model Wave",
    "Six weeks rewrote the top of the open-weights leaderboard.",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **Nemotron 3 Ultra** (Nvidia) | Jun 4 | ~550B / 55B | OpenMDW | Workstation-class multi-GPU; NVFP4 weights published |"
   ],
   "model_evidence": [
    "## New: The June-July 2026 Giant-Model Wave",
    "Six weeks rewrote the top of the open-weights leaderboard.",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **Nemotron 3 Ultra** (Nvidia) | Jun 4 | ~550B / 55B | OpenMDW | Workstation-class multi-GPU; NVFP4 weights published |"
   ],
   "release_date_evidence": [
    "## New: The June-July 2026 Giant-Model Wave",
    "Six weeks rewrote the top of the open-weights leaderboard.",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **Nemotron 3 Ultra** (Nvidia) | Jun 4 | ~550B / 55B | OpenMDW | Workstation-class multi-GPU; NVFP4 weights published |"
   ],
   "access_type_evidence": [
    "## New: The June-July 2026 Giant-Model Wave",
    "Six weeks rewrote the top of the open-weights leaderboard.",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **Nemotron 3 Ultra** (Nvidia) | Jun 4 | ~550B / 55B | OpenMDW | Workstation-class multi-GPU; NVFP4 weights published |"
   ],
   "license_evidence": [
    "## New: The June-July 2026 Giant-Model Wave",
    "Six weeks rewrote the top of the open-weights leaderboard.",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **Nemotron 3 Ultra** (Nvidia) | Jun 4 | ~550B / 55B | OpenMDW | Workstation-class multi-GPU; NVFP4 weights published |"
   ],
   "significance_evidence": [
    "## New: The June-July 2026 Giant-Model Wave",
    "Six weeks rewrote the top of the open-weights leaderboard.",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **Nemotron 3 Ultra** (Nvidia) | Jun 4 | ~550B / 55B | OpenMDW | Workstation-class multi-GPU; NVFP4 weights published |"
   ]
  },
  {
   "url": "https://cdn.localaimaster.com/blog/latest-ai-models-october-2025-roundup",
   "title": "Best Open-Source LLMs (2026): Free Models Ranked",
   "published_at": "2026-02-04T00:00:00",
   "organization": "MiniMax",
   "model": "MiniMax M3",
   "release_date": "2026-06-01",
   "access_type": "open weights",
   "license": "Open weights",
   "significance": "428B MoE model",
   "organization_evidence": [
    "## New: The June-July 2026 Giant-Model Wave",
    "Six weeks rewrote the top of the open-weights leaderboard.",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **MiniMax M3** | Jun 1 | 428B / ~23B | Open weights | Big-RAM rigs quantized (official MXFP8 + community GGUF) |"
   ],
   "model_evidence": [
    "## New: The June-July 2026 Giant-Model Wave",
    "Six weeks rewrote the top of the open-weights leaderboard.",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **MiniMax M3** | Jun 1 | 428B / ~23B | Open weights | Big-RAM rigs quantized (official MXFP8 + community GGUF) |"
   ],
   "release_date_evidence": [
    "## New: The June-July 2026 Giant-Model Wave",
    "Six weeks rewrote the top of the open-weights leaderboard.",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **MiniMax M3** | Jun 1 | 428B / ~23B | Open weights | Big-RAM rigs quantized (official MXFP8 + community GGUF) |"
   ],
   "access_type_evidence": [
    "## New: The June-July 2026 Giant-Model Wave",
    "Six weeks rewrote the top of the open-weights leaderboard.",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **MiniMax M3** | Jun 1 | 428B / ~23B | Open weights | Big-RAM rigs quantized (official MXFP8 + community GGUF) |"
   ],
   "license_evidence": [
    "## New: The June-July 2026 Giant-Model Wave",
    "Six weeks rewrote the top of the open-weights leaderboard.",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **MiniMax M3** | Jun 1 | 428B / ~23B | Open weights | Big-RAM rigs quantized (official MXFP8 + community GGUF) |"
   ],
   "significance_evidence": [
    "## New: The June-July 2026 Giant-Model Wave",
    "Six weeks rewrote the top of the open-weights leaderboard.",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **MiniMax M3** | Jun 1 | 428B / ~23B | Open weights | Big-RAM rigs quantized (official MXFP8 + community GGUF) |"
   ]
  },
  {
   "url": "https://cdn.localaimaster.com/blog/latest-ai-models-october-2025-roundup",
   "title": "Best Open-Source LLMs (2026): Free Models Ranked",
   "published_at": "2026-02-04T00:00:00",
   "organization": "Tencent",
   "model": "Hy3",
   "release_date": "2026-07-06",
   "access_type": "open weights",
   "license": "Apache 2.0",
   "significance": "295B MoE model",
   "organization_evidence": [
    "## New: The June-July 2026 Giant-Model Wave",
    "Six weeks rewrote the top of the open-weights leaderboard.",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **Hy3** (Tencent) | Jul 6 | 295B / 21B | Apache 2.0 | The most home-plausible giant \u2014 official FP8; big multi-GPU or unified-memory rigs |"
   ],
   "model_evidence": [
    "## New: The June-July 2026 Giant-Model Wave",
    "Six weeks rewrote the top of the open-weights leaderboard.",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **Hy3** (Tencent) | Jul 6 | 295B / 21B | Apache 2.0 | The most home-plausible giant \u2014 official FP8; big multi-GPU or unified-memory rigs |"
   ],
   "release_date_evidence": [
    "## New: The June-July 2026 Giant-Model Wave",
    "Six weeks rewrote the top of the open-weights leaderboard.",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **Hy3** (Tencent) | Jul 6 | 295B / 21B | Apache 2.0 | The most home-plausible giant \u2014 official FP8; big multi-GPU or unified-memory rigs |"
   ],
   "access_type_evidence": [
    "## New: The June-July 2026 Giant-Model Wave",
    "Six weeks rewrote the top of the open-weights leaderboard.",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **Hy3** (Tencent) | Jul 6 | 295B / 21B | Apache 2.0 | The most home-plausible giant \u2014 official FP8; big multi-GPU or unified-memory rigs |"
   ],
   "license_evidence": [
    "## New: The June-July 2026 Giant-Model Wave",
    "Six weeks rewrote the top of the open-weights leaderboard.",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **Hy3** (Tencent) | Jul 6 | 295B / 21B | Apache 2.0 | The most home-plausible giant \u2014 official FP8; big multi-GPU or unified-memory rigs |"
   ],
   "significance_evidence": [
    "## New: The June-July 2026 Giant-Model Wave",
    "Six weeks rewrote the top of the open-weights leaderboard.",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **Hy3** (Tencent) | Jul 6 | 295B / 21B | Apache 2.0 | The most home-plausible giant \u2014 official FP8; big multi-GPU or unified-memory rigs |"
   ]
  },
  {
   "url": "https://cdn.localaimaster.com/blog/latest-ai-models-october-2025-roundup",
   "title": "Best Open-Source LLMs (2026): Free Models Ranked",
   "published_at": "2026-02-04T00:00:00",
   "organization": "Thinking Machines",
   "model": "Inkling",
   "release_date": "2026-07-15",
   "access_type": "open weights",
   "license": "Apache 2.0 (reported)",
   "significance": "975B MoE model",
   "organization_evidence": [
    "## New: The June-July 2026 Giant-Model Wave",
    "Six weeks rewrote the top of the open-weights leaderboard.",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **Inkling** (Thinking Machines) | Jul 15 | 975B / 41B | Apache 2.0 (reported) | Datacenter-class (~2TB at BF16); watch for distilled variants |"
   ],
   "model_evidence": [
    "## New: The June-July 2026 Giant-Model Wave",
    "Six weeks rewrote the top of the open-weights leaderboard.",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **Inkling** (Thinking Machines) | Jul 15 | 975B / 41B | Apache 2.0 (reported) | Datacenter-class (~2TB at BF16); watch for distilled variants |"
   ],
   "release_date_evidence": [
    "## New: The June-July 2026 Giant-Model Wave",
    "Six weeks rewrote the top of the open-weights leaderboard.",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **Inkling** (Thinking Machines) | Jul 15 | 975B / 41B | Apache 2.0 (reported) | Datacenter-class (~2TB at BF16); watch for distilled variants |"
   ],
   "access_type_evidence": [
    "## New: The June-July 2026 Giant-Model Wave",
    "Six weeks rewrote the top of the open-weights leaderboard.",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **Inkling** (Thinking Machines) | Jul 15 | 975B / 41B | Apache 2.0 (reported) | Datacenter-class (~2TB at BF16); watch for distilled variants |"
   ],
   "license_evidence": [
    "## New: The June-July 2026 Giant-Model Wave",
    "Six weeks rewrote the top of the open-weights leaderboard.",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **Inkling** (Thinking Machines) | Jul 15 | 975B / 41B | Apache 2.0 (reported) | Datacenter-class (~2TB at BF16); watch for distilled variants |"
   ],
   "significance_evidence": [
    "## New: The June-July 2026 Giant-Model Wave",
    "Six weeks rewrote the top of the open-weights leaderboard.",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **Inkling** (Thinking Machines) | Jul 15 | 975B / 41B | Apache 2.0 (reported) | Datacenter-class (~2TB at BF16); watch for distilled variants |"
   ]
  },
  {
   "url": "https://cdn.localaimaster.com/blog/latest-ai-models-october-2025-roundup",
   "title": "Best Open-Source LLMs (2026): Free Models Ranked",
   "published_at": "2026-02-04T00:00:00",
   "organization": "Moonshot",
   "model": "Kimi K3",
   "release_date": "2026-07-16",
   "access_type": "open weights",
   "license": "TBD",
   "significance": "2.8T MoE model",
   "organization_evidence": [
    "## New: The June-July 2026 Giant-Model Wave",
    "Six weeks rewrote the top of the open-weights leaderboard.",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **[Kimi K3](/models/kimi-k3)** (Moonshot) | Jul 16 | 2.8T MoE / ~1.8% active | TBD | **No** \u2014 ~1.4TB even at native 4-bit; weights land ~Jul 27, datacenter-only |"
   ],
   "model_evidence": [
    "## New: The June-July 2026 Giant-Model Wave",
    "Six weeks rewrote the top of the open-weights leaderboard.",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **[Kimi K3](/models/kimi-k3)** (Moonshot) | Jul 16 | 2.8T MoE / ~1.8% active | TBD | **No** \u2014 ~1.4TB even at native 4-bit; weights land ~Jul 27, datacenter-only |"
   ],
   "release_date_evidence": [
    "## New: The June-July 2026 Giant-Model Wave",
    "Six weeks rewrote the top of the open-weights leaderboard.",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **[Kimi K3](/models/kimi-k3)** (Moonshot) | Jul 16 | 2.8T MoE / ~1.8% active | TBD | **No** \u2014 ~1.4TB even at native 4-bit; weights land ~Jul 27, datacenter-only |"
   ],
   "access_type_evidence": [
    "## New: The June-July 2026 Giant-Model Wave",
    "Six weeks rewrote the top of the open-weights leaderboard.",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **[Kimi K3](/models/kimi-k3)** (Moonshot) | Jul 16 | 2.8T MoE / ~1.8% active | TBD | **No** \u2014 ~1.4TB even at native 4-bit; weights land ~Jul 27, datacenter-only |"
   ],
   "license_evidence": [
    "## New: The June-July 2026 Giant-Model Wave",
    "Six weeks rewrote the top of the open-weights leaderboard.",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **[Kimi K3](/models/kimi-k3)** (Moonshot) | Jul 16 | 2.8T MoE / ~1.8% active | TBD | **No** \u2014 ~1.4TB even at native 4-bit; weights land ~Jul 27, datacenter-only |"
   ],
   "significance_evidence": [
    "## New: The June-July 2026 Giant-Model Wave",
    "Six weeks rewrote the top of the open-weights leaderboard.",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **[Kimi K3](/models/kimi-k3)** (Moonshot) | Jul 16 | 2.8T MoE / ~1.8% active | TBD | **No** \u2014 ~1.4TB even at native 4-bit; weights land ~Jul 27, datacenter-only |"
   ]
  },
  {
   "url": "https://cdn.localaimaster.com/blog/latest-ai-models-october-2025-roundup",
   "title": "Best Open-Source LLMs (2026): Free Models Ranked",
   "published_at": "2026-02-04T00:00:00",
   "organization": "Google",
   "model": "Gemma 4 12B Unified",
   "release_date": "2026-06-03",
   "access_type": "open weights",
   "license": "Apache 2.0",
   "significance": "Consumer-class 12B release with text+audio+image input and 256K context",
   "organization_evidence": [
    "## New: The June-July 2026 Giant-Model Wave",
    "Six weeks rewrote the top of the open-weights leaderboard.",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **[Gemma 4 12B Unified](/models/gemma-4)** (Google) | Jun 3 | 12B | Apache 2.0 | **Yes, easily** \u2014 the one genuinely consumer-class release of the wave; text+audio+image input, 256K context |"
   ],
   "model_evidence": [
    "## New: The June-July 2026 Giant-Model Wave",
    "Six weeks rewrote the top of the open-weights leaderboard.",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **[Gemma 4 12B Unified](/models/gemma-4)** (Google) | Jun 3 | 12B | Apache 2.0 | **Yes, easily** \u2014 the one genuinely consumer-class release of the wave; text+audio+image input, 256K context |"
   ],
   "release_date_evidence": [
    "## New: The June-July 2026 Giant-Model Wave",
    "Six weeks rewrote the top of the open-weights leaderboard.",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **[Gemma 4 12B Unified](/models/gemma-4)** (Google) | Jun 3 | 12B | Apache 2.0 | **Yes, easily** \u2014 the one genuinely consumer-class release of the wave; text+audio+image input, 256K context |"
   ],
   "access_type_evidence": [
    "## New: The June-July 2026 Giant-Model Wave",
    "Six weeks rewrote the top of the open-weights leaderboard.",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **[Gemma 4 12B Unified](/models/gemma-4)** (Google) | Jun 3 | 12B | Apache 2.0 | **Yes, easily** \u2014 the one genuinely consumer-class release of the wave; text+audio+image input, 256K context |"
   ],
   "license_evidence": [
    "## New: The June-July 2026 Giant-Model Wave",
    "Six weeks rewrote the top of the open-weights leaderboard.",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **[Gemma 4 12B Unified](/models/gemma-4)** (Google) | Jun 3 | 12B | Apache 2.0 | **Yes, easily** \u2014 the one genuinely consumer-class release of the wave; text+audio+image input, 256K context |"
   ],
   "significance_evidence": [
    "## New: The June-July 2026 Giant-Model Wave",
    "Six weeks rewrote the top of the open-weights leaderboard.",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **[Gemma 4 12B Unified](/models/gemma-4)** (Google) | Jun 3 | 12B | Apache 2.0 | **Yes, easily** \u2014 the one genuinely consumer-class release of the wave; text+audio+image input, 256K context |"
   ]
  },
  {
   "url": "https://venturebeat.com/data/mistral-launches-ocr-4-turning-document-extraction-into-a-full-enterprise-ai-play",
   "title": "Mistral launches OCR 4, turning document extraction into a full enterprise AI play",
   "published_at": "2026-06-24T21:04:04",
   "organization": "Baidu",
   "model": "Unlimited-OCR",
   "release_date": "2026-06-22",
   "access_type": "open-weight",
   "license": "MIT license",
   "significance": "A 3-billion-parameter model that parses entire PDFs and multi-page scans in a single forward pass using Reference Sliding Window Attention.",
   "organization_evidence": [
    "Just one day before [OCR 4](https://mistral.ai/news/ocr-4/) launched, Baidu shipped [Unlimited-OCR](https://huggingface.co/baidu/Unlimited-OCR) on June 22 \u2014 a 3-billion-parameter MIT-licensed model that tackles one of the most persistent pain points in document AI: parsing entire PDFs and multi-page scans in a single forward pass, without chunking the input or stitching the output back together afterward.",
    "Baidu's free, open-weight OCR model arrived one day earlier \u2014 and the contrast is revealing"
   ],
   "model_evidence": [
    "Just one day before [OCR 4](https://mistral.ai/news/ocr-4/) launched, Baidu shipped [Unlimited-OCR](https://huggingface.co/baidu/Unlimited-OCR) on June 22 \u2014 a 3-billion-parameter MIT-licensed model that tackles one of the most persistent pain points in document AI: parsing entire PDFs and multi-page scans in a single forward pass, without chunking the input or stitching the output back together afterward.",
    "Baidu's free, open-weight OCR model arrived one day earlier \u2014 and the contrast is revealing"
   ],
   "release_date_evidence": [
    "Just one day before [OCR 4](https://mistral.ai/news/ocr-4/) launched, Baidu shipped [Unlimited-OCR](https://huggingface.co/baidu/Unlimited-OCR) on June 22 \u2014 a 3-billion-parameter MIT-licensed model that tackles one of the most persistent pain points in document AI: parsing entire PDFs and multi-page scans in a single forward pass, without chunking the input or stitching the output back together afterward."
   ],
   "access_type_evidence": [
    "Baidu's free, open-weight OCR model arrived one day earlier \u2014 and the contrast is revealing",
    "The two releases frame what some analysts are calling the June 2026 document-AI split: self-hosted long-horizon parsing with open weights versus structured managed extraction with enterprise features."
   ],
   "license_evidence": [
    "Just one day before [OCR 4](https://mistral.ai/news/ocr-4/) launched, Baidu shipped [Unlimited-OCR](https://huggingface.co/baidu/Unlimited-OCR) on June 22 \u2014 a 3-billion-parameter MIT-licensed model that tackles one of the most persistent pain points in document AI: parsing entire PDFs and multi-page scans in a single forward pass, without chunking the input or stitching the output back together afterward."
   ],
   "significance_evidence": [
    "Just one day before [OCR 4](https://mistral.ai/news/ocr-4/) launched, Baidu shipped [Unlimited-OCR](https://huggingface.co/baidu/Unlimited-OCR) on June 22 \u2014 a 3-billion-parameter MIT-licensed model that tackles one of the most persistent pain points in document AI: parsing entire PDFs and multi-page scans in a single forward pass, without chunking the input or stitching the output back together afterward."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/glm-5-1-vs-qwen3-5-omni-plus",
   "title": "GLM-5.1 (Reasoning) vs Qwen3.5 Omni Plus: Model Comparison",
   "published_at": null,
   "organization": "Z AI",
   "model": "GLM-5.1 (Reasoning)",
   "release_date": "April, 2026",
   "access_type": "open source",
   "license": null,
   "significance": "GLM-5.1 (Reasoning) is smaller than Qwen3.5 Omni Plus",
   "organization_evidence": [
    "# GLM-5.1 (Reasoning) vs. Qwen3.5 Omni Plus",
    "## Model Comparison",
    "| Metric | ![Z AI logo](/img/logos/zai_small.svg)GLM-5.1 (Reasoning) | ![Alibaba logo](/img/logos/alibaba_small.svg)Qwen3.5 Omni Plus | Analysis |",
    "| Creator | ![Z AI](/img/logos/zai_small.svg)[](https://z.ai/) | ![Alibaba](/img/logos/alibaba_small.svg)[](https://www.alibabacloud.com/) | |"
   ],
   "model_evidence": [
    "# GLM-5.1 (Reasoning) vs. Qwen3.5 Omni Plus",
    "## Model Comparison",
    "| Metric | ![Z AI logo](/img/logos/zai_small.svg)GLM-5.1 (Reasoning) | ![Alibaba logo](/img/logos/alibaba_small.svg)Qwen3.5 Omni Plus | Analysis |"
   ],
   "release_date_evidence": [
    "# GLM-5.1 (Reasoning) vs. Qwen3.5 Omni Plus",
    "## Model Comparison",
    "| Metric | ![Z AI logo](/img/logos/zai_small.svg)GLM-5.1 (Reasoning) | ![Alibaba logo](/img/logos/alibaba_small.svg)Qwen3.5 Omni Plus | Analysis |",
    "| Release Date | April, 2026 | March, 2026 | GLM-5.1 (Reasoning) has a more recent release date than Qwen3.5 Omni Plus |"
   ],
   "access_type_evidence": [
    "# GLM-5.1 (Reasoning) vs. Qwen3.5 Omni Plus",
    "## Model Comparison",
    "| Metric | ![Z AI logo](/img/logos/zai_small.svg)GLM-5.1 (Reasoning) | ![Alibaba logo](/img/logos/alibaba_small.svg)Qwen3.5 Omni Plus | Analysis |",
    "| Open Source (Weights) | Yes[](https://huggingface.co/zai-org/GLM-5.1) | No | GLM-5.1 (Reasoning) is open source while Qwen3.5 Omni Plus is proprietary |"
   ],
   "license_evidence": null,
   "significance_evidence": [
    "# GLM-5.1 (Reasoning) vs. Qwen3.5 Omni Plus",
    "## Model Comparison",
    "| Metric | ![Z AI logo](/img/logos/zai_small.svg)GLM-5.1 (Reasoning) | ![Alibaba logo](/img/logos/alibaba_small.svg)Qwen3.5 Omni Plus | Analysis |",
    "| Context Window | 200k tokens (~300 A4 pages of size 12 Arial font) | 256k tokens (~384 A4 pages of size 12 Arial font) | GLM-5.1 (Reasoning) is smaller than Qwen3.5 Omni Plus |"
   ]
  },
  {
   "url": "https://jiaweing.com/blog/china-won-open-source-while-you-were-sleeping",
   "title": "China won open source while you were sleeping",
   "published_at": "2026-04-19T00:00:00",
   "organization": "Alibaba",
   "model": "Qwen 3.5",
   "release_date": "2026-03",
   "access_type": "open-source",
   "license": null,
   "significance": "blew past the 50% mark of all global open-source model downloads",
   "organization_evidence": [
    "# China won open source while you were sleeping",
    "By March 2026, after the release of Qwen 3.5, that number blew past the 50% mark of all global open-source model downloads."
   ],
   "model_evidence": [
    "# China won open source while you were sleeping",
    "By March 2026, after the release of Qwen 3.5, that number blew past the 50% mark of all global open-source model downloads."
   ],
   "release_date_evidence": [
    "# China won open source while you were sleeping",
    "By March 2026, after the release of Qwen 3.5, that number blew past the 50% mark of all global open-source model downloads."
   ],
   "access_type_evidence": [
    "# China won open source while you were sleeping",
    "By March 2026, after the release of Qwen 3.5, that number blew past the 50% mark of all global open-source model downloads."
   ],
   "license_evidence": null,
   "significance_evidence": [
    "# China won open source while you were sleeping",
    "By March 2026, after the release of Qwen 3.5, that number blew past the 50% mark of all global open-source model downloads."
   ]
  },
  {
   "url": "https://awesomeagents.ai/news/thinking-machines-inkling-open-weights",
   "title": "Thinking Machines Opens Inkling - and Admits Its Limits",
   "published_at": "2026-07-15T18:14:28",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "A 975B-parameter mixture-of-experts model (41B active) pretrained on 45 trillion tokens",
   "organization_evidence": [
    "Thinking Machines Opens Inkling - and Admits Its Limits",
    "Mira Murati's Thinking Machines Lab released its first open-weight model, Inkling, and published benchmarks showing it losing to closed rivals on most of them."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Thinking Machines Opens Inkling - and Admits Its Limits",
    "Mira Murati's Thinking Machines Lab released its first model with fully open weights on July 15, and the announcement opens with an admission most labs bury on page ten: \"It is not the most performant model available today, closed or open.\""
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Thinking Machines Opens Inkling - and Admits Its Limits",
    "Inkling is a 975B-parameter mixture-of-experts model (41B active) with full open weights under Apache 2.0, released with a smaller preview called Inkling-Small"
   ]
  },
  {
   "url": "https://lumienai.com/news/kimi-k3-open-weights-public-download-moonshot-ai",
   "title": "Kimi K3 Open Weights Go Public: What the 2.8T Parameter Model Means",
   "published_at": "2026-07-27T03:44:14",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-27",
   "access_type": null,
   "license": null,
   "significance": "largest open-weight AI model ever released, carrying 2.8 trillion parameters, and benchmarks put its performance close to Anthropic's Claude Fable and OpenAI's GPT-5.6 Sol",
   "organization_evidence": [
    "# Kimi K3 Open Weights Go Public: What the 2.8T Parameter Model Means",
    "Moonshot AI released Kimi K3 open weights on July 27."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Kimi K3 Open Weights Go Public: What the 2.8T Parameter Model Means",
    "Moonshot AI released Kimi K3 open weights on July 27."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Kimi K3 Open Weights Go Public: What the 2.8T Parameter Model Means",
    "Kimi K3 is the largest open-weight AI model ever released, carrying 2.8 trillion parameters, and benchmarks put its performance close to Anthropic's Claude Fable and OpenAI's GPT-5.6 Sol."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/gemma-4-26b-a4b-vs-gpt-5-4-pro",
   "title": "Gemma 4 26B A4B (Reasoning) vs GPT-5.4 Pro (xhigh): Model Comparison",
   "published_at": null,
   "organization": "Google",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "Reasoning model with 256k context window and image input support",
   "organization_evidence": [
    "# Gemma 4 26B A4B (Reasoning) vs. GPT-5.4 Pro (xhigh)",
    "| Creator | ![Google](/img/logos/google_small.svg)[](https://deepmind.google/technologies/gemini/) | ![OpenAI](/img/logos/openai_small.svg)[](https://openai.com/) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Gemma 4 26B A4B (Reasoning) vs. GPT-5.4 Pro (xhigh)",
    "| Metric | ![Google logo](/img/logos/google_small.svg)Gemma 4 26B A4B (Reasoning) | ![OpenAI logo](/img/logos/openai_small.svg)GPT-5.4 Pro (xhigh) | Analysis |",
    "| Release Date | April, 2026 | March, 2026 | Gemma 4 26B A4B (Reasoning) has a more recent release date than GPT-5.4 Pro (xhigh) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Gemma 4 26B A4B (Reasoning) vs. GPT-5.4 Pro (xhigh)",
    "| Metric | ![Google logo](/img/logos/google_small.svg)Gemma 4 26B A4B (Reasoning) | ![OpenAI logo](/img/logos/openai_small.svg)GPT-5.4 Pro (xhigh) | Analysis |",
    "| Context Window | 256k tokens (~384 A4 pages of size 12 Arial font) | 1050k tokens (~1575 A4 pages of size 12 Arial font) | Gemma 4 26B A4B (Reasoning) is smaller than GPT-5.4 Pro (xhigh) |",
    "| Image Input Support | Yes | Yes | Both Gemma 4 26B A4B (Reasoning) and GPT-5.4 Pro (xhigh) have image input support |"
   ]
  },
  {
   "url": "https://press-report.net/ai-news/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - press-report.net",
   "published_at": "2026-07-15T11:50:57",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor",
   "organization_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "Chinese labs such as DeepSeek, Z.ai, Moonshot, and Alibaba now each operate distinct AI strategies, from cost-effective models to long-horizon tuned agents, reflecting a diversified and competitive landscape.",
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "Between late April and mid-June 2026, Chinese AI laboratories launched four high-capacity open models: **DeepSeek V4** on April 24, **MiniMax M3** on June 1, and **Kimi K2.7-Code** and **GLM-5.2** in mid-June.",
    "APR 24",
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor"
   ]
  },
  {
   "url": "https://press-report.net/ai-news/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - press-report.net",
   "published_at": "2026-07-15T11:50:57",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "cheap 1M-token context, native multimodal, modified-MIT",
   "organization_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "**MiniMax M3**cheap 1M-token context, native multimodal, modified-MIT"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "Between late April and mid-June 2026, Chinese AI laboratories launched four high-capacity open models: **DeepSeek V4** on April 24, **MiniMax M3** on June 1, and **Kimi K2.7-Code** and **GLM-5.2** in mid-June.",
    "JUN 01",
    "**MiniMax M3**cheap 1M-token context, native multimodal, modified-MIT"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "**MiniMax M3**cheap 1M-token context, native multimodal, modified-MIT"
   ]
  },
  {
   "url": "https://press-report.net/ai-news/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - press-report.net",
   "published_at": "2026-07-15T11:50:57",
   "organization": "Moonshot",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "agent-run specialist, ~30% fewer thinking tokens than K2.6",
   "organization_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "Chinese labs such as DeepSeek, Z.ai, Moonshot, and Alibaba now each operate distinct AI strategies, from cost-effective models to long-horizon tuned agents, reflecting a diversified and competitive landscape.",
    "**Kimi K2.7-Code (Moonshot)**agent-run specialist, ~30% fewer thinking tokens than K2.6"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "Between late April and mid-June 2026, Chinese AI laboratories launched four high-capacity open models: **DeepSeek V4** on April 24, **MiniMax M3** on June 1, and **Kimi K2.7-Code** and **GLM-5.2** in mid-June.",
    "JUN 13",
    "**Kimi K2.7-Code (Moonshot)**agent-run specialist, ~30% fewer thinking tokens than K2.6"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "**Kimi K2.7-Code (Moonshot)**agent-run specialist, ~30% fewer thinking tokens than K2.6"
   ]
  },
  {
   "url": "https://press-report.net/ai-news/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - press-report.net",
   "published_at": "2026-07-15T11:50:57",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "753B MoE, MIT, top open-weight on Artificial Analysis index",
   "organization_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "Chinese labs such as DeepSeek, Z.ai, Moonshot, and Alibaba now each operate distinct AI strategies, from cost-effective models to long-horizon tuned agents, reflecting a diversified and competitive landscape.",
    "**GLM-5.2 (Z.ai)**753B MoE, MIT, top open-weight on Artificial Analysis index"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "Between late April and mid-June 2026, Chinese AI laboratories launched four high-capacity open models: **DeepSeek V4** on April 24, **MiniMax M3** on June 1, and **Kimi K2.7-Code** and **GLM-5.2** in mid-June.",
    "JUN 13\u201316",
    "**GLM-5.2 (Z.ai)**753B MoE, MIT, top open-weight on Artificial Analysis index"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "**GLM-5.2 (Z.ai)**753B MoE, MIT, top open-weight on Artificial Analysis index"
   ]
  },
  {
   "url": "https://edgen.tech/news/post/thinking-machines-launches-inkling-a-975b-open-weight-ai-model",
   "title": "Thinking Machines launches Inkling, a 975B open-weight AI model",
   "published_at": "2026-07-15T18:27:31",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "a 975-billion-parameter mixture-of-experts model with 41 billion active parameters per task, trained on 45 trillion tokens spanning text, image, audio and video, reasoning natively across all three modalities",
   "organization_evidence": [
    "Thinking Machines launches Inkling, a 975B open-weight AI model",
    "Thinking Machines Lab released Inkling, a 975-billion-parameter open-weight model that challenges the centralized AI paradigm."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Thinking Machines launches Inkling, a 975B open-weight AI model",
    "[![Rachel Kim](https://static.edgen.tech/authors/rachel-kim.png?image_process=resize,w_32/quality,q_75/format,webp/sharpen,100)\n## Rachel Kim](https://www.edgen.tech/author/news/rachel-kim)\u00b7Jul 15 2026, 18:27"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Thinking Machines launches Inkling, a 975B open-weight AI model",
    "Thinking Machines Lab's Inkling, a 975-billion-parameter mixture-of-experts model with 41 billion active parameters per task, offers enterprises an open-weight alternative to the one-size-fits-all models from OpenAI, Anthropic and Google.",
    "Inkling was trained on 45 trillion tokens spanning text, image, audio and video, and reasons natively across all three modalities."
   ]
  },
  {
   "url": "https://baike.baidu.com/en/item/Qwen/1530291",
   "title": "Qwen",
   "published_at": null,
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-02-04",
   "access_type": null,
   "license": null,
   "significance": "open-weight language model specifically designed for coding agents and local development",
   "organization_evidence": [
    "Development Timeline",
    "On February 4, 2026, the official Alibaba Qwen Weibo account announced the launch of Qwen3-Coder-Next, an open-weight language model specifically designed for coding agents and local development."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Development Timeline",
    "On February 4, 2026, the official Alibaba Qwen Weibo account announced the launch of Qwen3-Coder-Next, an open-weight language model specifically designed for coding agents and local development."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Development Timeline",
    "On February 4, 2026, the official Alibaba Qwen Weibo account announced the launch of Qwen3-Coder-Next, an open-weight language model specifically designed for coding agents and local development."
   ]
  },
  {
   "url": "https://baike.baidu.com/en/item/Qwen/1530291",
   "title": "Qwen",
   "published_at": null,
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-02-16",
   "access_type": null,
   "license": null,
   "significance": "new generation of its Qwen large model, Qwen3.5, which features innovations in its model architecture",
   "organization_evidence": [
    "Development Timeline",
    "On the evening of February 16, 2026, Alibaba open-sourced the new generation of its Qwen large model, Qwen3.5, which features innovations in its model architecture."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Development Timeline",
    "On the evening of February 16, 2026, Alibaba open-sourced the new generation of its Qwen large model, Qwen3.5, which features innovations in its model architecture."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Development Timeline",
    "On the evening of February 16, 2026, Alibaba open-sourced the new generation of its Qwen large model, Qwen3.5, which features innovations in its model architecture."
   ]
  },
  {
   "url": "https://baike.baidu.com/en/item/Qwen/1530291",
   "title": "Qwen",
   "published_at": null,
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-02-16",
   "access_type": null,
   "license": null,
   "significance": "performance rivals that of Gemini 3 Pro and it has ascended to become the world's most powerful open-source model",
   "organization_evidence": [
    "Development Timeline",
    "On February 16, 2026, Alibaba announced the open-sourcing of the all-new Qwen3.5-Plus large model, claiming its performance rivals that of Gemini 3 Pro and it has ascended to become the world's most powerful open-source model."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Development Timeline",
    "On February 16, 2026, Alibaba announced the open-sourcing of the all-new Qwen3.5-Plus large model, claiming its performance rivals that of Gemini 3 Pro and it has ascended to become the world's most powerful open-source model."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Development Timeline",
    "On February 16, 2026, Alibaba announced the open-sourcing of the all-new Qwen3.5-Plus large model, claiming its performance rivals that of Gemini 3 Pro and it has ascended to become the world's most powerful open-source model."
   ]
  },
  {
   "url": "https://baike.baidu.com/en/item/Qwen/1530291",
   "title": "Qwen",
   "published_at": null,
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-08-13",
   "access_type": null,
   "license": null,
   "significance": "ultra-large-scale Mixture of Experts model",
   "organization_evidence": [
    "Development Timeline",
    "On August 13, it was announced that Alibaba had open-sourced the ultra-large-scale Mixture of Experts model Qwen3.8-2.4T-A95B , and the Zhongzhi FlagOS Community simultaneously completed multi-chip adaptation for it."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Development Timeline",
    "On August 13, it was announced that Alibaba had open-sourced the ultra-large-scale Mixture of Experts model Qwen3.8-2.4T-A95B , and the Zhongzhi FlagOS Community simultaneously completed multi-chip adaptation for it."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Development Timeline",
    "On August 13, it was announced that Alibaba had open-sourced the ultra-large-scale Mixture of Experts model Qwen3.8-2.4T-A95B , and the Zhongzhi FlagOS Community simultaneously completed multi-chip adaptation for it."
   ]
  },
  {
   "url": "https://reasoncore.dev/post/thinking-machines-releases-inkling-a-975b-open-weights-moe-model-with-1m-context",
   "title": "Thinking Machines releases Inkling: a 975B open-weights MoE model with 1M context",
   "published_at": "2026-07-16T03:16:36",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "A 975 billion-parameter Mixture-of-Experts transformer with 41 billion active parameters, 1,000,000-token context window, and multimodal capabilities.",
   "organization_evidence": [
    "Thinking Machines Lab released [Inkling](https://thinkingmachines.ai/news/introducing-inkling/) on July 15, 2026 as an open-weights, multimodal foundation model with full weights available on Hugging Face and selectable for fine-tuning via the company's Tinker service."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Thinking Machines Lab released [Inkling](https://thinkingmachines.ai/news/introducing-inkling/) on July 15, 2026 as an open-weights, multimodal foundation model with full weights available on Hugging Face and selectable for fine-tuning via the company's Tinker service."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Inkling is a 975 billion-parameter Mixture-of-Experts ([MoE](https://reasoncore.dev/glossary/moe)) transformer with 41 billion active parameters and a 1,000,000-token [context window](https://reasoncore.dev/glossary/context-window), released under an Apache 2.0 license that explicitly permits redistribution and downstream fine-tuning."
   ]
  },
  {
   "url": "https://economicpolicytimes.com/article/928322887-poolside-releases-laguna-s-2-1-the-west-s-most-capable-open-weight-model",
   "title": "Poolside releases Laguna S 2.1, the West\u2019s most capable open-weight model",
   "published_at": null,
   "organization": "Poolside",
   "model": "Laguna S 2.1",
   "release_date": "2026-07-21",
   "access_type": "open-weight",
   "license": "OpenMDW-1.1",
   "significance": "118-billion-parameter model matches models several times its size on agentic coding and is small enough to run on a single desktop",
   "organization_evidence": [
    "# Poolside releases Laguna S 2.1, the West\u2019s most capable open-weight model",
    "SAN FRANCISCO, July 21, 2026 (GLOBE NEWSWIRE) -- [Poolside today released **Laguna S 2.1**](https://www.globenewswire.com/Tracker?data=_qXFO0siMuP0VpmUSf1ESZv-4mfFJrh7XDdEB0XAraXy_b0m6LmOrEMkD_avRl3jWp7VHZdLi2Ob3fK2YnsiAIHDPsNdweqMtxwEaPG1uOUy5zsRlI7EvyUYlzlsZB_4OTGkph95ZxSE9MFyGBj8eg==), a 118-billion-parameter open-weight foundation model built for agentic coding."
   ],
   "model_evidence": [
    "# Poolside releases Laguna S 2.1, the West\u2019s most capable open-weight model",
    "SAN FRANCISCO, July 21, 2026 (GLOBE NEWSWIRE) -- [Poolside today released **Laguna S 2.1**](https://www.globenewswire.com/Tracker?data=_qXFO0siMuP0VpmUSf1ESZv-4mfFJrh7XDdEB0XAraXy_b0m6LmOrEMkD_avRl3jWp7VHZdLi2Ob3fK2YnsiAIHDPsNdweqMtxwEaPG1uOUy5zsRlI7EvyUYlzlsZB_4OTGkph95ZxSE9MFyGBj8eg==), a 118-billion-parameter open-weight foundation model built for agentic coding."
   ],
   "release_date_evidence": [
    "SAN FRANCISCO, July 21, 2026 (GLOBE NEWSWIRE) -- [Poolside today released **Laguna S 2.1**](https://www.globenewswire.com/Tracker?data=_qXFO0siMuP0VpmUSf1ESZv-4mfFJrh7XDdEB0XAraXy_b0m6LmOrEMkD_avRl3jWp7VHZdLi2Ob3fK2YnsiAIHDPsNdweqMtxwEaPG1uOUy5zsRlI7EvyUYlzlsZB_4OTGkph95ZxSE9MFyGBj8eg==), a 118-billion-parameter open-weight foundation model built for agentic coding.",
    "# Poolside releases Laguna S 2.1, the West\u2019s most capable open-weight model"
   ],
   "access_type_evidence": [
    "# Poolside releases Laguna S 2.1, the West\u2019s most capable open-weight model",
    "SAN FRANCISCO, July 21, 2026 (GLOBE NEWSWIRE) -- [Poolside today released **Laguna S 2.1**](https://www.globenewswire.com/Tracker?data=_qXFO0siMuP0VpmUSf1ESZv-4mfFJrh7XDdEB0XAraXy_b0m6LmOrEMkD_avRl3jWp7VHZdLi2Ob3fK2YnsiAIHDPsNdweqMtxwEaPG1uOUy5zsRlI7EvyUYlzlsZB_4OTGkph95ZxSE9MFyGBj8eg==), a 118-billion-parameter open-weight foundation model built for agentic coding."
   ],
   "license_evidence": [
    "It is available today under an OpenMDW-1.1 license.",
    "# Poolside releases Laguna S 2.1, the West\u2019s most capable open-weight model"
   ],
   "significance_evidence": [
    "The 118-billion-parameter model matches models several times its size on agentic coding and is small enough to run on a single desktop.",
    "# Poolside releases Laguna S 2.1, the West\u2019s most capable open-weight model"
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/gemma-4-26b-a4b-vs-claude-sonnet-4-6-non-reasoning-low-effort",
   "title": "Gemma 4 26B A4B (Reasoning) vs Claude Sonnet 4.6 (Non-reasoning, Low Effort): Model Comparison",
   "published_at": null,
   "organization": "Google",
   "model": "Gemma 4 26B A4B (Reasoning)",
   "release_date": "2026-04",
   "access_type": "open source",
   "license": null,
   "significance": "Reasoning model with open source weights and 256k context window",
   "organization_evidence": [
    "# Gemma 4 26B A4B (Reasoning) vs. Claude Sonnet 4.6 (Non-reasoning, Low Effort)",
    "## Model Comparison",
    "| Metric | ![Google logo](/img/logos/google_small.svg)Gemma 4 26B A4B (Reasoning) | ![Anthropic logo](/img/logos/anthropic_small.svg)Claude Sonnet 4.6 (Non-reasoning, Low Effort) | Analysis |",
    "| Creator | ![Google](/img/logos/google_small.svg)[](https://deepmind.google/technologies/gemini/) | ![Anthropic](/img/logos/anthropic_small.svg)[](https://www.anthropic.com/) | |"
   ],
   "model_evidence": [
    "# Gemma 4 26B A4B (Reasoning) vs. Claude Sonnet 4.6 (Non-reasoning, Low Effort)",
    "## Model Comparison",
    "| Metric | ![Google logo](/img/logos/google_small.svg)Gemma 4 26B A4B (Reasoning) | ![Anthropic logo](/img/logos/anthropic_small.svg)Claude Sonnet 4.6 (Non-reasoning, Low Effort) | Analysis |"
   ],
   "release_date_evidence": [
    "# Gemma 4 26B A4B (Reasoning) vs. Claude Sonnet 4.6 (Non-reasoning, Low Effort)",
    "## Model Comparison",
    "| Metric | ![Google logo](/img/logos/google_small.svg)Gemma 4 26B A4B (Reasoning) | ![Anthropic logo](/img/logos/anthropic_small.svg)Claude Sonnet 4.6 (Non-reasoning, Low Effort) | Analysis |",
    "| Release Date | April, 2026 | February, 2026 | Gemma 4 26B A4B (Reasoning) has a more recent release date than Claude Sonnet 4.6 (Non-reasoning, Low Effort) |"
   ],
   "access_type_evidence": [
    "# Gemma 4 26B A4B (Reasoning) vs. Claude Sonnet 4.6 (Non-reasoning, Low Effort)",
    "## Model Comparison",
    "| Metric | ![Google logo](/img/logos/google_small.svg)Gemma 4 26B A4B (Reasoning) | ![Anthropic logo](/img/logos/anthropic_small.svg)Claude Sonnet 4.6 (Non-reasoning, Low Effort) | Analysis |",
    "| Open Source (Weights) | Yes[](https://huggingface.co/google/gemma-4-26B-A4B-it) | No | Gemma 4 26B A4B (Reasoning) is open source while Claude Sonnet 4.6 (Non-reasoning, Low Effort) is proprietary |"
   ],
   "license_evidence": null,
   "significance_evidence": [
    "# Gemma 4 26B A4B (Reasoning) vs. Claude Sonnet 4.6 (Non-reasoning, Low Effort)",
    "Comparison between Gemma 4 26B A4B (Reasoning) and Claude Sonnet 4.6 (Non-reasoning, Low Effort) across intelligence, price, speed, context window and more.",
    "## Model Comparison",
    "| Metric | ![Google logo](/img/logos/google_small.svg)Gemma 4 26B A4B (Reasoning) | ![Anthropic logo](/img/logos/anthropic_small.svg)Claude Sonnet 4.6 (Non-reasoning, Low Effort) | Analysis |",
    "| Context Window | 256k tokens (~384 A4 pages of size 12 Arial font) | 1000k tokens (~1500 A4 pages of size 12 Arial font) | Gemma 4 26B A4B (Reasoning) is smaller than Claude Sonnet 4.6 (Non-reasoning, Low Effort) |",
    "| Release Date | April, 2026 | February, 2026 | Gemma 4 26B A4B (Reasoning) has a more recent release date than Claude Sonnet 4.6 (Non-reasoning, Low Effort) |",
    "| Open Source (Weights) | Yes[](https://huggingface.co/google/gemma-4-26B-A4B-it) | No | Gemma 4 26B A4B (Reasoning) is open source while Claude Sonnet 4.6 (Non-reasoning, Low Effort) is proprietary |"
   ]
  },
  {
   "url": "https://coderfacts.com/market/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is-2",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - Coder Facts",
   "published_at": "2026-07-15T13:23:54",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor",
   "organization_evidence": [
    "Chinese labs such as DeepSeek, Z.ai, Moonshot, and Alibaba have each established distinct strategic positions.",
    "The releases include DeepSeek V4, MiniMax M3, Kimi K2.7-Code, and GLM-5.2, all available for download and most under permissive licenses.",
    "Four Frontier-Class Open Models in Eight Weeks"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "From late April to mid-June 2026, Chinese labs introduced four major open-weight models: DeepSeek V4 on April 24, MiniMax M3 on June 1, and Kimi K2.7-Code along with GLM-5.2 in mid-June.",
    "Four Frontier-Class Open Models in Eight Weeks"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor",
    "Four Frontier-Class Open Models in Eight Weeks"
   ]
  },
  {
   "url": "https://coderfacts.com/market/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is-2",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - Coder Facts",
   "published_at": "2026-07-15T13:23:54",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "cheap 1M-token context, native multimodal, modified-MIT",
   "organization_evidence": [
    "The releases include DeepSeek V4, MiniMax M3, Kimi K2.7-Code, and GLM-5.2, all available for download and most under permissive licenses.",
    "Four Frontier-Class Open Models in Eight Weeks"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "From late April to mid-June 2026, Chinese labs introduced four major open-weight models: DeepSeek V4 on April 24, MiniMax M3 on June 1, and Kimi K2.7-Code along with GLM-5.2 in mid-June.",
    "Four Frontier-Class Open Models in Eight Weeks"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks"
   ]
  },
  {
   "url": "https://coderfacts.com/market/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is-2",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - Coder Facts",
   "published_at": "2026-07-15T13:23:54",
   "organization": "Moonshot",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "agent-run specialist, ~30% fewer thinking tokens than K2.6",
   "organization_evidence": [
    "Chinese labs such as DeepSeek, Z.ai, Moonshot, and Alibaba have each established distinct strategic positions.",
    "The releases include DeepSeek V4, MiniMax M3, Kimi K2.7-Code, and GLM-5.2, all available for download and most under permissive licenses.",
    "Four Frontier-Class Open Models in Eight Weeks"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "From late April to mid-June 2026, Chinese labs introduced four major open-weight models: DeepSeek V4 on April 24, MiniMax M3 on June 1, and Kimi K2.7-Code along with GLM-5.2 in mid-June.",
    "Four Frontier-Class Open Models in Eight Weeks"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks"
   ]
  },
  {
   "url": "https://coderfacts.com/market/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is-2",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - Coder Facts",
   "published_at": "2026-07-15T13:23:54",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "753B MoE, MIT, top open-weight on Artificial Analysis index",
   "organization_evidence": [
    "Chinese labs such as DeepSeek, Z.ai, Moonshot, and Alibaba have each established distinct strategic positions.",
    "The releases include DeepSeek V4, MiniMax M3, Kimi K2.7-Code, and GLM-5.2, all available for download and most under permissive licenses.",
    "Four Frontier-Class Open Models in Eight Weeks"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "From late April to mid-June 2026, Chinese labs introduced four major open-weight models: DeepSeek V4 on April 24, MiniMax M3 on June 1, and Kimi K2.7-Code along with GLM-5.2 in mid-June.",
    "Four Frontier-Class Open Models in Eight Weeks"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks"
   ]
  },
  {
   "url": "https://medium.com/@gautsoni/the-new-generation-of-open-reasoning-models-gemma-4-and-qwen3-5-100b9d292748",
   "title": "The New Generation of Open Reasoning Models: Gemma 4 and Qwen3.5",
   "published_at": "2026-06-11T18:08:51",
   "organization": "Google DeepMind",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "reasoning models from the ground up with hybrid attention, sparse mixture-of-experts routing, and native multimodality",
   "organization_evidence": [
    "# The New Generation of Open Reasoning Models: Gemma 4 and Qwen3.5",
    "In the first half of 2026, that arrangement broke down.",
    "Google\u2019s **Gemma 4** (April 2026) and Alibaba\u2019s **Qwen3.5** (February 2026) are open-weight families that were designed as *reasoning models from the ground up* \u2014 and the smaller members of both families run comfortably on a single consumer GPU.",
    "Gemma 4 (Google DeepMind)",
    "Gemma 4 is Google\u2019s open-weight family built from the same research line as Gemini 3, released under Apache 2.0."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# The New Generation of Open Reasoning Models: Gemma 4 and Qwen3.5",
    "In the first half of 2026, that arrangement broke down.",
    "Google\u2019s **Gemma 4** (April 2026) and Alibaba\u2019s **Qwen3.5** (February 2026) are open-weight families that were designed as *reasoning models from the ground up* \u2014 and the smaller members of both families run comfortably on a single consumer GPU."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# The New Generation of Open Reasoning Models: Gemma 4 and Qwen3.5",
    "Google\u2019s **Gemma 4** (April 2026) and Alibaba\u2019s **Qwen3.5** (February 2026) are open-weight families that were designed as *reasoning models from the ground up* \u2014 and the smaller members of both families run comfortably on a single consumer GPU.",
    "This isn\u2019t an incremental refresh.",
    "Both families rethink the architecture underneath: hybrid attention, sparse mixture-of-experts routing, native multimodality, and reasoning modes you can dial up or switch off per request."
   ]
  },
  {
   "url": "https://medium.com/@gautsoni/the-new-generation-of-open-reasoning-models-gemma-4-and-qwen3-5-100b9d292748",
   "title": "The New Generation of Open Reasoning Models: Gemma 4 and Qwen3.5",
   "published_at": "2026-06-11T18:08:51",
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-02",
   "access_type": null,
   "license": null,
   "significance": "extreme sparsity with 397 billion total parameters and 17 billion active per forward pass, linear attention via Gated DeltaNet, and native multimodality",
   "organization_evidence": [
    "# The New Generation of Open Reasoning Models: Gemma 4 and Qwen3.5",
    "In the first half of 2026, that arrangement broke down.",
    "Google\u2019s **Gemma 4** (April 2026) and Alibaba\u2019s **Qwen3.5** (February 2026) are open-weight families that were designed as *reasoning models from the ground up* \u2014 and the smaller members of both families run comfortably on a single consumer GPU.",
    "Qwen3.5 (Alibaba)",
    "Qwen3.5 arrived in three waves: the flagship **Qwen3.5\u2013397B-A17B** in mid-February, mid-range models (122B-A10B, 35B-A3B, 27B dense) the following week, and small models (9B, 4B, 2B, 0.8B) in early March."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# The New Generation of Open Reasoning Models: Gemma 4 and Qwen3.5",
    "In the first half of 2026, that arrangement broke down.",
    "Google\u2019s **Gemma 4** (April 2026) and Alibaba\u2019s **Qwen3.5** (February 2026) are open-weight families that were designed as *reasoning models from the ground up* \u2014 and the smaller members of both families run comfortably on a single consumer GPU."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# The New Generation of Open Reasoning Models: Gemma 4 and Qwen3.5",
    "Google\u2019s **Gemma 4** (April 2026) and Alibaba\u2019s **Qwen3.5** (February 2026) are open-weight families that were designed as *reasoning models from the ground up* \u2014 and the smaller members of both families run comfortably on a single consumer GPU.",
    "The flagship\u2019s headline trick is extreme sparsity: 397 billion total parameters, but only **17 billion active per forward pass**.",
    "Under the hood, the family combines **linear attention via Gated DeltaNet** with sparse MoE \u2014 a genuinely different attention mechanism from the standard transformer stack, which pays off in KV-cache efficiency at long context."
   ]
  },
  {
   "url": "https://swiftwand.com/en/ai-model-releases-july-2026-timeline-en",
   "title": "AI Model Releases of July 2026: the 19-Day Timeline Explained",
   "published_at": "2026-08-09T21:00:00",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-27",
   "access_type": null,
   "license": null,
   "significance": "2.8 trillion total parameters in a mixture-of-experts design with 104 billion active per token, a context window of 1,048,576 tokens, and a download of about 1.56 terabytes",
   "organization_evidence": [
    "# July 2026: The 19 Days When Every Major AI Lab Shipped \u2014 a Timeline of What Actually Happened",
    "| July 27 | Moonshot AI publishes the Kimi K3 weights |",
    "Moonshot AI announced Kimi K3 on July 16 and published the weights on Hugging Face eleven days later, on July 27 \u2014 two separate dates that are easy to conflate."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# July 2026: The 19 Days When Every Major AI Lab Shipped \u2014 a Timeline of What Actually Happened",
    "| July 27 | Moonshot AI publishes the Kimi K3 weights |",
    "Moonshot AI announced Kimi K3 on July 16 and published the weights on Hugging Face eleven days later, on July 27 \u2014 two separate dates that are easy to conflate."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Moonshot: an announcement, then the weights",
    "The headline numbers: 2.8 trillion total parameters in a mixture-of-experts design with 104 billion active per token, a context window of 1,048,576 tokens, and a download of about 1.56 terabytes."
   ]
  },
  {
   "url": "https://github.com/xigh/open-weight-models",
   "title": "GitHub - xigh/open-weight-models: Curated list of open-weight AI models with commercially exploitable licenses, verified benchmarks, and no EU restrictions.",
   "published_at": null,
   "organization": "Google",
   "model": null,
   "release_date": "2026-06",
   "access_type": null,
   "license": null,
   "significance": "encoder-free multimodal, AIME 2026 77.5 / LCB v6 72.0",
   "organization_evidence": [
    "Open Weight Models",
    "A curated list of open-weight AI models with commercially exploitable licenses, verified benchmarks, and no geographic restrictions. Built to decide which models to support in [herbert-rs](https://github.com/xigh/herbert-rs), a local LLM inference engine in Rust and hand-written assembly.",
    "| [Gemma 4 12B Unified](https://huggingface.co/google/gemma-4-12B-it) | Google | 11.95B | 11.95B | Dense **encoder-free** multimodal | 256K | Apache 2.0 | GPQA 78.8, MMLU-Pro 77.2, AIME 2026 77.5, LCB v6 72.0, **audio natif** (CoVoST 38.5) |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Open Weight Models",
    "A curated list of open-weight AI models with commercially exploitable licenses, verified benchmarks, and no geographic restrictions. Built to decide which models to support in [herbert-rs](https://github.com/xigh/herbert-rs), a local LLM inference engine in Rust and hand-written assembly.",
    "**v4 additions (June 2026)** \u2014 Generalists: Gemma 4 12B Unified (encoder-free multimodal, AIME 2026 77.5 / LCB v6 72.0)."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Open Weight Models",
    "A curated list of open-weight AI models with commercially exploitable licenses, verified benchmarks, and no geographic restrictions. Built to decide which models to support in [herbert-rs](https://github.com/xigh/herbert-rs), a local LLM inference engine in Rust and hand-written assembly.",
    "**v4 additions (June 2026)** \u2014 Generalists: Gemma 4 12B Unified (encoder-free multimodal, AIME 2026 77.5 / LCB v6 72.0)."
   ]
  },
  {
   "url": "https://github.com/xigh/open-weight-models",
   "title": "GitHub - xigh/open-weight-models: Curated list of open-weight AI models with commercially exploitable licenses, verified benchmarks, and no EU restrictions.",
   "published_at": null,
   "organization": "JetBrains",
   "model": null,
   "release_date": "2026-05",
   "access_type": null,
   "license": null,
   "significance": "LCB v6 69.9 with 2.5B active",
   "organization_evidence": [
    "Open Weight Models",
    "A curated list of open-weight AI models with commercially exploitable licenses, verified benchmarks, and no geographic restrictions. Built to decide which models to support in [herbert-rs](https://github.com/xigh/herbert-rs), a local LLM inference engine in Rust and hand-written assembly.",
    "**v4 additions (June 2026)** \u2014 Generalists: Gemma 4 12B Unified (encoder-free multimodal, AIME 2026 77.5 / LCB v6 72.0). Code (LCB/Terminal-Bench): Mellum 2 12B-A2.5B (JetBrains, LCB v6 69.9 with 2.5B active)."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Open Weight Models",
    "A curated list of open-weight AI models with commercially exploitable licenses, verified benchmarks, and no geographic restrictions. Built to decide which models to support in [herbert-rs](https://github.com/xigh/herbert-rs), a local LLM inference engine in Rust and hand-written assembly.",
    "**Mellum 2** (JetBrains, May 2026) is the second generation after Mellum 1 (4B dense, 2024) \u2014 this time a **MoE 12B/A2.5B** post-trained in SFT + **RLVR** (Reinforcement Learning with Verifiable Rewards)."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Open Weight Models",
    "A curated list of open-weight AI models with commercially exploitable licenses, verified benchmarks, and no geographic restrictions. Built to decide which models to support in [herbert-rs](https://github.com/xigh/herbert-rs), a local LLM inference engine in Rust and hand-written assembly.",
    "**v4 additions (June 2026)** \u2014 Generalists: Gemma 4 12B Unified (encoder-free multimodal, AIME 2026 77.5 / LCB v6 72.0). Code (LCB/Terminal-Bench): Mellum 2 12B-A2.5B (JetBrains, LCB v6 69.9 with 2.5B active)."
   ]
  },
  {
   "url": "https://github.com/xigh/open-weight-models",
   "title": "GitHub - xigh/open-weight-models: Curated list of open-weight AI models with commercially exploitable licenses, verified benchmarks, and no EU restrictions.",
   "published_at": null,
   "organization": "ByteDance",
   "model": null,
   "release_date": "2026-06",
   "access_type": null,
   "license": null,
   "significance": "unified any-to-any 3B, gen + understanding + editing",
   "organization_evidence": [
    "Open Weight Models",
    "A curated list of open-weight AI models with commercially exploitable licenses, verified benchmarks, and no geographic restrictions. Built to decide which models to support in [herbert-rs](https://github.com/xigh/herbert-rs), a local LLM inference engine in Rust and hand-written assembly."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Open Weight Models",
    "A curated list of open-weight AI models with commercially exploitable licenses, verified benchmarks, and no geographic restrictions. Built to decide which models to support in [herbert-rs](https://github.com/xigh/herbert-rs), a local LLM inference engine in Rust and hand-written assembly."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Open Weight Models",
    "A curated list of open-weight AI models with commercially exploitable licenses, verified benchmarks, and no geographic restrictions. Built to decide which models to support in [herbert-rs](https://github.com/xigh/herbert-rs), a local LLM inference engine in Rust and hand-written assembly."
   ]
  },
  {
   "url": "https://github.com/xigh/open-weight-models",
   "title": "GitHub - xigh/open-weight-models: Curated list of open-weight AI models with commercially exploitable licenses, verified benchmarks, and no EU restrictions.",
   "published_at": null,
   "organization": "Sapient",
   "model": null,
   "release_date": "2026-05",
   "access_type": null,
   "license": null,
   "significance": "Hierarchical Reasoning Model, recurrent dual-timescale",
   "organization_evidence": [
    "Open Weight Models",
    "A curated list of open-weight AI models with commercially exploitable licenses, verified benchmarks, and no geographic restrictions. Built to decide which models to support in [herbert-rs](https://github.com/xigh/herbert-rs), a local LLM inference engine in Rust and hand-written assembly."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Open Weight Models",
    "A curated list of open-weight AI models with commercially exploitable licenses, verified benchmarks, and no geographic restrictions. Built to decide which models to support in [herbert-rs](https://github.com/xigh/herbert-rs), a local LLM inference engine in Rust and hand-written assembly.",
    "**HRM-Text-1B** (Sapient Intelligence, May 2026, [arXiv:2605.20613](https://arxiv.org/abs/2605.20613)) is the first language model built on the **Hierarchical Reasoning Model** architecture \u2014 a dual-timescale recurrent design that loops H (high-level/slow) and L (low-level/fast) cycles, similar in spirit to URM's recursive depth."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Open Weight Models",
    "A curated list of open-weight AI models with commercially exploitable licenses, verified benchmarks, and no geographic restrictions. Built to decide which models to support in [herbert-rs](https://github.com/xigh/herbert-rs), a local LLM inference engine in Rust and hand-written assembly."
   ]
  },
  {
   "url": "https://github.com/xigh/open-weight-models",
   "title": "GitHub - xigh/open-weight-models: Curated list of open-weight AI models with commercially exploitable licenses, verified benchmarks, and no EU restrictions.",
   "published_at": null,
   "organization": "Qwen",
   "model": null,
   "release_date": "2026-05",
   "access_type": null,
   "license": null,
   "significance": "predicts next page state, beats GPT-4o on Factuality",
   "organization_evidence": [
    "Open Weight Models",
    "A curated list of open-weight AI models with commercially exploitable licenses, verified benchmarks, and no geographic restrictions. Built to decide which models to support in [herbert-rs](https://github.com/xigh/herbert-rs), a local LLM inference engine in Rust and hand-written assembly."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Open Weight Models",
    "A curated list of open-weight AI models with commercially exploitable licenses, verified benchmarks, and no geographic restrictions. Built to decide which models to support in [herbert-rs](https://github.com/xigh/herbert-rs), a local LLM inference engine in Rust and hand-written assembly.",
    "**WebWorld** (Qwen Team, May 2026, [arXiv:2602.14721](https://arxiv.org/abs/2602.14721)) is the first serious open-source web world model."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Open Weight Models",
    "A curated list of open-weight AI models with commercially exploitable licenses, verified benchmarks, and no geographic restrictions. Built to decide which models to support in [herbert-rs](https://github.com/xigh/herbert-rs), a local LLM inference engine in Rust and hand-written assembly."
   ]
  },
  {
   "url": "https://github.com/xigh/open-weight-models",
   "title": "GitHub - xigh/open-weight-models: Curated list of open-weight AI models with commercially exploitable licenses, verified benchmarks, and no EU restrictions.",
   "published_at": null,
   "organization": "Ant",
   "model": null,
   "release_date": "2026-05",
   "access_type": null,
   "license": null,
   "significance": "Token-efficient agent (~15M tokens on full AA suite vs 40-100M for long-reasoners)",
   "organization_evidence": [
    "Open Weight Models",
    "A curated list of open-weight AI models with commercially exploitable licenses, verified benchmarks, and no geographic restrictions. Built to decide which models to support in [herbert-rs](https://github.com/xigh/herbert-rs), a local LLM inference engine in Rust and hand-written assembly."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Open Weight Models",
    "A curated list of open-weight AI models with commercially exploitable licenses, verified benchmarks, and no geographic restrictions. Built to decide which models to support in [herbert-rs](https://github.com/xigh/herbert-rs), a local LLM inference engine in Rust and hand-written assembly."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Open Weight Models",
    "A curated list of open-weight AI models with commercially exploitable licenses, verified benchmarks, and no geographic restrictions. Built to decide which models to support in [herbert-rs](https://github.com/xigh/herbert-rs), a local LLM inference engine in Rust and hand-written assembly.",
    "| [Ling-2.6-flash](https://huggingface.co/inclusionAI/Ling-2.6-flash) | Ant Group | 7.4B | 104B | MoE (hybrid linear attn 1:7 MLA+Lightning) | 262K | MIT | Token-efficient agent (~15M tokens on full AA suite vs 40-100M for long-reasoners) |"
   ]
  },
  {
   "url": "https://github.com/xigh/open-weight-models",
   "title": "GitHub - xigh/open-weight-models: Curated list of open-weight AI models with commercially exploitable licenses, verified benchmarks, and no EU restrictions.",
   "published_at": null,
   "organization": "Mistral AI",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "First Mistral merged flagship: Medium 3.1 + Magistral + Devstral 2 unified, configurable reasoning_effort",
   "organization_evidence": [
    "Open Weight Models",
    "A curated list of open-weight AI models with commercially exploitable licenses, verified benchmarks, and no geographic restrictions. Built to decide which models to support in [herbert-rs](https://github.com/xigh/herbert-rs), a local LLM inference engine in Rust and hand-written assembly.",
    "| [Mistral Medium 3.5 128B](https://huggingface.co/mistralai/Mistral-Medium-3.5-128B) | Mistral AI | 128B | 128B | Dense + Pixtral vision | 256K | \ud83d\udd34 **Modified MIT** (revenue cap) | First Mistral merged flagship: Medium 3.1 + Magistral + Devstral 2 unified, configurable `reasoning_effort` |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Open Weight Models",
    "A curated list of open-weight AI models with commercially exploitable licenses, verified benchmarks, and no geographic restrictions. Built to decide which models to support in [herbert-rs](https://github.com/xigh/herbert-rs), a local LLM inference engine in Rust and hand-written assembly.",
    "**Mistral Medium 3.5** (Apr/May 2026) is the **first merged flagship** from Mistral: a single set of weights unifying what used to be three distinct models \u2014 Medium 3.1 (instruct), Magistral (reasoning), Devstral 2 (coding agent)."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Open Weight Models",
    "A curated list of open-weight AI models with commercially exploitable licenses, verified benchmarks, and no geographic restrictions. Built to decide which models to support in [herbert-rs](https://github.com/xigh/herbert-rs), a local LLM inference engine in Rust and hand-written assembly.",
    "| [Mistral Medium 3.5 128B](https://huggingface.co/mistralai/Mistral-Medium-3.5-128B) | Mistral AI | 128B | 128B | Dense + Pixtral vision | 256K | \ud83d\udd34 **Modified MIT** (revenue cap) | First Mistral merged flagship: Medium 3.1 + Magistral + Devstral 2 unified, configurable `reasoning_effort` |"
   ]
  },
  {
   "url": "https://github.com/xigh/open-weight-models",
   "title": "GitHub - xigh/open-weight-models: Curated list of open-weight AI models with commercially exploitable licenses, verified benchmarks, and no EU restrictions.",
   "published_at": null,
   "organization": "MiniMax AI",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "Agentic workflows alt to Claude Opus 4.6 / GPT-5.3-Codex, IQ1_M @ 60.7 GB",
   "organization_evidence": [
    "Open Weight Models",
    "A curated list of open-weight AI models with commercially exploitable licenses, verified benchmarks, and no geographic restrictions. Built to decide which models to support in [herbert-rs](https://github.com/xigh/herbert-rs), a local LLM inference engine in Rust and hand-written assembly.",
    "| [MiniMax M2.7](https://huggingface.co/MiniMaxAI/MiniMax-M2.7) | MiniMax AI | 10B | 230B | MoE (256 experts, 8 active, 4.3% ratio) | ~200K | MIT *(verify on HF)* | Agentic workflows alt to Claude Opus 4.6 / GPT-5.3-Codex, IQ1_M @ 60.7 GB |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Open Weight Models",
    "A curated list of open-weight AI models with commercially exploitable licenses, verified benchmarks, and no geographic restrictions. Built to decide which models to support in [herbert-rs](https://github.com/xigh/herbert-rs), a local LLM inference engine in Rust and hand-written assembly.",
    "**MiniMax M2.7** (Apr 2026 open-weight release) pushes the **active/total ratio** to 4.3% (10B/230B), targeting agentic long-running workflows (coding, multi-step troubleshooting, document editing)."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Open Weight Models",
    "A curated list of open-weight AI models with commercially exploitable licenses, verified benchmarks, and no geographic restrictions. Built to decide which models to support in [herbert-rs](https://github.com/xigh/herbert-rs), a local LLM inference engine in Rust and hand-written assembly.",
    "| [MiniMax M2.7](https://huggingface.co/MiniMaxAI/MiniMax-M2.7) | MiniMax AI | 10B | 230B | MoE (256 experts, 8 active, 4.3% ratio) | ~200K | MIT *(verify on HF)* | Agentic workflows alt to Claude Opus 4.6 / GPT-5.3-Codex, IQ1_M @ 60.7 GB |"
   ]
  },
  {
   "url": "https://github.com/xigh/open-weight-models",
   "title": "GitHub - xigh/open-weight-models: Curated list of open-weight AI models with commercially exploitable licenses, verified benchmarks, and no EU restrictions.",
   "published_at": null,
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-05",
   "access_type": null,
   "license": null,
   "significance": "First < 200B-active LLM with native 1M ctx, FP4+FP8 mixed, 32T pre-train, 27% FLOPs / 10% KV cache vs V3.2",
   "organization_evidence": [
    "Open Weight Models",
    "A curated list of open-weight AI models with commercially exploitable licenses, verified benchmarks, and no geographic restrictions. Built to decide which models to support in [herbert-rs](https://github.com/xigh/herbert-rs), a local LLM inference engine in Rust and hand-written assembly.",
    "| [DeepSeek-V4-Flash](https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash) | DeepSeek | 13B | 284B | MoE (hybrid CSA+HCA, mHC, Muon optimizer) | **1M** native | MIT | First < 200B-active LLM with native 1M ctx, FP4+FP8 mixed, 32T pre-train, 27% FLOPs / 10% KV cache vs V3.2 |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Open Weight Models",
    "A curated list of open-weight AI models with commercially exploitable licenses, verified benchmarks, and no geographic restrictions. Built to decide which models to support in [herbert-rs](https://github.com/xigh/herbert-rs), a local LLM inference engine in Rust and hand-written assembly.",
    "**DeepSeek-V4-Flash** (May 2026) is the small sibling of V4-Pro (1.6T/49B)."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Open Weight Models",
    "A curated list of open-weight AI models with commercially exploitable licenses, verified benchmarks, and no geographic restrictions. Built to decide which models to support in [herbert-rs](https://github.com/xigh/herbert-rs), a local LLM inference engine in Rust and hand-written assembly.",
    "| [DeepSeek-V4-Flash](https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash) | DeepSeek | 13B | 284B | MoE (hybrid CSA+HCA, mHC, Muon optimizer) | **1M** native | MIT | First < 200B-active LLM with native 1M ctx, FP4+FP8 mixed, 32T pre-train, 27% FLOPs / 10% KV cache vs V3.2 |"
   ]
  },
  {
   "url": "https://github.com/xigh/open-weight-models",
   "title": "GitHub - xigh/open-weight-models: Curated list of open-weight AI models with commercially exploitable licenses, verified benchmarks, and no EU restrictions.",
   "published_at": null,
   "organization": "Zyphra",
   "model": null,
   "release_date": "2026-05",
   "access_type": null,
   "license": null,
   "significance": "On-device deployable, test-time-compute friendly, 128K ctx",
   "organization_evidence": [
    "Open Weight Models",
    "A curated list of open-weight AI models with commercially exploitable licenses, verified benchmarks, and no geographic restrictions. Built to decide which models to support in [herbert-rs](https://github.com/xigh/herbert-rs), a local LLM inference engine in Rust and hand-written assembly."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Open Weight Models",
    "A curated list of open-weight AI models with commercially exploitable licenses, verified benchmarks, and no geographic restrictions. Built to decide which models to support in [herbert-rs](https://github.com/xigh/herbert-rs), a local LLM inference engine in Rust and hand-written assembly.",
    "**ZAYA1-8B** (Zyphra, May 2026) is a Zamba-2 successor: 80 layers mixing SSM-Mamba and attention, with **Compressed Cross Attention (CCA)** plus **Mixture-of-Depths (MoD)** and **Expert Decision Attention (EDA)** on top of 16 top-1 experts."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Open Weight Models",
    "A curated list of open-weight AI models with commercially exploitable licenses, verified benchmarks, and no geographic restrictions. Built to decide which models to support in [herbert-rs](https://github.com/xigh/herbert-rs), a local LLM inference engine in Rust and hand-written assembly.",
    "| [ZAYA1-8B](https://huggingface.co/Zyphra/ZAYA1-8B) | Hybrid Mamba + Compressed Cross Attention (CCA) + MoD + EDA | 760M / 8.4B (**9% active**) | On-device deployable, test-time-compute friendly, 128K ctx | Apache 2.0 |"
   ]
  },
  {
   "url": "https://github.com/xigh/open-weight-models",
   "title": "GitHub - xigh/open-weight-models: Curated list of open-weight AI models with commercially exploitable licenses, verified benchmarks, and no EU restrictions.",
   "published_at": null,
   "organization": "Stanford",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "7B beats 671B pass@4 on D3k",
   "organization_evidence": [
    "Open Weight Models",
    "A curated list of open-weight AI models with commercially exploitable licenses, verified benchmarks, and no geographic restrictions. Built to decide which models to support in [herbert-rs](https://github.com/xigh/herbert-rs), a local LLM inference engine in Rust and hand-written assembly."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Open Weight Models",
    "A curated list of open-weight AI models with commercially exploitable licenses, verified benchmarks, and no geographic restrictions. Built to decide which models to support in [herbert-rs](https://github.com/xigh/herbert-rs), a local LLM inference engine in Rust and hand-written assembly.",
    "**SGS \u2014 Self-Guided Self-Play** (Stanford, [arXiv:2604.20209](https://arxiv.org/abs/2604.20209), Apr 2026) is not a model but an RL self-play **algorithm** applied to DeepSeek-Prover-V2-7B."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Open Weight Models",
    "A curated list of open-weight AI models with commercially exploitable licenses, verified benchmarks, and no geographic restrictions. Built to decide which models to support in [herbert-rs](https://github.com/xigh/herbert-rs), a local LLM inference engine in Rust and hand-written assembly."
   ]
  },
  {
   "url": "https://github.com/xigh/open-weight-models",
   "title": "GitHub - xigh/open-weight-models: Curated list of open-weight AI models with commercially exploitable licenses, verified benchmarks, and no EU restrictions.",
   "published_at": null,
   "organization": "Prism ML",
   "model": null,
   "release_date": "2026-03",
   "access_type": null,
   "license": null,
   "significance": "1-bit Qwen3-8B fine-tune, CUDA/Metal/CPU/Android/iPhone",
   "organization_evidence": [
    "Open Weight Models",
    "A curated list of open-weight AI models with commercially exploitable licenses, verified benchmarks, and no geographic restrictions. Built to decide which models to support in [herbert-rs](https://github.com/xigh/herbert-rs), a local LLM inference engine in Rust and hand-written assembly.",
    "**Bonsai-8B** (Prism ML, Mar 2026) is a **1-bit end-to-end** fine-tune of Qwen3-8B: every projection + the LM head quantized to 1 bit (GGUF Q1_0), shrinking the deployed model to **1.15 GB**."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Open Weight Models",
    "A curated list of open-weight AI models with commercially exploitable licenses, verified benchmarks, and no geographic restrictions. Built to decide which models to support in [herbert-rs](https://github.com/xigh/herbert-rs), a local LLM inference engine in Rust and hand-written assembly.",
    "**Bonsai-8B** (Prism ML, Mar 2026) is a **1-bit end-to-end** fine-tune of Qwen3-8B: every projection + the LM head quantized to 1 bit (GGUF Q1_0), shrinking the deployed model to **1.15 GB**."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Open Weight Models",
    "A curated list of open-weight AI models with commercially exploitable licenses, verified benchmarks, and no geographic restrictions. Built to decide which models to support in [herbert-rs](https://github.com/xigh/herbert-rs), a local LLM inference engine in Rust and hand-written assembly."
   ]
  },
  {
   "url": "https://github.com/xigh/open-weight-models",
   "title": "GitHub - xigh/open-weight-models: Curated list of open-weight AI models with commercially exploitable licenses, verified benchmarks, and no EU restrictions.",
   "published_at": null,
   "organization": "Covenant AI + Mila",
   "model": null,
   "release_date": "2026-03",
   "access_type": null,
   "license": null,
   "significance": "Permissionless P2P, SparseLoCo optimizer, Bittensor blockchain (Subnet 3)",
   "organization_evidence": [
    "Open Weight Models",
    "A curated list of open-weight AI models with commercially exploitable licenses, verified benchmarks, and no geographic restrictions. Built to decide which models to support in [herbert-rs](https://github.com/xigh/herbert-rs), a local LLM inference engine in Rust and hand-written assembly."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Open Weight Models",
    "A curated list of open-weight AI models with commercially exploitable licenses, verified benchmarks, and no geographic restrictions. Built to decide which models to support in [herbert-rs](https://github.com/xigh/herbert-rs), a local LLM inference engine in Rust and hand-written assembly.",
    "Covenant-72B (Mar 2026) pre-trained a 72B dense LLaMA-3-style model over a permissionless blockchain network (Bittensor Subnet 3) on 1.1T tokens."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Open Weight Models",
    "A curated list of open-weight AI models with commercially exploitable licenses, verified benchmarks, and no geographic restrictions. Built to decide which models to support in [herbert-rs](https://github.com/xigh/herbert-rs), a local LLM inference engine in Rust and hand-written assembly.",
    "| [Covenant-72B](https://arxiv.org/abs/2603.08163) | Permissionless P2P, SparseLoCo optimizer, Bittensor blockchain (Subnet 3) | 72B dense | 1.1T (+14.8B SFT) | LLaMA-3 style, GQA, 80 layers, d=8192, 64 heads, 8 KV heads, RoPE 500K, ctx 2048\u21928192 | Apache 2.0 (checkpoints) |"
   ]
  },
  {
   "url": "https://github.com/xigh/open-weight-models",
   "title": "GitHub - xigh/open-weight-models: Curated list of open-weight AI models with commercially exploitable licenses, verified benchmarks, and no EU restrictions.",
   "published_at": null,
   "organization": "NVIDIA",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "any-to-any (text+audio+image+video \u2192 text), 256K ctx, Mamba-Transformer hybrid MoE",
   "organization_evidence": [
    "Open Weight Models",
    "A curated list of open-weight AI models with commercially exploitable licenses, verified benchmarks, and no geographic restrictions. Built to decide which models to support in [herbert-rs](https://github.com/xigh/herbert-rs), a local LLM inference engine in Rust and hand-written assembly.",
    "**Nemotron 3 Nano Omni** (NVIDIA, Apr 2026) extends the Nano family with native audio/video/image inputs."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Open Weight Models",
    "A curated list of open-weight AI models with commercially exploitable licenses, verified benchmarks, and no geographic restrictions. Built to decide which models to support in [herbert-rs](https://github.com/xigh/herbert-rs), a local LLM inference engine in Rust and hand-written assembly.",
    "**Nemotron 3 Nano Omni** (NVIDIA, Apr 2026) extends the Nano family with native audio/video/image inputs."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Open Weight Models",
    "A curated list of open-weight AI models with commercially exploitable licenses, verified benchmarks, and no geographic restrictions. Built to decide which models to support in [herbert-rs](https://github.com/xigh/herbert-rs), a local LLM inference engine in Rust and hand-written assembly.",
    "| [Nemotron 3 Nano Omni 30B-A3B](https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-Omni-30B-A3B) | -- | 3B | **any-to-any** (text+audio+image+video \u2192 text), 256K ctx, Mamba-Transformer hybrid MoE | Nemotron OML |"
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/gemma-4-31b-vs-claude-sonnet-4-6",
   "title": "Gemma 4 31B (Reasoning) vs Claude Sonnet 4.6 (Non-reasoning, High Effort): Model Comparison",
   "published_at": null,
   "organization": "Google",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "31B model with reasoning capabilities and open weights",
   "organization_evidence": [
    "# Gemma 4 31B (Reasoning) vs. Claude Sonnet 4.6 (Non-reasoning, High Effort)",
    "| Creator | ![Google](/img/logos/google_small.svg)[](https://deepmind.google/technologies/gemini/) | ![Anthropic](/img/logos/anthropic_small.svg)[](https://www.anthropic.com/) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Gemma 4 31B (Reasoning) vs. Claude Sonnet 4.6 (Non-reasoning, High Effort)",
    "| Release Date | April, 2026 | February, 2026 | Gemma 4 31B (Reasoning) has a more recent release date than Claude Sonnet 4.6 (Non-reasoning, High Effort) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Gemma 4 31B (Reasoning) vs. Claude Sonnet 4.6 (Non-reasoning, High Effort)",
    "| Metric | ![Google logo](/img/logos/google_small.svg)Gemma 4 31B (Reasoning) | ![Anthropic logo](/img/logos/anthropic_small.svg)Claude Sonnet 4.6 (Non-reasoning, High Effort) | Analysis |",
    "| Open Source (Weights) | Yes[](https://huggingface.co/google/gemma-4-31B-it) | No | Gemma 4 31B (Reasoning) is open source while Claude Sonnet 4.6 (Non-reasoning, High Effort) is proprietary |"
   ]
  },
  {
   "url": "https://chinabizinsider.com/moonshot-ai-detonates-open-source-race-with-kimi-k3-triggering-immediate-global-adoption",
   "title": "Moonshot AI Opens Kimi K3: 2.8T-Parameter MoE Goes Global",
   "published_at": "2026-07-28T01:02:50",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-28",
   "access_type": null,
   "license": null,
   "significance": "a 2.8-trillion-parameter Mixture-of-Experts model with native vision understanding and a one-million-token context window",
   "organization_evidence": [
    "Moonshot AI Detonates Open-Source Race With Kimi K3, Triggering Immediate Global Adoption",
    "Moonshot AI on July 28, 2026 open-sourced Kimi K3 \u2014 a 2.8-trillion-parameter Mixture-of-Experts model with native vision understanding and a one-million-token context window \u2014 alongside the full training infrastructure stack, a move that within 30 minutes made it the fastest-rising model in Hugging Face history and prompted immediate Day-0 integration commitments from U.S. AI infrastructure providers."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Moonshot AI Detonates Open-Source Race With Kimi K3, Triggering Immediate Global Adoption",
    "Moonshot AI on July 28, 2026 open-sourced Kimi K3 \u2014 a 2.8-trillion-parameter Mixture-of-Experts model with native vision understanding and a one-million-token context window \u2014 alongside the full training infrastructure stack, a move that within 30 minutes made it the fastest-rising model in Hugging Face history and prompted immediate Day-0 integration commitments from U.S. AI infrastructure providers."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Moonshot AI Detonates Open-Source Race With Kimi K3, Triggering Immediate Global Adoption",
    "Moonshot AI on July 28, 2026 open-sourced Kimi K3 \u2014 a 2.8-trillion-parameter Mixture-of-Experts model with native vision understanding and a one-million-token context window \u2014 alongside the full training infrastructure stack, a move that within 30 minutes made it the fastest-rising model in Hugging Face history and prompted immediate Day-0 integration commitments from U.S. AI infrastructure providers."
   ]
  },
  {
   "url": "https://aiskimmer.com/digest/2310003",
   "title": "Thinking Machines Unveils First Open-Weight Model Inkling",
   "published_at": "2026-07-16T00:05:03",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-16",
   "access_type": null,
   "license": null,
   "significance": "975B MoE multimodal open-weight model supporting a 1M token context window",
   "organization_evidence": [
    "# Thinking Machines Unveils First Open-Weight Model Inkling",
    "Mira Murati's Thinking Machines Lab released Inkling, a 975B MoE multimodal open-weight model."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Thinking Machines Unveils First Open-Weight Model Inkling",
    "**\ubaa8\ub378 \ucd9c\uc2dc** | Thu Jul 16 2026 00:00:00 GMT+0000 (Coordinated Universal Time) | 4 sources"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Thinking Machines Unveils First Open-Weight Model Inkling",
    "Mira Murati's Thinking Machines Lab released Inkling, a 975B MoE multimodal open-weight model."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/hy3-preview",
   "title": "Hy3-preview - Intelligence, Performance & Price Analysis | Artificial Analysis",
   "published_at": null,
   "organization": "Tencent",
   "model": null,
   "release_date": "2026-04-23",
   "access_type": null,
   "license": null,
   "significance": "A Mixture of Experts (MoE) model with 295 billion total parameters and 256k context window.",
   "organization_evidence": [
    "Hy3-preview (Reasoning) was created by Tencent.",
    "Tencent has launched a newer model, [Hy3](/models/hy3)."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Hy3-preview (Reasoning) was released on April 23, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Hy3-preview (Reasoning) is above average in intelligence and well priced when comparing to other open weight models of similar size. The model supports text input, outputs text, and has a 256k tokens context window.",
    "Hy3-preview (Reasoning) has 295 billion parameters (21 billion active).",
    "Hy3-preview (Reasoning) is a Mixture of Experts (MoE) model with 295 billion total parameters, but only 21 billion active parameters are used during inference."
   ]
  },
  {
   "url": "https://techdefused.com/a/cZFlsT8/meituan-open-sources-longcat-20-proving-china-can-train-frontier-ai-models-on-domestic-chips",
   "title": "Meituan open-sources LongCat-2.0, proving China can train frontier AI models on domestic chips",
   "published_at": "2026-06-30T12:57:21",
   "organization": "Meituan",
   "model": "LongCat-2.0",
   "release_date": "2026-06-30",
   "access_type": "open-source",
   "license": "MIT license",
   "significance": "1.6 trillion parameters trained entirely on domestically produced Chinese chips with a 1 million-token context window",
   "organization_evidence": [
    "Article",
    "# Meituan open-sources LongCat-2.0, proving China can train frontier AI models on domestic chips",
    "Meituan released LongCat-2.0 on June 30, 2026, an open-source large language model with 1.6 trillion parameters trained entirely on domestically produced Chinese chips."
   ],
   "model_evidence": [
    "Article",
    "# Meituan open-sources LongCat-2.0, proving China can train frontier AI models on domestic chips",
    "Meituan released LongCat-2.0 on June 30, 2026, an open-source large language model with 1.6 trillion parameters trained entirely on domestically produced Chinese chips."
   ],
   "release_date_evidence": [
    "Article",
    "# Meituan open-sources LongCat-2.0, proving China can train frontier AI models on domestic chips",
    "Meituan released LongCat-2.0 on June 30, 2026, an open-source large language model with 1.6 trillion parameters trained entirely on domestically produced Chinese chips."
   ],
   "access_type_evidence": [
    "Article",
    "# Meituan open-sources LongCat-2.0, proving China can train frontier AI models on domestic chips",
    "Meituan released LongCat-2.0 on June 30, 2026, an open-source large language model with 1.6 trillion parameters trained entirely on domestically produced Chinese chips."
   ],
   "license_evidence": [
    "Article",
    "# Meituan open-sources LongCat-2.0, proving China can train frontier AI models on domestic chips",
    "The model is open-sourced under the MIT license, making it available for enterprise and commercial use without the legal restrictions of copyleft licenses."
   ],
   "significance_evidence": [
    "Article",
    "# Meituan open-sources LongCat-2.0, proving China can train frontier AI models on domestic chips",
    "Meituan released LongCat-2.0 on June 30, 2026, an open-source large language model with 1.6 trillion parameters trained entirely on domestically produced Chinese chips.",
    "The model features a 1 million-token context window and uses a Mixture-of-Experts architecture that dynamically activates between 33 billion and 56 billion parameters per token, making it computationally efficient despite its massive scale."
   ]
  },
  {
   "url": "https://futureagi.com/blogs/best-llms-may-2026",
   "title": "Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
   "published_at": "2026-05-06T00:00:00",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-23",
   "access_type": null,
   "license": null,
   "significance": "1.6T parameter MoE with Hybrid Attention Architecture, scoring 80.6% on SWE-bench Verified at extremely low cost",
   "organization_evidence": [
    "Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "### DeepSeek V4-Pro and V4-Flash. Best cost-performance on the frontier",
    "*DeepSeek. Released April 23\u201324, 2026. MIT license, downloadable weights.*"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "### DeepSeek V4-Pro and V4-Flash. Best cost-performance on the frontier",
    "*DeepSeek. Released April 23\u201324, 2026. MIT license, downloadable weights.*"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "### DeepSeek V4-Pro and V4-Flash. Best cost-performance on the frontier",
    "**V4-Pro architecture:** 1.6 trillion total parameters with 49 billion active per token, pre-trained on more than 32 trillion tokens (per DeepSeek model card). MoE. **Hybrid Attention Architecture** combining Compressed Sparse Attention (CSA) and Heavily Compressed Attention (HCA). the first frontier-class model to ship this combination."
   ]
  },
  {
   "url": "https://futureagi.com/blogs/best-llms-may-2026",
   "title": "Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
   "published_at": "2026-05-06T00:00:00",
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-04-11",
   "access_type": null,
   "license": null,
   "significance": "Apache 2.0 multilingual open-weight family in three sizes beating Gemma 4 on every coding benchmark",
   "organization_evidence": [
    "Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "### Qwen 3.6 open-weight family. Best Apache 2.0 multilingual base",
    "*Alibaba. Released April 11\u201321, 2026. Apache 2.0, three sizes (72B / 35B-A3B / 27B).*"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "### Qwen 3.6 open-weight family. Best Apache 2.0 multilingual base",
    "*Alibaba. Released April 11\u201321, 2026. Apache 2.0, three sizes (72B / 35B-A3B / 27B).*"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "### Qwen 3.6 open-weight family. Best Apache 2.0 multilingual base",
    "- **Qwen 3.6-72B-dense** (April 11). flagship dense, 256K context, 94.8% HumanEval, 68.2% SWE-bench Verified, 71.4% LiveCodeBench. Beats Gemma 4 on every coding benchmark."
   ]
  },
  {
   "url": "https://futureagi.com/blogs/best-llms-may-2026",
   "title": "Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
   "published_at": "2026-05-06T00:00:00",
   "organization": "Mistral AI",
   "model": null,
   "release_date": "2026-03-16",
   "access_type": null,
   "license": null,
   "significance": "Smallest-footprint frontier MoE pick with hybrid reasoning and 6.5B active parameters",
   "organization_evidence": [
    "Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "### Mistral Small 4. Smallest frontier-class MoE for self-hosting",
    "*Mistral AI. Released March 16, 2026. Apache 2.0, 119B total / 6.5B active.*"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "### Mistral Small 4. Smallest frontier-class MoE for self-hosting",
    "*Mistral AI. Released March 16, 2026. Apache 2.0, 119B total / 6.5B active.*"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "### Mistral Small 4. Smallest frontier-class MoE for self-hosting",
    "The smallest-footprint frontier pick. 119B total parameters with 6.5B active per forward pass."
   ]
  },
  {
   "url": "https://futureagi.com/blogs/best-llms-may-2026",
   "title": "Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
   "published_at": "2026-05-06T00:00:00",
   "organization": "Google",
   "model": null,
   "release_date": "2026-04-02",
   "access_type": null,
   "license": null,
   "significance": "Google's open-weight family in four sizes up to 31B parameters",
   "organization_evidence": [
    "Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "### Gemma 4. Best small fine-tuning base in the Google ecosystem",
    "*Google. Released April 2, 2026. Apache 2.0, sizes up to 31B.*"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "### Gemma 4. Best small fine-tuning base in the Google ecosystem",
    "*Google. Released April 2, 2026. Apache 2.0, sizes up to 31B.*"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "### Gemma 4. Best small fine-tuning base in the Google ecosystem",
    "Google\u2019s open-weight family. Four sizes ranging up to 31B parameters."
   ]
  },
  {
   "url": "https://futureagi.com/blogs/best-llms-may-2026",
   "title": "Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
   "published_at": "2026-05-06T00:00:00",
   "organization": "Z.AI / Zhipu",
   "model": null,
   "release_date": "2026-04-07",
   "access_type": null,
   "license": null,
   "significance": "754B parameter MoE under MIT license, briefly leading SWE-bench Pro",
   "organization_evidence": [
    "Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "### GLM-5.1. Largest open-weight MIT-licensed flagship",
    "*Z.AI / Zhipu. Released April 7, 2026. MIT license, 754B parameter MoE.*"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "### GLM-5.1. Largest open-weight MIT-licensed flagship",
    "*Z.AI / Zhipu. Released April 7, 2026. MIT license, 754B parameter MoE.*"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "### GLM-5.1. Largest open-weight MIT-licensed flagship",
    "**754B parameter MoE under MIT license.** GLM-5.1 is the largest practically self-hostable MIT-licensed flagship in May 2026"
   ]
  },
  {
   "url": "https://futureagi.com/blogs/best-llms-may-2026",
   "title": "Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
   "published_at": "2026-05-06T00:00:00",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-04-20",
   "access_type": null,
   "license": null,
   "significance": "Current #1 open-weight model on the Artificial Analysis Intelligence Index with 1.1T parameters MoE",
   "organization_evidence": [
    "Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "### Kimi K2.6. Top open-weight on AA Intelligence Index",
    "*Moonshot AI. Released April 20, 2026. Open-weight, 1.1T MoE.*"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "### Kimi K2.6. Top open-weight on AA Intelligence Index",
    "*Moonshot AI. Released April 20, 2026. Open-weight, 1.1T MoE.*"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "### Kimi K2.6. Top open-weight on AA Intelligence Index",
    "**The current #1 open-weight model on the Artificial Analysis Intelligence Index.** Moonshot\u2019s latest, an upgrade to K2.5 with the primary improvement being **agentic stability over extended sessions**."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/glm-5-1-non-reasoning-vs-gpt-5-4-mini-medium",
   "title": "GLM-5.1 (Non-reasoning) vs GPT-5.4 mini (medium): Model Comparison",
   "published_at": null,
   "organization": "Z AI",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": null,
   "organization_evidence": [
    "# GLM-5.1 (Non-reasoning) vs. GPT-5.4 mini (medium)",
    "| Metric | ![Z AI logo](/img/logos/zai_small.svg)GLM-5.1 (Non-reasoning) | ![OpenAI logo](/img/logos/openai_small.svg)GPT-5.4 mini (medium) | Analysis |",
    "| Creator | ![Z AI](/img/logos/zai_small.svg)[](https://z.ai/) | ![OpenAI](/img/logos/openai_small.svg)[](https://openai.com/) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# GLM-5.1 (Non-reasoning) vs. GPT-5.4 mini (medium)",
    "| Metric | ![Z AI logo](/img/logos/zai_small.svg)GLM-5.1 (Non-reasoning) | ![OpenAI logo](/img/logos/openai_small.svg)GPT-5.4 mini (medium) | Analysis |",
    "| Release Date | April, 2026 | March, 2026 | GLM-5.1 (Non-reasoning) has a more recent release date than GPT-5.4 mini (medium) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": null
  },
  {
   "url": "https://bigtech.news/2026-07-20-china-moonshot-ai-releases-open-source-model.html",
   "title": "China\u2019s Moonshot AI Releases Open-Source Kimi K3 Model, Challenging Anthropic\u2019s Opus",
   "published_at": "2026-07-20T00:00:00",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-17",
   "access_type": null,
   "license": null,
   "significance": "contains 2.8 trillion parameters and is comparable to Anthropic\u2019s Claude Opus in performance",
   "organization_evidence": [
    "Chinese AI startup Moonshot AI released its Kimi K3 large language model as an open-source model on July 17, 2026, according to the company."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Chinese AI startup Moonshot AI released its Kimi K3 large language model as an open-source model on July 17, 2026, according to the company."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Chinese AI startup Moonshot AI released its Kimi K3 large language model as an open-source model on July 17, 2026, according to the company.",
    "The model is described by Moonshot AI as comparable to Anthropic\u2019s Claude Opus in performance, based on internal benchmarks.",
    "Kimi K3 contains 2.8 trillion parameters, serving as a measure of the AI\u2019s scale and processing power, according to a report by the BBC."
   ]
  },
  {
   "url": "https://telnyx.com/resources/best-open-source-llms",
   "title": "The 7 best open-source LLMs to know in 2026",
   "published_at": "2025-01-31T22:00:00",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-27",
   "access_type": null,
   "license": null,
   "significance": "uses a mixture-of-experts architecture with 2.8T total parameters, scoring 93.5% on GPQA Diamond, the highest of any open model per the Onyx leaderboard",
   "organization_evidence": [
    "## 4. Kimi K3",
    "Moonshot AI released Kimi K3 in July 2026 and published full weights on July 27, 2026."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## 4. Kimi K3",
    "Moonshot AI released Kimi K3 in July 2026 and published full weights on July 27, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## 4. Kimi K3",
    "The model uses a mixture-of-experts architecture with 2.8T total parameters, activating 16 of 896 experts per token, per Moonshot's model card.",
    "The headline number is reasoning.",
    "Kimi K3 scores 93.5% on GPQA Diamond, the highest of any open model per the Onyx leaderboard."
   ]
  },
  {
   "url": "https://telnyx.com/resources/best-open-source-llms",
   "title": "The 7 best open-source LLMs to know in 2026",
   "published_at": "2025-01-31T22:00:00",
   "organization": "Google",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "dense 32.2B-parameter model bringing dense-model performance to 85.2% MMLU and 80.0% SWE-Bench",
   "organization_evidence": [
    "## 7. Google Gemma 4 31B",
    "Google released Gemma 4 31B IT in April 2026 under an Apache 2.0 license, per Google's model card."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## 7. Google Gemma 4 31B",
    "Google released Gemma 4 31B IT in April 2026 under an Apache 2.0 license, per Google's model card."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## 7. Google Gemma 4 31B",
    "Unlike the MoE giants on this list, it is a dense 32.2B-parameter model.",
    "Gemma 4 31B brings dense-model performance to 85.2% MMLU and 80.0% SWE-Bench."
   ]
  },
  {
   "url": "https://datanorth.ai/news/nvidia-releases-nemotron-3-ultra-and-cosmos-3",
   "title": "NVIDIA Releases Nemotron 3 Ultra and Cosmos 3",
   "published_at": "2026-06-03T09:54:23",
   "organization": "NVIDIA",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "NVIDIA\u2019s largest open-weight language model to date, featuring approximately 550 billion total parameters with 55 billion active parameters per token and a 1 million token context window",
   "organization_evidence": [
    "# NVIDIA Releases Nemotron 3 Ultra and Cosmos 3",
    "NVIDIA announced Nemotron 3 Ultra and [Cosmos 3](https://datanorth.ai/news/nvidia-launches-cosmos-3) at its Computex 2026 keynote on June 1 in Taipei."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "NVIDIA announced Nemotron 3 Ultra and [Cosmos 3](https://datanorth.ai/news/nvidia-launches-cosmos-3) at its Computex 2026 keynote on June 1 in Taipei."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# NVIDIA Releases Nemotron 3 Ultra and Cosmos 3",
    "Nemotron 3 Ultra is NVIDIA\u2019s largest open-weight language model to date, featuring approximately 550 billion total parameters with 55 billion active parameters per token and a 1 million token context window."
   ]
  },
  {
   "url": "https://datanorth.ai/news/nvidia-releases-nemotron-3-ultra-and-cosmos-3",
   "title": "NVIDIA Releases Nemotron 3 Ultra and Cosmos 3",
   "published_at": "2026-06-03T09:54:23",
   "organization": "NVIDIA",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "the world\u2019s first fully open omnimodel, unifying language, image, video, audio, and action generation in a single architecture for robotics and physical AI applications",
   "organization_evidence": [
    "# NVIDIA Releases Nemotron 3 Ultra and Cosmos 3",
    "NVIDIA announced Nemotron 3 Ultra and [Cosmos 3](https://datanorth.ai/news/nvidia-launches-cosmos-3) at its Computex 2026 keynote on June 1 in Taipei."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "NVIDIA announced Nemotron 3 Ultra and [Cosmos 3](https://datanorth.ai/news/nvidia-launches-cosmos-3) at its Computex 2026 keynote on June 1 in Taipei."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# NVIDIA Releases Nemotron 3 Ultra and Cosmos 3",
    "Cosmos 3 is the world\u2019s first fully open omnimodel, unifying language, image, video, audio, and action generation in a single architecture for robotics and physical AI applications."
   ]
  },
  {
   "url": "https://memeburn.com/inkling-ai-model-specs-benchmarks-and-how-to-run-it",
   "title": "Inkling AI Model Specs, Benchmarks and How to Run It",
   "published_at": "2026-07-20T13:51:07",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "A 975-billion-parameter sparse Mixture-of-Experts system scoring 41 on the Artificial Analysis Intelligence Index and handling context lengths up to 1 million tokens.",
   "organization_evidence": [
    "# Inkling AI Model Specs, Benchmarks and How to Run It",
    "Thinking Machines Lab, the startup founded by former OpenAI CTO Mira Murati,[ released](https://thinkingmachines.ai/news/introducing-inkling/) its first AI model on July 15, 2026."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Inkling AI Model Specs, Benchmarks and How to Run It",
    "Thinking Machines Lab, the startup founded by former OpenAI CTO Mira Murati,[ released](https://thinkingmachines.ai/news/introducing-inkling/) its first AI model on July 15, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Inkling AI Model Specs, Benchmarks and How to Run It",
    "The Inkling AI model is Thinking Machines Lab's first release, a 975-billion-parameter open-weights system scoring 41 on the Artificial Analysis Intelligence Index.",
    "The Inkling AI model packs **975 billion parameters** into a sparse Mixture-of-Experts design, with roughly **41 billion active** on any given inference pass."
   ]
  },
  {
   "url": "https://neomanex.com/news/poolside-laguna-s-2-1-launch",
   "title": "Laguna S 2.1: The West's Open-Weight Answer Arrives | AI News",
   "published_at": "2026-07-22T07:30:00",
   "organization": "Poolside",
   "model": null,
   "release_date": "2026-07-21",
   "access_type": null,
   "license": null,
   "significance": "118B MoE open-weight coding model that beats DeepSeek-V4-Pro-Max, Inkling, and Nemotron 3 Ultra on Terminal-Bench 2.1",
   "organization_evidence": [
    "# Laguna S 2.1: The West's Open-Weight Answer Arrives",
    "Poolside released Laguna S 2.1, a 118B MoE open-weight coding model that beats DeepSeek-V4-Pro-Max, Inkling, and Nemotron 3 Ultra on Terminal-Bench 2.1."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Laguna S 2.1: The West's Open-Weight Answer Arrives",
    "PoolsideImportantJuly 22, 2026Models",
    "Poolside released Laguna S 2.1 on July 21 \u2014 a 118B-parameter Mixture-of-Experts coding model that activates only 8B parameters per token and decisively beats larger open-weight rivals on agentic coding benchmarks."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Laguna S 2.1: The West's Open-Weight Answer Arrives",
    "Poolside released Laguna S 2.1, a 118B MoE open-weight coding model that beats DeepSeek-V4-Pro-Max, Inkling, and Nemotron 3 Ultra on Terminal-Bench 2.1."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/claude-sonnet-4-6-vs-glm-5",
   "title": "Claude Sonnet 4.6 (Non-reasoning, High Effort) vs GLM-5 (Reasoning): Model Comparison",
   "published_at": null,
   "organization": "Z AI",
   "model": null,
   "release_date": "2026-02",
   "access_type": null,
   "license": null,
   "significance": "Reasoning model with open source weights",
   "organization_evidence": [
    "# Claude Sonnet 4.6 (Non-reasoning, High Effort) vs. GLM-5 (Reasoning)",
    "| Creator | ![Anthropic](/img/logos/anthropic_small.svg)[](https://www.anthropic.com/) | ![Z AI](/img/logos/zai_small.svg)[](https://z.ai/) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Claude Sonnet 4.6 (Non-reasoning, High Effort) vs. GLM-5 (Reasoning)",
    "| Release Date | February, 2026 | February, 2026 | Claude Sonnet 4.6 (Non-reasoning, High Effort) has a more recent release date than GLM-5 (Reasoning) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Claude Sonnet 4.6 (Non-reasoning, High Effort) vs. GLM-5 (Reasoning)",
    "| Open Source (Weights) | No | Yes[](https://huggingface.co/zai-org/GLM-5) | GLM-5 (Reasoning) is open source while Claude Sonnet 4.6 (Non-reasoning, High Effort) is proprietary |"
   ]
  },
  {
   "url": "https://chatforest.com/reviews/thinking-machines-lab-inkling-975b-open-weight-multimodal-mira-murati-july-2026",
   "title": "Mira Murati Ships Inkling: A 975B-Parameter Open-Weight AI Built for Fine-Tuning, Not Benchmark Wins \u2014 ChatForest",
   "published_at": null,
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "A 975B-parameter (41B active) multimodal mixture-of-experts model with 45 trillion tokens of pretraining data and a 1 million token context window, built for custom fine-tuning",
   "organization_evidence": [
    "MCP Server Review Published Jul 21, 2026",
    "Mira Murati Ships Inkling: A 975B-Parameter Open-Weight AI Built for Fine-Tuning, Not Benchmark Wins"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "MCP Server Review Published Jul 21, 2026",
    "Mira Murati Ships Inkling: A 975B-Parameter Open-Weight AI Built for Fine-Tuning, Not Benchmark Wins",
    "Released: July 15, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "MCP Server Review Published Jul 21, 2026",
    "Mira Murati Ships Inkling: A 975B-Parameter Open-Weight AI Built for Fine-Tuning, Not Benchmark Wins"
   ]
  },
  {
   "url": "https://yottalabs.ai/post/best-open-source-llms-2026",
   "title": "Best Open-Source LLMs in 2026: What Actually Runs in Production",
   "published_at": "2026-08-06T14:49:15",
   "organization": "Moonshot",
   "model": null,
   "release_date": "2026-07-27",
   "access_type": null,
   "license": null,
   "significance": "2.8 trillion total parameters with 104B active, 1M context, native vision, ranking second on Vals AI index and top five on Artificial Analysis.",
   "organization_evidence": [
    "# Best Open-Source LLMs in 2026: What Actually Runs in Production",
    "Moonshot is unusually honest about this, explicitly calling Kimi K3 \"open weight\" rather than open source.",
    "If the question is \"what is the strongest model whose weights I can hold,\" Moonshot's Kimi K3 is the current answer."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Best Open-Source LLMs in 2026: What Actually Runs in Production",
    "2.8 trillion total parameters with 104B active, 1M context, native vision, weights shipped July 27 on Hugging Face."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Best Open-Source LLMs in 2026: What Actually Runs in Production",
    "2.8 trillion total parameters with 104B active, 1M context, native vision, weights shipped July 27 on Hugging Face.",
    "It ranks second on the Vals AI index and top five on Artificial Analysis, the best third-party showing of any open-weight model to date."
   ]
  },
  {
   "url": "https://yottalabs.ai/post/best-open-source-llms-2026",
   "title": "Best Open-Source LLMs in 2026: What Actually Runs in Production",
   "published_at": "2026-08-06T14:49:15",
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-08-03",
   "access_type": null,
   "license": null,
   "significance": "A 2.4T flagship Qwen 3.8-Max and a new Qwen3.8-27B.",
   "organization_evidence": [
    "# Best Open-Source LLMs in 2026: What Actually Runs in Production",
    "Alibaba launched Qwen 3.8 on August 3 and committed to open weights for both models in the family, the 2.4T flagship Qwen 3.8-Max and a new Qwen3.8-27B, within about a week."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Best Open-Source LLMs in 2026: What Actually Runs in Production",
    "Alibaba launched Qwen 3.8 on August 3 and committed to open weights for both models in the family, the 2.4T flagship Qwen 3.8-Max and a new Qwen3.8-27B, within about a week."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Best Open-Source LLMs in 2026: What Actually Runs in Production",
    "Alibaba launched Qwen 3.8 on August 3 and committed to open weights for both models in the family, the 2.4T flagship Qwen 3.8-Max and a new Qwen3.8-27B, within about a week."
   ]
  },
  {
   "url": "https://medium.com/@rominur/the-state-of-open-source-large-language-models-d96948af9f41",
   "title": "The State of Open-Source Large Language Models:",
   "published_at": "2026-03-01T12:38:36",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-01-27",
   "access_type": null,
   "license": null,
   "significance": "Trillion-parameter titan with 32 billion active per forward pass via MoE architecture",
   "organization_evidence": [
    "## Introduction: The Open-Source AI Revolution",
    "Kimi K2.5 from Moonshot AI stands as the largest model on the entire list at a staggering 1 trillion parameters, with 32 billion active per forward pass via its MoE architecture."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## S-Tier: The Open-Source Elite",
    "Launched on January 27, 2026, it was the opening salvo of what became an extraordinary week of Chinese model releases."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## S-Tier: The Open-Source Elite",
    "Only four models have earned this distinction: **Kimi K2.5 (1T)**, **GLM-5 (744B)**, **MiniMax M2.5 (230B)**, and **Qwen 3.5 (397B)**.",
    "Kimi K2.5 from Moonshot AI stands as the largest model on the entire list at a staggering 1 trillion parameters, with 32 billion active per forward pass via its MoE architecture."
   ]
  },
  {
   "url": "https://medium.com/@rominur/the-state-of-open-source-large-language-models-d96948af9f41",
   "title": "The State of Open-Source Large Language Models:",
   "published_at": "2026-03-01T12:38:36",
   "organization": "Zhipu AI",
   "model": null,
   "release_date": "2026-02-11",
   "access_type": null,
   "license": null,
   "significance": "744 billion parameter model trained entirely on Huawei Ascend 910B chips",
   "organization_evidence": [
    "## S-Tier: The Open-Source Elite",
    "GLM-5 from Zhipu AI (now branded as Z.ai), at 744 billion parameters with 40 billion active, arrived on February 11, 2026 \u2014 just days before the Lunar New Year."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## S-Tier: The Open-Source Elite",
    "GLM-5 from Zhipu AI (now branded as Z.ai), at 744 billion parameters with 40 billion active, arrived on February 11, 2026 \u2014 just days before the Lunar New Year."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## S-Tier: The Open-Source Elite",
    "Only four models have earned this distinction: **Kimi K2.5 (1T)**, **GLM-5 (744B)**, **MiniMax M2.5 (230B)**, and **Qwen 3.5 (397B)**.",
    "GLM-5 from Zhipu AI (now branded as Z.ai), at 744 billion parameters with 40 billion active, arrived on February 11, 2026 \u2014 just days before the Lunar New Year.",
    "What makes GLM-5\u2019s story especially compelling is that it was trained entirely on 100,000 Huawei Ascend 910B chips without a single NVIDIA GPU, making it a geopolitical milestone as much as a technical one."
   ]
  },
  {
   "url": "https://medium.com/@rominur/the-state-of-open-source-large-language-models-d96948af9f41",
   "title": "The State of Open-Source Large Language Models:",
   "published_at": "2026-03-01T12:38:36",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-02-12",
   "access_type": null,
   "license": null,
   "significance": "230 billion parameter efficiency champion scoring 80.2% on SWE-Bench Verified",
   "organization_evidence": [
    "## S-Tier: The Open-Source Elite",
    "Perhaps the most impressive story in S-tier belongs to MiniMax M2.5, released on February 12, 2026 \u2014 just one day after GLM-5."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## S-Tier: The Open-Source Elite",
    "Perhaps the most impressive story in S-tier belongs to MiniMax M2.5, released on February 12, 2026 \u2014 just one day after GLM-5."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## S-Tier: The Open-Source Elite",
    "Only four models have earned this distinction: **Kimi K2.5 (1T)**, **GLM-5 (744B)**, **MiniMax M2.5 (230B)**, and **Qwen 3.5 (397B)**.",
    "At 230 billion total parameters with only 10 billion active per forward pass, it is the smallest model in the S-tier by a significant margin, yet it punches dramatically above its weight class."
   ]
  },
  {
   "url": "https://medium.com/@rominur/the-state-of-open-source-large-language-models-d96948af9f41",
   "title": "The State of Open-Source Large Language Models:",
   "published_at": "2026-03-01T12:38:36",
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-02-16",
   "access_type": null,
   "license": null,
   "significance": "397 billion parameter global contender supporting 201 languages and released under Apache 2.0",
   "organization_evidence": [
    "## S-Tier: The Open-Source Elite",
    "Rounding out the S-tier is Qwen 3.5 from Alibaba\u2019s Qwen team, released on February 16, 2026 \u2014 Chinese New Year\u2019s Day \u2014 at 397 billion total parameters with only 17 billion active."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## S-Tier: The Open-Source Elite",
    "Rounding out the S-tier is Qwen 3.5 from Alibaba\u2019s Qwen team, released on February 16, 2026 \u2014 Chinese New Year\u2019s Day \u2014 at 397 billion total parameters with only 17 billion active."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## S-Tier: The Open-Source Elite",
    "Only four models have earned this distinction: **Kimi K2.5 (1T)**, **GLM-5 (744B)**, **MiniMax M2.5 (230B)**, and **Qwen 3.5 (397B)**.",
    "Rounding out the S-tier is Qwen 3.5 from Alibaba\u2019s Qwen team, released on February 16, 2026 \u2014 Chinese New Year\u2019s Day \u2014 at 397 billion total parameters with only 17 billion active."
   ]
  },
  {
   "url": "https://machineherald.io/article/2026-05/06-mistral-medium-35-folds-chat-reasoning-and-coding-into-a-single-128-billion-parameter-open-weight-flagship",
   "title": "Mistral Medium 3.5 Folds Chat, Reasoning, and Coding Into a Single 128-Billion-Parameter Open-Weight Flagship",
   "published_at": "2026-05-06T09:57:02",
   "organization": "Mistral",
   "model": "Mistral Medium 3.5",
   "release_date": "2026-05",
   "access_type": "open-weight",
   "license": "Modified MIT License",
   "significance": "folds chat, reasoning, and coding into a single 128-billion-parameter open-weight flagship with a 256k context window",
   "organization_evidence": [
    "# Mistral Medium 3.5 Folds Chat, Reasoning, and Coding Into a Single 128-Billion-Parameter Open-Weight Flagship",
    "Mistral released Medium 3.5, a dense 128B open-weight model with a 256k context window that consolidates Medium 3.1, Magistral, and Devstral 2 under a Modified MIT license, with a per-query reasoning toggle."
   ],
   "model_evidence": [
    "# Mistral Medium 3.5 Folds Chat, Reasoning, and Coding Into a Single 128-Billion-Parameter Open-Weight Flagship",
    "Mistral released Medium 3.5, a dense 128B open-weight model with a 256k context window that consolidates Medium 3.1, Magistral, and Devstral 2 under a Modified MIT license, with a per-query reasoning toggle."
   ],
   "release_date_evidence": [
    "# Mistral Medium 3.5 Folds Chat, Reasoning, and Coding Into a Single 128-Billion-Parameter Open-Weight Flagship",
    "Sources: 3 Publisher: signed Contributor: signed Hash: 6372074616 [ View ](/provenance/2026-05/06-mistral-medium-35-folds-chat-reasoning-and-coding-into-a-single-128-billion-parameter-open-weight-flagship)"
   ],
   "access_type_evidence": [
    "# Mistral Medium 3.5 Folds Chat, Reasoning, and Coding Into a Single 128-Billion-Parameter Open-Weight Flagship",
    "Mistral released Medium 3.5, a dense 128B open-weight model with a 256k context window that consolidates Medium 3.1, Magistral, and Devstral 2 under a Modified MIT license, with a per-query reasoning toggle."
   ],
   "license_evidence": [
    "# Mistral Medium 3.5 Folds Chat, Reasoning, and Coding Into a Single 128-Billion-Parameter Open-Weight Flagship",
    "Mistral released Medium 3.5, a dense 128B open-weight model with a 256k context window that consolidates Medium 3.1, Magistral, and Devstral 2 under a Modified MIT license, with a per-query reasoning toggle."
   ],
   "significance_evidence": [
    "# Mistral Medium 3.5 Folds Chat, Reasoning, and Coding Into a Single 128-Billion-Parameter Open-Weight Flagship",
    "Mistral released Medium 3.5, a dense 128B open-weight model with a 256k context window that consolidates Medium 3.1, Magistral, and Devstral 2 under a Modified MIT license, with a per-query reasoning toggle."
   ]
  },
  {
   "url": "https://en.wedoany.com/shortnews/407398.html",
   "title": "2.8 Trillion! China's Moonshot AI Releases the World's Largest Open-Source Model by Parameters",
   "published_at": null,
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-16",
   "access_type": null,
   "license": null,
   "significance": "With a total parameter scale of 2.8 trillion, it is currently the world's largest open-source artificial intelligence model.",
   "organization_evidence": [
    "2.8 Trillion! China's Moonshot AI Releases the World's Largest Open-Source Model by Parameters",
    "Beijing-based Moonshot AI officially released its next-generation large model, Kimi K3, on July 16, 2026."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "2.8 Trillion! China's Moonshot AI Releases the World's Largest Open-Source Model by Parameters",
    "Beijing-based Moonshot AI officially released its next-generation large model, Kimi K3, on July 16, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "2.8 Trillion! China's Moonshot AI Releases the World's Largest Open-Source Model by Parameters",
    "With a total parameter scale of 2.8 trillion, it is currently the world's largest open-source artificial intelligence model."
   ]
  },
  {
   "url": "https://instadatanews.com/articles/2026-07-20-thinking-machines-inkling-open-weights-model.html",
   "title": "Mira Murati's Thinking Machines Ships Inkling, Its First Open Model",
   "published_at": "2026-07-20T04:00:00",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "Mixture-of-Experts transformer with 975 billion total parameters and 41 billion active per token, supporting a 1-million-token context window and native reasoning across text, images, and audio",
   "organization_evidence": [
    "AI Research",
    "# Mira Murati's Thinking Machines Ships Inkling, Its First Open Model",
    "**Thinking Machines Lab**, the AI startup founded by former OpenAI chief technology officer **Mira Murati**, released its first in-house model on July 15, 2026."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "AI Research",
    "# Mira Murati's Thinking Machines Ships Inkling, Its First Open Model",
    "**Thinking Machines Lab**, the AI startup founded by former OpenAI chief technology officer **Mira Murati**, released its first in-house model on July 15, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "AI Research",
    "## What Inkling Is",
    "Inkling is a **Mixture-of-Experts transformer** with **975 billion total parameters** and **41 billion active** per token."
   ]
  },
  {
   "url": "https://aitoolradar.io/blog/open-source-ai-2026-local-gemma-deepseek-llama",
   "title": "Open-Source AI 2026: Gemma 4, DeepSeek V4, Llama 4",
   "published_at": "2026-05-25T00:00:00",
   "organization": "DeepSeek",
   "model": "DeepSeek V4",
   "release_date": "2026-04-24",
   "access_type": "open weights",
   "license": "MIT license",
   "significance": "1.6T total / 49B active parameters and V4-Flash is 284B total / 13B active, both with a 1 million token context window and output up to 384K tokens",
   "organization_evidence": [
    "# Open-Source AI 2026: Gemma 4, DeepSeek V4, Llama 4",
    "DeepSeek, Alibaba and Google are shipping models that trade blows with frontier systems, and they publish the weights.",
    "**DeepSeek V4** shipped on 24 April 2026 as two models, both open weights under the MIT license."
   ],
   "model_evidence": [
    "# Open-Source AI 2026: Gemma 4, DeepSeek V4, Llama 4",
    "## The 2026 open flagships (server-class)",
    "**DeepSeek V4** shipped on 24 April 2026 as two models, both open weights under the MIT license."
   ],
   "release_date_evidence": [
    "# Open-Source AI 2026: Gemma 4, DeepSeek V4, Llama 4",
    "## The 2026 open flagships (server-class)",
    "**DeepSeek V4** shipped on 24 April 2026 as two models, both open weights under the MIT license."
   ],
   "access_type_evidence": [
    "# Open-Source AI 2026: Gemma 4, DeepSeek V4, Llama 4",
    "The big story in open-weight models is real: the distance between the best open models and the best closed ones has narrowed to a handful of benchmark points.",
    "**DeepSeek V4** shipped on 24 April 2026 as two models, both open weights under the MIT license."
   ],
   "license_evidence": [
    "# Open-Source AI 2026: Gemma 4, DeepSeek V4, Llama 4",
    "**DeepSeek V4** shipped on 24 April 2026 as two models, both open weights under the MIT license."
   ],
   "significance_evidence": [
    "# Open-Source AI 2026: Gemma 4, DeepSeek V4, Llama 4",
    "## The 2026 open flagships (server-class)",
    "[Per DeepSeek's own release notes](https://api-docs.deepseek.com/news/news260424), V4-Pro is 1.6T total / 49B active parameters and V4-Flash is 284B total / 13B active, both with a 1 million token context window and output up to 384K tokens."
   ]
  },
  {
   "url": "https://aitoolradar.io/blog/open-source-ai-2026-local-gemma-deepseek-llama",
   "title": "Open-Source AI 2026: Gemma 4, DeepSeek V4, Llama 4",
   "published_at": "2026-05-25T00:00:00",
   "organization": "Alibaba",
   "model": "Qwen 3.5",
   "release_date": "2026-02-16",
   "access_type": "available",
   "license": null,
   "significance": "carries 397B total parameters with 17B active, on a hybrid architecture, with native context up to 262K tokens",
   "organization_evidence": [
    "# Open-Source AI 2026: Gemma 4, DeepSeek V4, Llama 4",
    "## The 2026 open flagships (server-class)",
    "**Qwen 3.5** from Alibaba arrived 16 February 2026."
   ],
   "model_evidence": [
    "# Open-Source AI 2026: Gemma 4, DeepSeek V4, Llama 4",
    "## The 2026 open flagships (server-class)",
    "**Qwen 3.5** from Alibaba arrived 16 February 2026."
   ],
   "release_date_evidence": [
    "# Open-Source AI 2026: Gemma 4, DeepSeek V4, Llama 4",
    "## The 2026 open flagships (server-class)",
    "**Qwen 3.5** from Alibaba arrived 16 February 2026."
   ],
   "access_type_evidence": [
    "# Open-Source AI 2026: Gemma 4, DeepSeek V4, Llama 4",
    "## The 2026 open flagships (server-class)",
    "The flagship Qwen3.5-397B-A17B carries 397B total parameters with 17B active, on a hybrid architecture, with native context up to 262K tokens (extendable toward 1M)."
   ],
   "license_evidence": null,
   "significance_evidence": [
    "# Open-Source AI 2026: Gemma 4, DeepSeek V4, Llama 4",
    "## The 2026 open flagships (server-class)",
    "The flagship Qwen3.5-397B-A17B carries 397B total parameters with 17B active, on a hybrid architecture, with native context up to 262K tokens (extendable toward 1M)."
   ]
  },
  {
   "url": "https://aitoolradar.io/blog/open-source-ai-2026-local-gemma-deepseek-llama",
   "title": "Open-Source AI 2026: Gemma 4, DeepSeek V4, Llama 4",
   "published_at": "2026-05-25T00:00:00",
   "organization": "Google",
   "model": "Gemma 4",
   "release_date": "2026-04",
   "access_type": "open model",
   "license": null,
   "significance": "byte for byte the most capable open model",
   "organization_evidence": [
    "# Open-Source AI 2026: Gemma 4, DeepSeek V4, Llama 4",
    "## What actually runs locally (workstation-class)",
    "**Gemma 4** is Google's open-model family, released in April 2026."
   ],
   "model_evidence": [
    "# Open-Source AI 2026: Gemma 4, DeepSeek V4, Llama 4",
    "## What actually runs locally (workstation-class)",
    "**Gemma 4** is Google's open-model family, released in April 2026."
   ],
   "release_date_evidence": [
    "# Open-Source AI 2026: Gemma 4, DeepSeek V4, Llama 4",
    "## What actually runs locally (workstation-class)",
    "**Gemma 4** is Google's open-model family, released in April 2026."
   ],
   "access_type_evidence": [
    "# Open-Source AI 2026: Gemma 4, DeepSeek V4, Llama 4",
    "## What actually runs locally (workstation-class)",
    "**Gemma 4** is Google's open-model family, released in April 2026."
   ],
   "license_evidence": null,
   "significance_evidence": [
    "# Open-Source AI 2026: Gemma 4, DeepSeek V4, Llama 4",
    "## What actually runs locally (workstation-class)",
    "Google calls it [\"byte for byte the most capable open model\"](https://blog.google/innovation-and-ai/technology/ai/google-ai-updates-april-2026/) and notes Gemma has now been downloaded over 500 million times."
   ]
  },
  {
   "url": "https://scottharvanek.com/moonshot-kimi-k3-largest-open-model-thinking-machines-inkling-openai-gpt-red",
   "title": "Moonshot\u2019s Kimi K3 Becomes the Largest Open Model, Thinking Machines Debuts Inkling, and OpenAI Details GPT-Red Adversarial Safety Training",
   "published_at": "2026-07-18T12:25:59",
   "organization": "Moonshot AI",
   "model": "Kimi K3",
   "release_date": "2026-07-16",
   "access_type": "open",
   "license": null,
   "significance": "2.8-trillion-parameter model it calls the first open 3T-class model",
   "organization_evidence": [
    "# Moonshot\u2019s Kimi K3 Becomes the Largest Open Model, Thinking Machines Debuts Inkling, and OpenAI Details GPT-Red Adversarial Safety Training",
    "### Moonshot AI launches Kimi K3, a 2.8-trillion-parameter model it calls the first open 3T-class model",
    "**Moonshot AI \u00b7 July 16, 2026**",
    "Moonshot introduced Kimi K3, a 2.8T-parameter Mixture-of-Experts model (16 of 896 experts active) built on its Kimi Delta Attention and Attention Residuals architectures, with native vision and a 1-million-token context window."
   ],
   "model_evidence": [
    "# Moonshot\u2019s Kimi K3 Becomes the Largest Open Model, Thinking Machines Debuts Inkling, and OpenAI Details GPT-Red Adversarial Safety Training",
    "### Moonshot AI launches Kimi K3, a 2.8-trillion-parameter model it calls the first open 3T-class model",
    "Moonshot introduced Kimi K3, a 2.8T-parameter Mixture-of-Experts model (16 of 896 experts active) built on its Kimi Delta Attention and Attention Residuals architectures, with native vision and a 1-million-token context window."
   ],
   "release_date_evidence": [
    "# Moonshot\u2019s Kimi K3 Becomes the Largest Open Model, Thinking Machines Debuts Inkling, and OpenAI Details GPT-Red Adversarial Safety Training",
    "### Moonshot AI launches Kimi K3, a 2.8-trillion-parameter model it calls the first open 3T-class model",
    "**Moonshot AI \u00b7 July 16, 2026**"
   ],
   "access_type_evidence": [
    "# Moonshot\u2019s Kimi K3 Becomes the Largest Open Model, Thinking Machines Debuts Inkling, and OpenAI Details GPT-Red Adversarial Safety Training",
    "It was a heavyweight stretch for open-weight models: Moonshot shipped the largest open model ever announced, and Mira Murati\u2019s Thinking Machines Lab released its first model, while OpenAI published three posts spanning safety research, teen policy, and AI economics, and Google DeepMind and Isomorphic Labs laid out a joint biosecurity strategy.",
    "K3 is live on Kimi.com, Kimi Work, Kimi Code, and the Kimi API, with full open weights promised by July 27, 2026."
   ],
   "license_evidence": null,
   "significance_evidence": [
    "# Moonshot\u2019s Kimi K3 Becomes the Largest Open Model, Thinking Machines Debuts Inkling, and OpenAI Details GPT-Red Adversarial Safety Training",
    "### Moonshot AI launches Kimi K3, a 2.8-trillion-parameter model it calls the first open 3T-class model"
   ]
  },
  {
   "url": "https://scottharvanek.com/moonshot-kimi-k3-largest-open-model-thinking-machines-inkling-openai-gpt-red",
   "title": "Moonshot\u2019s Kimi K3 Becomes the Largest Open Model, Thinking Machines Debuts Inkling, and OpenAI Details GPT-Red Adversarial Safety Training",
   "published_at": "2026-07-18T12:25:59",
   "organization": "Thinking Machines Lab",
   "model": "Inkling",
   "release_date": "2026-07-15",
   "access_type": "open-weights",
   "license": null,
   "significance": "975B-parameter Mixture-of-Experts model trained from scratch on 45 trillion tokens",
   "organization_evidence": [
    "# Moonshot\u2019s Kimi K3 Becomes the Largest Open Model, Thinking Machines Debuts Inkling, and OpenAI Details GPT-Red Adversarial Safety Training",
    "### Thinking Machines Lab releases Inkling, its first open-weights model",
    "**Thinking Machines Lab \u00b7 July 15, 2026**"
   ],
   "model_evidence": [
    "# Moonshot\u2019s Kimi K3 Becomes the Largest Open Model, Thinking Machines Debuts Inkling, and OpenAI Details GPT-Red Adversarial Safety Training",
    "### Thinking Machines Lab releases Inkling, its first open-weights model",
    "Mira Murati\u2019s Thinking Machines Lab released Inkling, a 975B-parameter Mixture-of-Experts model (41B active) trained from scratch on 45 trillion tokens of text, images, audio, and video, with a 1M-token context window and controllable thinking effort."
   ],
   "release_date_evidence": [
    "# Moonshot\u2019s Kimi K3 Becomes the Largest Open Model, Thinking Machines Debuts Inkling, and OpenAI Details GPT-Red Adversarial Safety Training",
    "### Thinking Machines Lab releases Inkling, its first open-weights model",
    "**Thinking Machines Lab \u00b7 July 15, 2026**"
   ],
   "access_type_evidence": [
    "# Moonshot\u2019s Kimi K3 Becomes the Largest Open Model, Thinking Machines Debuts Inkling, and OpenAI Details GPT-Red Adversarial Safety Training",
    "### Thinking Machines Lab releases Inkling, its first open-weights model",
    "Full weights are on Hugging Face, and a lighter Inkling-Small (12B active) is previewed."
   ],
   "license_evidence": null,
   "significance_evidence": [
    "# Moonshot\u2019s Kimi K3 Becomes the Largest Open Model, Thinking Machines Debuts Inkling, and OpenAI Details GPT-Red Adversarial Safety Training",
    "### Thinking Machines Lab releases Inkling, its first open-weights model",
    "Mira Murati\u2019s Thinking Machines Lab released Inkling, a 975B-parameter Mixture-of-Experts model (41B active) trained from scratch on 45 trillion tokens of text, images, audio, and video, with a 1M-token context window and controllable thinking effort."
   ]
  },
  {
   "url": "https://heybeagle.com/blog/kimi-k2-7-code-ships-open-weights-for-a-1-trillion-parameter-coding-agent",
   "title": "Kimi K2.7-Code Ships Open Weights for a 1-Trillion-Parameter Coding Agent",
   "published_at": "2026-06-16T04:01:37",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-06-12",
   "access_type": null,
   "license": null,
   "significance": "A coding-focused 1-trillion-parameter update featuring a 30% reduction in thinking-token usage.",
   "organization_evidence": [
    "# Kimi K2.7-Code Ships Open Weights for a 1-Trillion-Parameter Coding Agent",
    "Moonshot AI released K2.7-Code on June 12, 2026 - a coding-focused update to their 1T MoE K2 line."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Kimi K2.7-Code Ships Open Weights for a 1-Trillion-Parameter Coding Agent",
    "Moonshot AI released K2.7-Code on June 12, 2026 - a coding-focused update to their 1T MoE K2 line."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Kimi K2.7-Code Ships Open Weights for a 1-Trillion-Parameter Coding Agent",
    "Moonshot AI released K2.7-Code on June 12, 2026 - a coding-focused update to their 1T MoE K2 line.",
    "The headline number is a claimed 30% reduction in thinking-token usage."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/qwen3-6-plus-vs-kimi-k2-5-non-reasoning",
   "title": "Qwen3.6 Plus vs Kimi K2.5 (Non-reasoning): Model Comparison",
   "published_at": null,
   "organization": "Kimi",
   "model": "Kimi K2.5 (Non-reasoning)",
   "release_date": "2026-01",
   "access_type": "open source",
   "license": null,
   "significance": "256k tokens context window with image input support and open weights availability",
   "organization_evidence": [
    "# Qwen3.6 Plus vs. Kimi K2.5 (Non-reasoning)",
    "## Model Comparison",
    "| Creator | ![Alibaba](/img/logos/alibaba_small.svg)[](https://www.alibabacloud.com/) | ![Kimi](/img/logos/kimi_small.png)[](https://www.moonshot.cn/) | |"
   ],
   "model_evidence": [
    "# Qwen3.6 Plus vs. Kimi K2.5 (Non-reasoning)",
    "## Model Comparison",
    "| Metric | ![Alibaba logo](/img/logos/alibaba_small.svg)Qwen3.6 Plus | ![Kimi logo](/img/logos/kimi_small.png)Kimi K2.5 (Non-reasoning) | Analysis |"
   ],
   "release_date_evidence": [
    "# Qwen3.6 Plus vs. Kimi K2.5 (Non-reasoning)",
    "## Model Comparison",
    "| Release Date | April, 2026 | January, 2026 | Qwen3.6 Plus has a more recent release date than Kimi K2.5 (Non-reasoning) |"
   ],
   "access_type_evidence": [
    "# Qwen3.6 Plus vs. Kimi K2.5 (Non-reasoning)",
    "## Model Comparison",
    "| Metric | ![Alibaba logo](/img/logos/alibaba_small.svg)Qwen3.6 Plus | ![Kimi logo](/img/logos/kimi_small.png)Kimi K2.5 (Non-reasoning) | Analysis |",
    "| Open Source (Weights) | No | Yes[](https://huggingface.co/moonshotai/Kimi-K2.5) | Kimi K2.5 (Non-reasoning) is open source while Qwen3.6 Plus is proprietary |"
   ],
   "license_evidence": null,
   "significance_evidence": [
    "# Qwen3.6 Plus vs. Kimi K2.5 (Non-reasoning)",
    "## Model Comparison",
    "| Context Window | 1000k tokens (~1500 A4 pages of size 12 Arial font) | 256k tokens (~384 A4 pages of size 12 Arial font) | Qwen3.6 Plus is larger than Kimi K2.5 (Non-reasoning) |",
    "| Image Input Support | Yes | Yes | Both Qwen3.6 Plus and Kimi K2.5 (Non-reasoning) have image input support |",
    "| Open Source (Weights) | No | Yes[](https://huggingface.co/moonshotai/Kimi-K2.5) | Kimi K2.5 (Non-reasoning) is open source while Qwen3.6 Plus is proprietary |"
   ]
  },
  {
   "url": "https://theideamagazine.com/science-nature/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - The Idea Magazine",
   "published_at": "2026-07-15T12:08:14",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor",
   "organization_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "Chinese labs such as DeepSeek, Z.ai, Moonshot, and Alibaba now each offer distinct approaches: DeepSeek emphasizes affordability with a large parameter count and low activation cost; Z.ai\u2019s GLM-5.2 leads in open-weight intelligence; Moonshot\u2019s Kimi line targets long-horizon stability; Alibaba\u2019s Qwen family offers compact, self-hostable variants.",
    "APR 24",
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "The first, DeepSeek V4, launched on April 24, features 1.6 trillion parameters but activates only 49 billion per pass, with a 1 million token context, and is priced at the lower end of the market.",
    "APR 24",
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "The first, DeepSeek V4, launched on April 24, features 1.6 trillion parameters but activates only 49 billion per pass, with a 1 million token context, and is priced at the lower end of the market.",
    "APR 24",
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor"
   ]
  },
  {
   "url": "https://theideamagazine.com/science-nature/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - The Idea Magazine",
   "published_at": "2026-07-15T12:08:14",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "cheap 1M-token context, native multimodal, modified-MIT",
   "organization_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "The subsequent releases include MiniMax M3 on June 1, and Kimi K2.7-Code along with GLM-5.2 in mid-June, all of which are openly downloadable and mostly licensed under permissive terms such as MIT.",
    "JUN 01"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "The subsequent releases include MiniMax M3 on June 1, and Kimi K2.7-Code along with GLM-5.2 in mid-June, all of which are openly downloadable and mostly licensed under permissive terms such as MIT.",
    "JUN 01"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "The subsequent releases include MiniMax M3 on June 1, and Kimi K2.7-Code along with GLM-5.2 in mid-June, all of which are openly downloadable and mostly licensed under permissive terms such as MIT.",
    "JUN 01"
   ]
  },
  {
   "url": "https://theideamagazine.com/science-nature/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - The Idea Magazine",
   "published_at": "2026-07-15T12:08:14",
   "organization": "Moonshot",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "agent-run specialist, ~30% fewer thinking tokens than K2.6",
   "organization_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "Chinese labs such as DeepSeek, Z.ai, Moonshot, and Alibaba now each offer distinct approaches: DeepSeek emphasizes affordability with a large parameter count and low activation cost; Z.ai\u2019s GLM-5.2 leads in open-weight intelligence; Moonshot\u2019s Kimi line targets long-horizon stability; Alibaba\u2019s Qwen family offers compact, self-hostable variants.",
    "JUN 13"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "The subsequent releases include MiniMax M3 on June 1, and Kimi K2.7-Code along with GLM-5.2 in mid-June, all of which are openly downloadable and mostly licensed under permissive terms such as MIT.",
    "JUN 13"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "The subsequent releases include MiniMax M3 on June 1, and Kimi K2.7-Code along with GLM-5.2 in mid-June, all of which are openly downloadable and mostly licensed under permissive terms such as MIT.",
    "JUN 13"
   ]
  },
  {
   "url": "https://theideamagazine.com/science-nature/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - The Idea Magazine",
   "published_at": "2026-07-15T12:08:14",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "753B MoE, MIT, top open-weight on Artificial Analysis index",
   "organization_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "Chinese labs such as DeepSeek, Z.ai, Moonshot, and Alibaba now each offer distinct approaches: DeepSeek emphasizes affordability with a large parameter count and low activation cost; Z.ai\u2019s GLM-5.2 leads in open-weight intelligence; Moonshot\u2019s Kimi line targets long-horizon stability; Alibaba\u2019s Qwen family offers compact, self-hostable variants.",
    "JUN 13\u201316"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "The subsequent releases include MiniMax M3 on June 1, and Kimi K2.7-Code along with GLM-5.2 in mid-June, all of which are openly downloadable and mostly licensed under permissive terms such as MIT.",
    "JUN 13\u201316"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "The subsequent releases include MiniMax M3 on June 1, and Kimi K2.7-Code along with GLM-5.2 in mid-June, all of which are openly downloadable and mostly licensed under permissive terms such as MIT.",
    "JUN 13\u201316"
   ]
  },
  {
   "url": "https://intelligibberish.com/articles/2026-07-28-kimi-k3-open-weights-2-8t-params-gpu-reality",
   "title": "Kimi K3 Weights Are Public: 2.8T Params, 64-GPU Reality",
   "published_at": "2026-07-28T00:00:00",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-27",
   "access_type": null,
   "license": null,
   "significance": "the world's first open 3T-class model, a 2.8-trillion-parameter mixture-of-experts with 104 billion active parameters and a one-million-token context window",
   "organization_evidence": [
    "# Kimi K3 Weights Are Public: 2.8T Params, 64-GPU Reality",
    "Moonshot AI made the full Kimi K3 weights public on July 27, 2026, turning the model from an API story into an open-weight one."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Kimi K3 Weights Are Public: 2.8T Params, 64-GPU Reality",
    "Moonshot released the full Kimi K3 weights on July 27, 2026: 2.8T params, 1M context, MXFP4. Read the license before you plan a deployment."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Kimi K3 Weights Are Public: 2.8T Params, 64-GPU Reality",
    "The card describes K3 as \u201cthe world\u2019s first open 3T-class model,\u201d a 2.8-trillion-parameter mixture-of-experts with 104 billion active parameters and a one-million-token context window ([Hugging Face model card](https://huggingface.co/moonshotai/Kimi-K3))."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/qwen3-5-2b-non-reasoning-vs-claude-opus-4-6-adaptive",
   "title": "Qwen3.5 2B (Non-reasoning) vs Claude Opus 4.6 (Adaptive Reasoning, Max Effort): Model Comparison",
   "published_at": null,
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-03",
   "access_type": null,
   "license": null,
   "significance": "Smaller context window compared to Claude Opus 4.6, open source model",
   "organization_evidence": [
    "## Model Comparison",
    "| Creator | ![Alibaba](/img/logos/alibaba_small.svg)[](https://www.alibabacloud.com/) | ![Anthropic](/img/logos/anthropic_small.svg)[](https://www.anthropic.com/) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## Model Comparison",
    "| Release Date | March, 2026 | February, 2026 | Qwen3.5 2B (Non-reasoning) has a more recent release date than Claude Opus 4.6 (Adaptive Reasoning, Max Effort) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## Model Comparison",
    "| Context Window | 262k tokens (~393 A4 pages of size 12 Arial font) | 1000k tokens (~1500 A4 pages of size 12 Arial font) | Qwen3.5 2B (Non-reasoning) is smaller than Claude Opus 4.6 (Adaptive Reasoning, Max Effort) |",
    "| Open Source (Weights) | Yes[](https://huggingface.co/Qwen/Qwen3.5-2B) | No | Qwen3.5 2B (Non-reasoning) is open source while Claude Opus 4.6 (Adaptive Reasoning, Max Effort) is proprietary |"
   ]
  },
  {
   "url": "https://baike.baidu.com/en/item/Qwen3.5-Plus/1478276",
   "title": "Qwen3.5-Plus",
   "published_at": null,
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-02-16",
   "access_type": null,
   "license": null,
   "significance": "has 397 billion parameters, supports 1M context, matches GPT-5.2 and Gemini-3-pro performance",
   "organization_evidence": [
    "Qwen3.5-Plus is an open-source large language model released by Alibaba on February 16, 2026, and belongs to the new generation of models in the Qianwen series.",
    "Owning company\nAlibaba"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Qwen3.5-Plus is an open-source large language model released by Alibaba on February 16, 2026, and belongs to the new generation of models in the Qianwen series.",
    "Release date\nFebruary 16, 2026"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "It has a total of 397 billion parameters, with 17 billion active parameters.",
    "The model supports a context length of 1M tokens and can directly process video content up to 2 hours long.",
    "In multiple benchmark tests, its performance is comparable to that of closed-source models like GPT-5.2 and Gemini-3-pro."
   ]
  },
  {
   "url": "https://aisocratic.org/updates/thinking-machines-introducing-inkling",
   "title": "Thinking Machines: Introducing Inkling",
   "published_at": "2026-07-22T04:00:00",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-22",
   "access_type": null,
   "license": null,
   "significance": "975B total parameters, 41B active \u2014 a Mixture-of-Experts transformer with 1M-token context window and native multimodality",
   "organization_evidence": [
    "# Thinking Machines: Introducing Inkling",
    "**Thinking Machines Lab** announced [**Inkling**](https://thinkingmachines.ai/news/introducing-inkling/) \u2014 its open-weights model \u2014 [via @thinkymachines](https://x.com/thinkymachines/status/2077454609551921208)."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Thinking Machines: Introducing Inkling",
    "July 22, 20262 min read"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Thinking Machines: Introducing Inkling",
    "- **975B total parameters, 41B active** \u2014 a Mixture-of-Experts transformer.",
    "- **Natively multimodal**: text, images, audio, and video, over a **1M-token** context window."
   ]
  },
  {
   "url": "https://pinggy.io/amp/blog/best_open_source_self_hosted_llms_for_coding",
   "title": "Best Open Source Self-Hosted LLMs for Coding in 2026",
   "published_at": null,
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-27",
   "access_type": null,
   "license": null,
   "significance": "largest open-weight model released to date with 2.8 trillion total parameters, 1M-token context, and top open-weight scores",
   "organization_evidence": [
    "# Best Open Source Self-Hosted LLMs for Coding in 2026",
    "Moonshot AI announced [Kimi K3](https://kimi.com/) on **July 16, 2026** and published the weights on [Hugging Face](https://huggingface.co/moonshotai/Kimi-K3) on **July 27, 2026**, along with a technical report and three of the infrastructure tools used to train it."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Best Open Source Self-Hosted LLMs for Coding in 2026",
    "Moonshot AI announced [Kimi K3](https://kimi.com/) on **July 16, 2026** and published the weights on [Hugging Face](https://huggingface.co/moonshotai/Kimi-K3) on **July 27, 2026**, along with a technical report and three of the infrastructure tools used to train it."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Best Open Source Self-Hosted LLMs for Coding in 2026",
    "At **2.8 trillion total parameters** it is the largest open-weight model released to date, and it hit the top of Hugging Face\u2019s trending chart within half an hour of going live."
   ]
  },
  {
   "url": "https://pinggy.io/amp/blog/best_open_source_self_hosted_llms_for_coding",
   "title": "Best Open Source Self-Hosted LLMs for Coding in 2026",
   "published_at": null,
   "organization": "Z.AI",
   "model": null,
   "release_date": "2026-06",
   "access_type": null,
   "license": null,
   "significance": "first open-weight model to beat GPT-5.5 on SWE-Bench Pro with a 1M-token context window",
   "organization_evidence": [
    "# Best Open Source Self-Hosted LLMs for Coding in 2026"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Best Open Source Self-Hosted LLMs for Coding in 2026"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Best Open Source Self-Hosted LLMs for Coding in 2026",
    "It runs a **1M-token context** and was the first open-weight model to beat GPT-5.5 on SWE-Bench Pro."
   ]
  },
  {
   "url": "https://pinggy.io/amp/blog/best_open_source_self_hosted_llms_for_coding",
   "title": "Best Open Source Self-Hosted LLMs for Coding in 2026",
   "published_at": null,
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "combines frontier coding performance, 1M-token context, and native multimodal input",
   "organization_evidence": [
    "# Best Open Source Self-Hosted LLMs for Coding in 2026"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Best Open Source Self-Hosted LLMs for Coding in 2026"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Best Open Source Self-Hosted LLMs for Coding in 2026"
   ]
  },
  {
   "url": "https://pinggy.io/amp/blog/best_open_source_self_hosted_llms_for_coding",
   "title": "Best Open Source Self-Hosted LLMs for Coding in 2026",
   "published_at": null,
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "MoE model family with 1M context in Pro and Flash variants",
   "organization_evidence": [
    "# Best Open Source Self-Hosted LLMs for Coding in 2026",
    "DeepSeek shipped **DeepSeek-V4** on **April 24, 2026** in two variants: [V4-Pro](https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro) (**1.6T total / 49B active**) and [V4-Flash](https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash) (**284B total / 13B active**), both **1M context**, MIT-licensed."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Best Open Source Self-Hosted LLMs for Coding in 2026",
    "DeepSeek shipped **DeepSeek-V4** on **April 24, 2026** in two variants: [V4-Pro](https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro) (**1.6T total / 49B active**) and [V4-Flash](https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash) (**284B total / 13B active**), both **1M context**, MIT-licensed."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Best Open Source Self-Hosted LLMs for Coding in 2026",
    "DeepSeek shipped **DeepSeek-V4** on **April 24, 2026** in two variants: [V4-Pro](https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro) (**1.6T total / 49B active**) and [V4-Flash](https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash) (**284B total / 13B active**), both **1M context**, MIT-licensed."
   ]
  },
  {
   "url": "https://pinggy.io/amp/blog/best_open_source_self_hosted_llms_for_coding",
   "title": "Best Open Source Self-Hosted LLMs for Coding in 2026",
   "published_at": null,
   "organization": "Xiaomi",
   "model": null,
   "release_date": "2026-04-22",
   "access_type": null,
   "license": null,
   "significance": "1.02T total parameter MoE model with 42B active parameters and a 1M token context window",
   "organization_evidence": [
    "# Best Open Source Self-Hosted LLMs for Coding in 2026"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Best Open Source Self-Hosted LLMs for Coding in 2026"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Best Open Source Self-Hosted LLMs for Coding in 2026",
    "It\u2019s a 1.02T total parameter MoE model with 42B active parameters and a 1M token context window - broadly comparable in scale to DeepSeek-V4 Pro."
   ]
  },
  {
   "url": "https://pinggy.io/amp/blog/best_open_source_self_hosted_llms_for_coding",
   "title": "Best Open Source Self-Hosted LLMs for Coding in 2026",
   "published_at": null,
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "best self-hostable option that fits a single consumer GPU",
   "organization_evidence": [
    "# Best Open Source Self-Hosted LLMs for Coding in 2026",
    "The general-purpose **Qwen 3.6** line (a 27B dense variant and the 35B-A3B MoE, both Apache 2.0, both released April 2026) is what shows up in this guide\u2019s LiveBench table: **71.78 Coding Avg / 39.29 Agentic Coding Avg** for the 27B, which makes it the best self-hostable option that fits a single consumer GPU."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Best Open Source Self-Hosted LLMs for Coding in 2026",
    "The general-purpose **Qwen 3.6** line (a 27B dense variant and the 35B-A3B MoE, both Apache 2.0, both released April 2026) is what shows up in this guide\u2019s LiveBench table: **71.78 Coding Avg / 39.29 Agentic Coding Avg** for the 27B, which makes it the best self-hostable option that fits a single consumer GPU."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Best Open Source Self-Hosted LLMs for Coding in 2026",
    "The general-purpose **Qwen 3.6** line (a 27B dense variant and the 35B-A3B MoE, both Apache 2.0, both released April 2026) is what shows up in this guide\u2019s LiveBench table: **71.78 Coding Avg / 39.29 Agentic Coding Avg** for the 27B, which makes it the best self-hostable option that fits a single consumer GPU."
   ]
  },
  {
   "url": "https://pinggy.io/amp/blog/best_open_source_self_hosted_llms_for_coding",
   "title": "Best Open Source Self-Hosted LLMs for Coding in 2026",
   "published_at": null,
   "organization": "Meta",
   "model": null,
   "release_date": "2026-08-10",
   "access_type": null,
   "license": null,
   "significance": "30B dense multimodal model built for agents running on a single consumer GPU",
   "organization_evidence": [
    "# Best Open Source Self-Hosted LLMs for Coding in 2026",
    "Meta Superintelligence Labs released [Muse Glimmer](https://research.meta.ai/blog/introducing-muse-glimmer-open-agentic-model) on **August 10, 2026**: a 30B dense multimodal model under **Apache 2.0**, built for agents that run on your own machine instead of in a datacenter."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Best Open Source Self-Hosted LLMs for Coding in 2026",
    "Meta Superintelligence Labs released [Muse Glimmer](https://research.meta.ai/blog/introducing-muse-glimmer-open-agentic-model) on **August 10, 2026**: a 30B dense multimodal model under **Apache 2.0**, built for agents that run on your own machine instead of in a datacenter."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Best Open Source Self-Hosted LLMs for Coding in 2026",
    "Meta Superintelligence Labs released [Muse Glimmer](https://research.meta.ai/blog/introducing-muse-glimmer-open-agentic-model) on **August 10, 2026**: a 30B dense multimodal model under **Apache 2.0**, built for agents that run on your own machine instead of in a datacenter."
   ]
  },
  {
   "url": "https://tesseract.academy/lessons/the-open-weights-wave-months-not-years-behind",
   "title": "The open-weights wave: months, not years, behind - Tesseract Academy",
   "published_at": "2026-07-29T16:44:35",
   "organization": null,
   "model": "V4",
   "release_date": "2026-04",
   "access_type": "open-weight",
   "license": null,
   "significance": "introduced a one-million-token context window and agent-oriented post-training capabilities",
   "organization_evidence": null,
   "model_evidence": [
    "Open-weight model evolution",
    "Subsequent models such as V3.2 (late 2025) and V4 (April 2026) further advanced the field."
   ],
   "release_date_evidence": [
    "Open-weight model evolution",
    "Subsequent models such as V3.2 (late 2025) and V4 (April 2026) further advanced the field."
   ],
   "access_type_evidence": [
    "The open-weights wave: months, not years, behind",
    "Subsequent models such as V3.2 (late 2025) and V4 (April 2026) further advanced the field."
   ],
   "license_evidence": null,
   "significance_evidence": [
    "Open-weight model evolution",
    "V4 introduced a one-million-token context window and agent-oriented post-training capabilities."
   ]
  },
  {
   "url": "https://tesseract.academy/lessons/the-open-weights-wave-months-not-years-behind",
   "title": "The open-weights wave: months, not years, behind - Tesseract Academy",
   "published_at": "2026-07-29T16:44:35",
   "organization": "Zhipu",
   "model": "GLM-5",
   "release_date": "2026-02",
   "access_type": "open-weight",
   "license": "MIT",
   "significance": "totalled 744 billion parameters with approximately 40 billion active parameters and achieved a 77.8% SWE-bench Verified score",
   "organization_evidence": [
    "China\u2019s open-weight contributions",
    "China\u2019s open-weight models have made significant inroads, particularly with Zhipu\u2019s GLM-5 series."
   ],
   "model_evidence": [
    "China\u2019s open-weight contributions",
    "Released in February 2026, GLM-5 totalled 744 billion parameters with approximately 40 billion active parameters."
   ],
   "release_date_evidence": [
    "China\u2019s open-weight contributions",
    "Released in February 2026, GLM-5 totalled 744 billion parameters with approximately 40 billion active parameters."
   ],
   "access_type_evidence": [
    "China\u2019s open-weight contributions",
    "China\u2019s open-weight models have made significant inroads, particularly with Zhipu\u2019s GLM-5 series."
   ],
   "license_evidence": [
    "China\u2019s open-weight contributions",
    "It was MIT licensed and trained on 100,000 Huawei Ascend chips without any US hardware."
   ],
   "significance_evidence": [
    "China\u2019s open-weight contributions",
    "Released in February 2026, GLM-5 totalled 744 billion parameters with approximately 40 billion active parameters.",
    "GLM-5 achieved a 77.8% SWE-bench Verified score, posting the strongest open coding results of its time."
   ]
  },
  {
   "url": "https://tesseract.academy/lessons/the-open-weights-wave-months-not-years-behind",
   "title": "The open-weights wave: months, not years, behind - Tesseract Academy",
   "published_at": "2026-07-29T16:44:35",
   "organization": "Moonshot",
   "model": "Kimi K2",
   "release_date": "2026",
   "access_type": "open-weight",
   "license": "Apache 2.0",
   "significance": "reached a trillion total parameters, with 32 billion active parameters and an 80.2% SWE-bench Verified score",
   "organization_evidence": [
    "China\u2019s open-weight contributions",
    "Moonshot\u2019s Kimi K2 line also reached a trillion total parameters, with 32 billion active parameters and an 80.2% SWE-bench Verified score."
   ],
   "model_evidence": [
    "China\u2019s open-weight contributions",
    "Moonshot\u2019s Kimi K2 line also reached a trillion total parameters, with 32 billion active parameters and an 80.2% SWE-bench Verified score."
   ],
   "release_date_evidence": [
    "The open-weights wave: months, not years, behind",
    "Moonshot\u2019s Kimi K2 line also reached a trillion total parameters, with 32 billion active parameters and an 80.2% SWE-bench Verified score."
   ],
   "access_type_evidence": [
    "China\u2019s open-weight contributions",
    "China\u2019s open-weight models have made significant inroads, particularly with Zhipu\u2019s GLM-5 series.",
    "Moonshot\u2019s Kimi K2 line also reached a trillion total parameters, with 32 billion active parameters and an 80.2% SWE-bench Verified score."
   ],
   "license_evidence": [
    "China\u2019s open-weight contributions",
    "The Kimi K2 line was released under the Apache 2.0 license."
   ],
   "significance_evidence": [
    "China\u2019s open-weight contributions",
    "Moonshot\u2019s Kimi K2 line also reached a trillion total parameters, with 32 billion active parameters and an 80.2% SWE-bench Verified score."
   ]
  },
  {
   "url": "https://courionai.com/news/2026-08-01-inkling-small-open-weights",
   "title": "Mira Murati's lab shrank its own model to a quarter of the size and lost one point",
   "published_at": "2026-08-01T00:00:00",
   "organization": "Thinking Machines",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "an open-weights reasoning model with 276 billion total parameters and 12 billion active, scoring 40 on the Artificial Analysis Intelligence Index where the 975-billion-parameter original scores 41",
   "organization_evidence": [
    "open-weights 3 min read",
    "# Mira Murati's lab shrank its own model to a quarter of the size and lost one point",
    "Thinking Machines released Inkling Small, an open-weights reasoning model with 276 billion parameters that scores 40 where the 975-billion-parameter original scores 41."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "open-weights 3 min read",
    "# Mira Murati's lab shrank its own model to a quarter of the size and lost one point",
    "Thinking Machines released Inkling Small, an open-weights reasoning model with 276 billion parameters that scores 40 where the 975-billion-parameter original scores 41."
   ]
  },
  {
   "url": "https://theairesourcehub.com/new",
   "title": "Model Release Desk - The AI Resource Hub",
   "published_at": null,
   "organization": "Google",
   "model": null,
   "release_date": "2026-04-02",
   "access_type": null,
   "license": null,
   "significance": "Open-weight Gemma 4 family anchor for local/open-source coverage",
   "organization_evidence": [
    "Tracked releases",
    "## Releases that matter for local model guides",
    "Gemma 4 31B is a currently tracked release from Google."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Tracked releases",
    "## Releases that matter for local model guides",
    "Backfill Needs research 2 Apr 2026",
    "Gemma 4 31B is a currently tracked release from Google."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Tracked releases",
    "## Releases that matter for local model guides",
    "Open-weight Gemma 4 family anchor for local/open-source coverage."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/deepseek-v4-flash",
   "title": "DeepSeek V4 Flash 0731 (max) - Intelligence, Performance & Price Analysis | Artificial Analysis",
   "published_at": null,
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-07-31",
   "access_type": null,
   "license": null,
   "significance": "A 284B parameter Mixture of Experts reasoning model with a 1M token context window, scoring 52 on the Artificial Analysis Intelligence Index.",
   "organization_evidence": [
    "DeepSeek V4 Flash 0731 (Reasoning, Max Effort) Intelligence, Performance & Price Analysis",
    "DeepSeek V4 Flash 0731 (Reasoning, Max Effort) was created by DeepSeek."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Released July 2026",
    "DeepSeek V4 Flash 0731 (Reasoning, Max Effort) Intelligence, Performance & Price Analysis",
    "DeepSeek V4 Flash 0731 (Reasoning, Max Effort) was released on July 31, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "DeepSeek V4 Flash 0731 (Reasoning, Max Effort) Intelligence, Performance & Price Analysis",
    "DeepSeek V4 Flash 0731 (Reasoning, Max Effort) is amongst the leading models in intelligence and reasonably priced when comparing to other open weight models of similar size.",
    "DeepSeek V4 Flash 0731 (Reasoning, Max Effort) scores 52 on the Artificial Analysis Intelligence Index, placing it well above average among comparable models (median: 27).",
    "| Context window | 1M ~1500 A4 pages of size 12 Arial font |",
    "| Total parameters | 284B |"
   ]
  },
  {
   "url": "http://english.scio.gov.cn/m/chinavoices/2026-07/20/content_118608213.html",
   "title": "China launches global open-source initiative for AI weather model",
   "published_at": "2026-07-20T00:00:00",
   "organization": "China Meteorological Administration",
   "model": "Fenghe",
   "release_date": "2026-07-17",
   "access_type": "open-source",
   "license": null,
   "significance": "trained on 50 million tokens of meteorological service data and integrated with authoritative weather datasets to support weather forecasting, risk assessment and meteorological services",
   "organization_evidence": [
    "# China launches global open-source initiative for AI weather model",
    "The announcement was made by the China Meteorological Administration (CMA) at a meteorological sub-forum of the 2026 World Artificial Intelligence Conference and High-Level Meeting on Global AI Governance in east China's Shanghai."
   ],
   "model_evidence": [
    "# China launches global open-source initiative for AI weather model",
    "China on Friday unveiled Fenghe, an open-source large language model dedicated to meteorological services, and launched a global open-source initiative for this model."
   ],
   "release_date_evidence": [
    "# China launches global open-source initiative for AI weather model",
    " Xinhua | July 20, 2026",
    "China on Friday unveiled Fenghe, an open-source large language model dedicated to meteorological services, and launched a global open-source initiative for this model.",
    "The announcement was made by the China Meteorological Administration (CMA) at a meteorological sub-forum of the 2026 World Artificial Intelligence Conference and High-Level Meeting on Global AI Governance in east China's Shanghai."
   ],
   "access_type_evidence": [
    "# China launches global open-source initiative for AI weather model",
    "China on Friday unveiled Fenghe, an open-source large language model dedicated to meteorological services, and launched a global open-source initiative for this model."
   ],
   "license_evidence": null,
   "significance_evidence": [
    "# China launches global open-source initiative for AI weather model",
    "Developed by the CMA, Fenghe is trained on 50 million tokens of meteorological service data and integrated with authoritative weather datasets to support weather forecasting, risk assessment and meteorological services."
   ]
  },
  {
   "url": "https://genalphai.com/the-2026-ai-model-landscape",
   "title": "AI Models 2026: The Mid-Year Frontier and Open-Weight Map",
   "published_at": "2026-06-16T05:53:05",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "80.6% on SWE-bench with 49B active parameters out of 1.6T total in a MoE architecture",
   "organization_evidence": [
    "The open-weight market is where the 2026 story gets interesting. Six families dominate, and most of the structural innovation came from Chinese labs.",
    "### DeepSeek",
    "The V4 line (March, April 2026) is the current generation, with the [DeepSeek V4 model card on NVIDIA NIM](https://build.nvidia.com/deepseek-ai/deepseek-v4-flash) anchoring the family."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The open-weight family table",
    "| Family | Frontier model | Active / total | License | SWE-bench Verified | Note |",
    "| DeepSeek | V4-Pro (Apr 2026) | 49B / 1.6T* | Apache 2.0 | 80.6% (Pro-Max) | Cheapest inference in tier |",
    "The V4 line (March, April 2026) is the current generation, with the [DeepSeek V4 model card on NVIDIA NIM](https://build.nvidia.com/deepseek-ai/deepseek-v4-flash) anchoring the family."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The open-weight family table",
    "| Family | Frontier model | Active / total | License | SWE-bench Verified | Note |",
    "| DeepSeek | V4-Pro (Apr 2026) | 49B / 1.6T* | Apache 2.0 | 80.6% (Pro-Max) | Cheapest inference in tier |",
    "DeepSeek's own post reports V4-Pro-Max at 80.6% on SWE-bench Verified."
   ]
  },
  {
   "url": "https://genalphai.com/the-2026-ai-model-landscape",
   "title": "AI Models 2026: The Mid-Year Frontier and Open-Weight Map",
   "published_at": "2026-06-16T05:53:05",
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "35B-total / 3B-active MoE reporting 73.4% SWE-bench in a size class that fits a 32GB GPU",
   "organization_evidence": [
    "The open-weight market is where the 2026 story gets interesting. Six families dominate, and most of the structural innovation came from Chinese labs.",
    "### Alibaba (Qwen)",
    "Qwen3.6-35B-A3B (April 2026) is a 35B-total / 3B-active MoE reporting 73.4% SWE-bench in a size class that fits a 32GB GPU."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The open-weight family table",
    "| Family | Frontier model | Active / total | License | SWE-bench Verified | Note |",
    "| Qwen | Qwen3.6-35B-A3B | 3B / 35B | Apache 2.0 | 73.4% | Best coding-per-GB |",
    "Qwen3.6-35B-A3B (April 2026) is a 35B-total / 3B-active MoE reporting 73.4% SWE-bench in a size class that fits a 32GB GPU."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The open-weight family table",
    "| Family | Frontier model | Active / total | License | SWE-bench Verified | Note |",
    "| Qwen | Qwen3.6-35B-A3B | 3B / 35B | Apache 2.0 | 73.4% | Best coding-per-GB |",
    "Qwen3.6-35B-A3B (April 2026) is a 35B-total / 3B-active MoE reporting 73.4% SWE-bench in a size class that fits a 32GB GPU."
   ]
  },
  {
   "url": "https://genalphai.com/the-2026-ai-model-landscape",
   "title": "AI Models 2026: The Mid-Year Frontier and Open-Weight Map",
   "published_at": "2026-06-16T05:53:05",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "32B active / 1T total parameters with ~75% SWE-bench and aggressive open reasoning",
   "organization_evidence": [
    "The open-weight market is where the 2026 story gets interesting. Six families dominate, and most of the structural innovation came from Chinese labs.",
    "### Moonshot AI (Kimi)"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The open-weight family table",
    "| Family | Frontier model | Active / total | License | SWE-bench Verified | Note |",
    "| Kimi | K2.6 (Apr 2026) | 32B / 1T* | Apache 2.0 | ~75%\u2020 | Aggressive open reasoning |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The open-weight family table",
    "| Family | Frontier model | Active / total | License | SWE-bench Verified | Note |",
    "| Kimi | K2.6 (Apr 2026) | 32B / 1T* | Apache 2.0 | ~75%\u2020 | Aggressive open reasoning |"
   ]
  },
  {
   "url": "https://kvmnode.com/en/blog/2026-kimi-k3-review-open-source-llm.html",
   "title": "Kimi K3 Review: The 2.8-Trillion-Parameter Open-Source Model That Challenges Claude and GPT",
   "published_at": "2026-07-17T00:00:00",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-27",
   "access_type": null,
   "license": null,
   "significance": "largest open-source AI model by parameter count \u2014 2.8T parameters, sparse MoE activating 16 of 896 experts, paired with a 1-million-token context window",
   "organization_evidence": [
    "For AI developers and engineering teams tracking **Kimi K3**, **Moonshot AI**, and **open-source LLMs**: on the night of July 16, 2026, Moonshot AI quietly added a **2.8-trillion-parameter** model to the top of its API docs \u2014 Kimi K3, now the largest open-source AI model in the world.",
    "On the night of July 16, 2026, Moonshot AI posted a banner at the top of its API documentation \u2014 \"\ud83c\udf89 Kimi K3 is live!\" \u2014 with no keynote, no hype cycle, just a technical blog post, a pricing page, and an immediately callable model ID `kimi-k3`."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Full weights open on **July 27**, at pricing roughly 40% below Claude Opus 4.8.",
    "| **Open weights** | July 27, 2026 (Hugging Face) |",
    "## What Is Kimi K3? The 2.8T Open-Source Model and Launch Context"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "**One-line definition:** Kimi K3 is currently the **largest open-source AI model by parameter count** \u2014 2.8T parameters, roughly 75% larger than DeepSeek V4 Pro (1.6T), 2.7x Xiaomi's open model (1.02T), and more than 7x Alibaba (397B).",
    "It uses sparse MoE architecture, activating 16 of 896 experts at inference; paired with a **1-million-token context window** (roughly five full copies of *Dream of the Red Chamber*) and native vision understanding, it targets complex coding, long-document reasoning, and knowledge work.",
    "## What Is Kimi K3? The 2.8T Open-Source Model and Launch Context"
   ]
  },
  {
   "url": "https://felloai.com/ko/kimi-k3",
   "title": "Kimi K3: Moonshot\u2019s 2.8T Open-Weight Model Explained, Specs, Pricing and Benchmarks",
   "published_at": "2026-07-16T23:59:00",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-27",
   "access_type": null,
   "license": null,
   "significance": "At 2.8 trillion total parameters, with 104 billion active per token, K3 is the largest open model anyone has released.",
   "organization_evidence": [
    "Kimi K3: Moonshot\u2019s 2.8T Open-Weight Model Explained, Specs, Pricing and Benchmarks",
    "Moonshot AI published the full **Kimi K3** weights on **July 27, 2026**, eleven days after the model went live in the Kimi app."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Kimi K3: Moonshot\u2019s 2.8T Open-Weight Model Explained, Specs, Pricing and Benchmarks",
    "Moonshot AI published the full **Kimi K3** weights on **July 27, 2026**, eleven days after the model went live in the Kimi app."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Kimi K3: Moonshot\u2019s 2.8T Open-Weight Model Explained, Specs, Pricing and Benchmarks",
    "At **2.8 trillion** total parameters, with **104 billion** active per token, K3 is the largest open model anyone has released."
   ]
  },
  {
   "url": "https://semifundamental.substack.com/p/china-ai-model-fundamentals",
   "title": "China AI Model Fundamentals",
   "published_at": "2026-05-07T13:49:09",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "1M-token context hybrid MoE with advanced compressed sparse attention",
   "organization_evidence": [
    "# China AI Model Fundamentals",
    "#### **4.** **DeepSeek (DeepSeek)**",
    "V4-Pro (1.6T total / 49B active MoE) and V4-Flash (284B total / 13B active) dropped as a preview on April 24, 2026 under MIT license."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# China AI Model Fundamentals",
    "#### **4.** **DeepSeek (DeepSeek)**",
    "V4-Pro (1.6T total / 49B active MoE) and V4-Flash (284B total / 13B active) dropped as a preview on April 24, 2026 under MIT license."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# China AI Model Fundamentals",
    "#### **4.** **DeepSeek (DeepSeek)**",
    "Both ship with 1M-token context, supported by a hybrid attention scheme that combines Compressed Sparse Attention (CSA) for moderately distant context and Heavily Compressed Attention (HCA) for very distant tokens."
   ]
  },
  {
   "url": "https://officeforge.co/blog/inkling-thinking-machines-open-weight",
   "title": "Inkling: 975B Open-Weights MoE Model Built for Customization",
   "published_at": "2026-08-02T00:00:00",
   "organization": "Thinking Machines Lab",
   "model": "Inkling",
   "release_date": "2026-07-15",
   "access_type": "open-weights",
   "license": null,
   "significance": "a broad, multimodal foundation model optimized for teams that want to fine-tune, customize, and deploy their own variants",
   "organization_evidence": [
    "On July 15, 2026, Thinking Machines Lab released Inkling \u2014 an open-weights, mixture-of-experts transformer with 975 billion total parameters and 41 billion active."
   ],
   "model_evidence": [
    "On July 15, 2026, Thinking Machines Lab released Inkling \u2014 an open-weights, mixture-of-experts transformer with 975 billion total parameters and 41 billion active."
   ],
   "release_date_evidence": [
    "On July 15, 2026, Thinking Machines Lab released Inkling \u2014 an open-weights, mixture-of-experts transformer with 975 billion total parameters and 41 billion active."
   ],
   "access_type_evidence": [
    "On July 15, 2026, Thinking Machines Lab released Inkling \u2014 an open-weights, mixture-of-experts transformer with 975 billion total parameters and 41 billion active."
   ],
   "license_evidence": null,
   "significance_evidence": [
    "In a landscape where every release is framed as a new leaderboard champion, Inkling takes a deliberately different position: it is a broad, multimodal foundation model optimized for teams that want to fine-tune, customize, and deploy their own variants \u2014 not just consume someone else's API."
   ]
  },
  {
   "url": "https://awesomeagents.ai/guides/state-of-open-source-llms-2026",
   "title": "State of Open-Source LLMs 2026: Rankings and Trends",
   "published_at": "2026-05-24T08:50:37",
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-02",
   "access_type": null,
   "license": null,
   "significance": "shipped Qwen 3.5, small models, and Qwen 3.6 covering the full range from tiny edge models to large multi-purpose ones",
   "organization_evidence": [
    "The state of open-source large language models in 2026 - who leads, how close they are to proprietary models, which licenses allow commercial use, and how to access them.",
    "### Qwen (Alibaba)",
    "Alibaba shipped Qwen 3.5 in February, followed by small models (0.8B to 9B) in March, and Qwen 3.6 in April with both a 35B MoE variant and a 27B dense model."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The state of open-source large language models in 2026 - who leads, how close they are to proprietary models, which licenses allow commercial use, and how to access them.",
    "### Qwen (Alibaba)",
    "Alibaba shipped Qwen 3.5 in February, followed by small models (0.8B to 9B) in March, and Qwen 3.6 in April with both a 35B MoE variant and a 27B dense model."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The state of open-source large language models in 2026 - who leads, how close they are to proprietary models, which licenses allow commercial use, and how to access them.",
    "### Qwen (Alibaba)",
    "Alibaba shipped Qwen 3.5 in February, followed by small models (0.8B to 9B) in March, and Qwen 3.6 in April with both a 35B MoE variant and a 27B dense model.",
    "The Qwen lineup covers the full range from tiny edge models to large multi-purpose ones."
   ]
  },
  {
   "url": "https://awesomeagents.ai/guides/state-of-open-source-llms-2026",
   "title": "State of Open-Source LLMs 2026: Rankings and Trends",
   "published_at": "2026-05-24T08:50:37",
   "organization": "DeepSeek AI",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "current benchmark leader in open-weight AI for coding and reasoning tasks",
   "organization_evidence": [
    "The state of open-source large language models in 2026 - who leads, how close they are to proprietary models, which licenses allow commercial use, and how to access them.",
    "### DeepSeek (DeepSeek AI)"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The state of open-source large language models in 2026 - who leads, how close they are to proprietary models, which licenses allow commercial use, and how to access them.",
    "### DeepSeek (DeepSeek AI)"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The state of open-source large language models in 2026 - who leads, how close they are to proprietary models, which licenses allow commercial use, and how to access them.",
    "### DeepSeek (DeepSeek AI)"
   ]
  },
  {
   "url": "https://onepin.ai/blog/minimax-m3-open-weights-model-2026",
   "title": "MiniMax M3: The First Open-Weight Model With Frontier Coding, 1M Context, and Native Multimodality",
   "published_at": "2026-06-01T00:00:00",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "the first open-weight model to combine frontier-level coding, a 1-million-token context window, and native multimodality in a single package",
   "organization_evidence": [
    "MiniMax M3: The First Open-Weight Model With Frontier Coding, 1M Context, and Native Multimodality",
    "MiniMax M3: The First Open-Weight Model With Frontier Coding, 1M Context, and Native Multimodality\n\n# TLDR\n\n[MiniMax](https://www.minimax.io) just released M3, the first open-weight model to combine frontier-level coding, a 1-million-token context window, and native multimodality in a single package."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "MiniMax M3: The First Open-Weight Model With Frontier Coding, 1M Context, and Native Multimodality"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "MiniMax M3: The First Open-Weight Model With Frontier Coding, 1M Context, and Native Multimodality",
    "MiniMax M3: The First Open-Weight Model With Frontier Coding, 1M Context, and Native Multimodality\n\n# TLDR\n\n[MiniMax](https://www.minimax.io) just released M3, the first open-weight model to combine frontier-level coding, a 1-million-token context window, and native multimodality in a single package."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/nvidia-nemotron-3-super-120b-a12b-vs-minimax-m2-5",
   "title": "Nemotron 3 Super 120B A12B (Reasoning) vs MiniMax-M2.5: Model Comparison",
   "published_at": null,
   "organization": "NVIDIA",
   "model": null,
   "release_date": "2026-03",
   "access_type": null,
   "license": null,
   "significance": "120.6B total parameters, 12.7B active at inference time, reasoning capabilities, 1M token context window",
   "organization_evidence": [
    "# Nemotron 3 Super 120B A12B (Reasoning) vs. MiniMax-M2.5",
    "| | ![NVIDIA logo](/img/logos/nvidia_small.svg)NVIDIA Nemotron 3 Super 120B A12B (Reasoning) | ![MiniMax logo](/img/logos/minimax_small.svg)MiniMax MiniMax-M2.5 | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Nemotron 3 Super 120B A12B (Reasoning) vs. MiniMax-M2.5",
    "| Release Date | March 2026 | February 2026 | Nemotron 3 Super 120B A12B (Reasoning) has a more recent release date than MiniMax-M2.5 |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Nemotron 3 Super 120B A12B (Reasoning) vs. MiniMax-M2.5",
    "| Parameters | 120.6B, 12.7B active at inference time | 230B, 10B active at inference time | MiniMax-M2.5 has more parameters than Nemotron 3 Super 120B A12B (Reasoning) |",
    "| Reasoning | Yes | Yes | Both Nemotron 3 Super 120B A12B (Reasoning) and MiniMax-M2.5 have reasoning |",
    "| Context Window | 1000k tokens~1,500 A4 pages of size 12 Arial font | 205k tokens~307 A4 pages of size 12 Arial font | Nemotron 3 Super 120B A12B (Reasoning) has a larger context window than MiniMax-M2.5 |"
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/nvidia-nemotron-3-super-120b-a12b-vs-minimax-m2-5",
   "title": "Nemotron 3 Super 120B A12B (Reasoning) vs MiniMax-M2.5: Model Comparison",
   "published_at": null,
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-02",
   "access_type": null,
   "license": null,
   "significance": "230B total parameters, 10B active at inference time, reasoning capabilities, Intelligence Index of 34",
   "organization_evidence": [
    "# Nemotron 3 Super 120B A12B (Reasoning) vs. MiniMax-M2.5",
    "| | ![NVIDIA logo](/img/logos/nvidia_small.svg)NVIDIA Nemotron 3 Super 120B A12B (Reasoning) | ![MiniMax logo](/img/logos/minimax_small.svg)MiniMax MiniMax-M2.5 | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Nemotron 3 Super 120B A12B (Reasoning) vs. MiniMax-M2.5",
    "| Release Date | March 2026 | February 2026 | Nemotron 3 Super 120B A12B (Reasoning) has a more recent release date than MiniMax-M2.5 |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Nemotron 3 Super 120B A12B (Reasoning) vs. MiniMax-M2.5",
    "| Intelligence Index | 26 | 34* | MiniMax-M2.5 is more intelligent than Nemotron 3 Super 120B A12B (Reasoning) |",
    "| Parameters | 120.6B, 12.7B active at inference time | 230B, 10B active at inference time | MiniMax-M2.5 has more parameters than Nemotron 3 Super 120B A12B (Reasoning) |",
    "| Reasoning | Yes | Yes | Both Nemotron 3 Super 120B A12B (Reasoning) and MiniMax-M2.5 have reasoning |"
   ]
  },
  {
   "url": "https://felloai.com/de/longcat-2-0",
   "title": "LongCat-2.0: Meituan\u2019s 1.6 Trillion-Parameter Open-Source Coding Model",
   "published_at": "2026-06-30T18:28:44",
   "organization": "Meituan",
   "model": "LongCat-2.0",
   "release_date": "2026-06-30",
   "access_type": "open source",
   "license": "MIT license",
   "significance": "1.6 trillion-parameter Mixture-of-Experts model with a native 1-million-token context window",
   "organization_evidence": [
    "# LongCat-2.0: Meituan\u2019s 1.6 Trillion-Parameter Open-Source Coding Model",
    "**Meituan** has open-sourced **LongCat-2.0**, a **1.6 trillion-parameter** Mixture-of-Experts model with a native **1-million-token context window**, released on **June 30, 2026** under a permissive **MIT license**."
   ],
   "model_evidence": [
    "# LongCat-2.0: Meituan\u2019s 1.6 Trillion-Parameter Open-Source Coding Model",
    "**Meituan** has open-sourced **LongCat-2.0**, a **1.6 trillion-parameter** Mixture-of-Experts model with a native **1-million-token context window**, released on **June 30, 2026** under a permissive **MIT license**."
   ],
   "release_date_evidence": [
    "# LongCat-2.0: Meituan\u2019s 1.6 Trillion-Parameter Open-Source Coding Model",
    "**Meituan** has open-sourced **LongCat-2.0**, a **1.6 trillion-parameter** Mixture-of-Experts model with a native **1-million-token context window**, released on **June 30, 2026** under a permissive **MIT license**."
   ],
   "access_type_evidence": [
    "# LongCat-2.0: Meituan\u2019s 1.6 Trillion-Parameter Open-Source Coding Model",
    "**Meituan** has open-sourced **LongCat-2.0**, a **1.6 trillion-parameter** Mixture-of-Experts model with a native **1-million-token context window**, released on **June 30, 2026** under a permissive **MIT license**."
   ],
   "license_evidence": [
    "**Meituan** has open-sourced **LongCat-2.0**, a **1.6 trillion-parameter** Mixture-of-Experts model with a native **1-million-token context window**, released on **June 30, 2026** under a permissive **MIT license**."
   ],
   "significance_evidence": [
    "# LongCat-2.0: Meituan\u2019s 1.6 Trillion-Parameter Open-Source Coding Model",
    "**Meituan** has open-sourced **LongCat-2.0**, a **1.6 trillion-parameter** Mixture-of-Experts model with a native **1-million-token context window**, released on **June 30, 2026** under a permissive **MIT license**."
   ]
  },
  {
   "url": "https://aitoolsreview.co.uk/insights/thinking-machines-inkling",
   "title": "Inkling Review: Thinking Machines' First Open Model",
   "published_at": "2026-07-17T11:19:27",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "A 975-billion-parameter Mixture-of-Experts model with a context window up to 1 million tokens, pretrained on 45 trillion multimodal tokens",
   "organization_evidence": [
    "Quick answer:",
    "**Inkling is Thinking Machines Lab's first in-house AI model**, released on 15 July 2026 by Mira Murati's startup."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Quick answer:",
    "**Inkling is Thinking Machines Lab's first in-house AI model**, released on 15 July 2026 by Mira Murati's startup."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Quick answer:",
    "It is a 975-billion-parameter Mixture-of-Experts model (41B active) with a context window up to 1 million tokens, pretrained on 45 trillion multimodal tokens, and released open-weight on Hugging Face."
   ]
  },
  {
   "url": "https://stateofopensource.ai/changelog",
   "title": "The State of Open Source AI",
   "published_at": null,
   "organization": "Moonshot",
   "model": null,
   "release_date": "2026-07-27",
   "access_type": null,
   "license": null,
   "significance": "debuts at #4, four points off the closed frontier, ahead of Opus 4.8, GPT-5.6 Terra, Grok 4.5 and Sonnet 5",
   "organization_evidence": [
    "The State of Open Source AI",
    "Moonshot shipped a frontier-class open model, API on 16 July and weights on 27 July, between v1.0 and this release."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The State of Open Source AI",
    "Moonshot shipped a frontier-class open model, API on 16 July and weights on 27 July, between v1.0 and this release."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The State of Open Source AI",
    "Section 01 \u00b7 The current state",
    "**\"The frontier's top three are closed. The fourth is open.\"** Artificial Analysis Intelligence Index v4.1. Kimi K3 debuts at #4, four points off the closed frontier, ahead of Opus 4.8, GPT-5.6 Terra, Grok 4.5 and Sonnet 5."
   ]
  },
  {
   "url": "https://onthewire.ai/article/the-independent-numbers-are-in-inkling-leads-us-open-weights-with-one-asterisk-t",
   "title": "The Independent Numbers Are In: Inkling Leads US Open Weights \u2014 With One Asterisk That Matters",
   "published_at": "2026-07-16T00:00:00",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "975B open-weights model that is the new leading U.S. open weights model",
   "organization_evidence": [
    "# The Independent Numbers Are In: Inkling Leads US Open Weights \u2014 With One Asterisk That Matters",
    "When Thinking Machines Lab [released its 975B open-weights model yesterday](/article/daily-update-16-july-2026-america-gets-its-open-weights-giant-grok-bares-its-sou), the loudest early take on Hacker News was \"not as good as GLM 5.2 for agentic workflows while also being bigger.\""
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# The Independent Numbers Are In: Inkling Leads US Open Weights \u2014 With One Asterisk That Matters",
    "By **Relay** \u2014 AI EditorAI",
    "16 July 2026",
    "When Thinking Machines Lab [released its 975B open-weights model yesterday](/article/daily-update-16-july-2026-america-gets-its-open-weights-giant-grok-bares-its-sou), the loudest early take on Hacker News was \"not as good as GLM 5.2 for agentic workflows while also being bigger.\""
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# The Independent Numbers Are In: Inkling Leads US Open Weights \u2014 With One Asterisk That Matters",
    "When Thinking Machines Lab [released its 975B open-weights model yesterday](/article/daily-update-16-july-2026-america-gets-its-open-weights-giant-grok-bares-its-sou), the loudest early take on Hacker News was \"not as good as GLM 5.2 for agentic workflows while also being bigger.\""
   ]
  },
  {
   "url": "https://aitechconnect.in/news/open-weight-coding-models-wave-kimi-glm-minimax-2026",
   "title": "The open-weight coding wave: Kimi K2.6, GLM-5.1, MiniMax M3",
   "published_at": "2026-06-19T08:00:00",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-04-07",
   "access_type": null,
   "license": null,
   "significance": "A Mixture-of-Experts model of roughly 754 billion total parameters with around 40 billion active per token, and an approximately 200K-token context window, reported at the top of SWE-Bench Pro at around 58.4.",
   "organization_evidence": [
    "## What actually shipped between April and June 2026",
    "**GLM-5.1**, from Z.ai (the lab formerly known as Zhipu AI), arrived first, on 7 April 2026."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## What actually shipped between April and June 2026",
    "**GLM-5.1**, from Z.ai (the lab formerly known as Zhipu AI), arrived first, on 7 April 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## What actually shipped between April and June 2026",
    "It is a Mixture-of-Experts model of roughly 754 billion total parameters with around 40 billion active per token, and an approximately 200K-token context window.",
    "Z.ai reported it at the top of SWE-Bench Pro at around 58.4, narrowly ahead of GPT-5.4 (about 57.7) and Claude Opus 4.6 (about 57.3) on that particular benchmark."
   ]
  },
  {
   "url": "https://aitechconnect.in/news/open-weight-coding-models-wave-kimi-glm-minimax-2026",
   "title": "The open-weight coding wave: Kimi K2.6, GLM-5.1, MiniMax M3",
   "published_at": "2026-06-19T08:00:00",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-04-21",
   "access_type": null,
   "license": null,
   "significance": "A 1-trillion-parameter MoE with around 32 billion active parameters and a 256K context window, natively multimodal with an 'Agent Swarm' orchestration mode.",
   "organization_evidence": [
    "## What actually shipped between April and June 2026",
    "**Kimi K2.6**, from Moonshot AI, followed on 21 April 2026."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## What actually shipped between April and June 2026",
    "**Kimi K2.6**, from Moonshot AI, followed on 21 April 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## What actually shipped between April and June 2026",
    "It is larger still: a 1-trillion-parameter MoE with around 32 billion active parameters and a 256K context window.",
    "It is natively multimodal, with a roughly 400-million-parameter MoonViT vision encoder bolted in, and Moonshot reported a SWE-Bench Pro figure of about 58.6.",
    "Its headline feature is an \"Agent Swarm\" orchestration mode that the lab says scales to around 300 sub-agents and roughly 4,000 steps for long-horizon tasks."
   ]
  },
  {
   "url": "https://aitechconnect.in/news/open-weight-coding-models-wave-kimi-glm-minimax-2026",
   "title": "The open-weight coding wave: Kimi K2.6, GLM-5.1, MiniMax M3",
   "published_at": "2026-06-19T08:00:00",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "Combined frontier coding, a 1-million-token context window and native multimodality in one package, reported to top the open-weight field on SWE-Bench Pro at around 59.0.",
   "organization_evidence": [
    "## What actually shipped between April and June 2026",
    "**MiniMax M3** rounds out the wave."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## What actually shipped between April and June 2026",
    "It shipped on 1 June 2026 with open weights following around 11 June 2026, and we covered it in detail in our [MiniMax M3 deep-dive](/news/minimax-m3-open-weight-frontier-coding-1m-context-2026)."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## What actually shipped between April and June 2026",
    "It was billed as the first open-weight model to combine frontier coding, a 1-million-token context window and native multimodality in one package, and it was reported to top the open-weight field on SWE-Bench Pro at around 59.0."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/qwen3-5-122b-a10b-vs-kimi-k2-5",
   "title": "Qwen3.5 122B A10B (Reasoning) vs Kimi K2.5 (Reasoning): Model Comparison",
   "published_at": null,
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-02",
   "access_type": null,
   "license": null,
   "significance": "125B, 10B active at inference time",
   "organization_evidence": [
    "# Qwen3.5 122B A10B (Reasoning) vs. Kimi K2.5 (Reasoning)",
    "## Model Comparison",
    "| Creator | ![Alibaba](/img/logos/alibaba_small.svg)[](https://www.alibabacloud.com/) | ![Kimi](/img/logos/kimi_small.png)[](https://www.moonshot.cn/) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Qwen3.5 122B A10B (Reasoning) vs. Kimi K2.5 (Reasoning)",
    "## Model Comparison",
    "| Release Date | February, 2026 | January, 2026 | Qwen3.5 122B A10B (Reasoning) has a more recent release date than Kimi K2.5 (Reasoning) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Qwen3.5 122B A10B (Reasoning) vs. Kimi K2.5 (Reasoning)",
    "## Model Comparison",
    "| Parameters | 125B, 10B active at inference time | 1000B, 32B active at inference time | Qwen3.5 122B A10B (Reasoning) is smaller than Kimi K2.5 (Reasoning) |"
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/qwen3-5-122b-a10b-vs-kimi-k2-5",
   "title": "Qwen3.5 122B A10B (Reasoning) vs Kimi K2.5 (Reasoning): Model Comparison",
   "published_at": null,
   "organization": "Kimi",
   "model": null,
   "release_date": "2026-01",
   "access_type": null,
   "license": null,
   "significance": "1000B, 32B active at inference time",
   "organization_evidence": [
    "# Qwen3.5 122B A10B (Reasoning) vs. Kimi K2.5 (Reasoning)",
    "## Model Comparison",
    "| Creator | ![Alibaba](/img/logos/alibaba_small.svg)[](https://www.alibabacloud.com/) | ![Kimi](/img/logos/kimi_small.png)[](https://www.moonshot.cn/) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Qwen3.5 122B A10B (Reasoning) vs. Kimi K2.5 (Reasoning)",
    "## Model Comparison",
    "| Release Date | February, 2026 | January, 2026 | Qwen3.5 122B A10B (Reasoning) has a more recent release date than Kimi K2.5 (Reasoning) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Qwen3.5 122B A10B (Reasoning) vs. Kimi K2.5 (Reasoning)",
    "## Model Comparison",
    "| Parameters | 125B, 10B active at inference time | 1000B, 32B active at inference time | Qwen3.5 122B A10B (Reasoning) is smaller than Kimi K2.5 (Reasoning) |"
   ]
  },
  {
   "url": "https://dreamridiculous.com/artificial-intelligence/the-accelerated-pace-of-china-s-ai-releases-four-frontier-models-in-record-time",
   "title": "The Accelerated Pace Of China\u2019s AI Releases: Four Frontier Models In Record Time - DreamRidiculous",
   "published_at": "2026-07-15T06:14:01",
   "organization": "DeepSeek",
   "model": "DeepSeek V4",
   "release_date": "2026-04-24",
   "access_type": "open-weight",
   "license": "MIT",
   "significance": "1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor",
   "organization_evidence": [
    "**Chinese AI labs have released four frontier-class open-weight models in roughly eight weeks,** marking a record-breaking cadence that significantly accelerates China\u2019s position in the global AI race.",
    "The Accelerated Pace Of China\u2019s AI Releases: Four Frontier Models In Record Time"
   ],
   "model_evidence": [
    "From late April to mid-June 2026, Chinese laboratories introduced four major open-weight models: **DeepSeek V4** on April 24, **MiniMax M3** on June 1, and **Kimi K2.7-Code** along with **GLM-5.2** in mid-June.",
    "The Accelerated Pace Of China\u2019s AI Releases: Four Frontier Models In Record Time"
   ],
   "release_date_evidence": [
    "From late April to mid-June 2026, Chinese laboratories introduced four major open-weight models: **DeepSeek V4** on April 24, **MiniMax M3** on June 1, and **Kimi K2.7-Code** along with **GLM-5.2** in mid-June.",
    "The Accelerated Pace Of China\u2019s AI Releases: Four Frontier Models In Record Time"
   ],
   "access_type_evidence": [
    "In just eight weeks, Chinese AI labs launched four frontier-class open-weight models, accelerating China\u2019s lead in open AI development.",
    "The Accelerated Pace Of China\u2019s AI Releases: Four Frontier Models In Record Time",
    "## Four Frontier-Class Open Models in Eight Weeks"
   ],
   "license_evidence": [
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor",
    "### The production line \u2014 spring 2026",
    "The Accelerated Pace Of China\u2019s AI Releases: Four Frontier Models In Record Time"
   ],
   "significance_evidence": [
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor",
    "### The production line \u2014 spring 2026",
    "The Accelerated Pace Of China\u2019s AI Releases: Four Frontier Models In Record Time"
   ]
  },
  {
   "url": "https://dreamridiculous.com/artificial-intelligence/the-accelerated-pace-of-china-s-ai-releases-four-frontier-models-in-record-time",
   "title": "The Accelerated Pace Of China\u2019s AI Releases: Four Frontier Models In Record Time - DreamRidiculous",
   "published_at": "2026-07-15T06:14:01",
   "organization": "MiniMax",
   "model": "MiniMax M3",
   "release_date": "2026-06-01",
   "access_type": "open-weight",
   "license": "modified-MIT",
   "significance": "cheap 1M-token context, native multimodal, modified-MIT",
   "organization_evidence": [
    "**Chinese AI labs have released four frontier-class open-weight models in roughly eight weeks,** marking a record-breaking cadence that significantly accelerates China\u2019s position in the global AI race.",
    "The Accelerated Pace Of China\u2019s AI Releases: Four Frontier Models In Record Time"
   ],
   "model_evidence": [
    "From late April to mid-June 2026, Chinese laboratories introduced four major open-weight models: **DeepSeek V4** on April 24, **MiniMax M3** on June 1, and **Kimi K2.7-Code** along with **GLM-5.2** in mid-June.",
    "The Accelerated Pace Of China\u2019s AI Releases: Four Frontier Models In Record Time"
   ],
   "release_date_evidence": [
    "From late April to mid-June 2026, Chinese laboratories introduced four major open-weight models: **DeepSeek V4** on April 24, **MiniMax M3** on June 1, and **Kimi K2.7-Code** along with **GLM-5.2** in mid-June.",
    "The Accelerated Pace Of China\u2019s AI Releases: Four Frontier Models In Record Time"
   ],
   "access_type_evidence": [
    "In just eight weeks, Chinese AI labs launched four frontier-class open-weight models, accelerating China\u2019s lead in open AI development.",
    "The Accelerated Pace Of China\u2019s AI Releases: Four Frontier Models In Record Time",
    "## Four Frontier-Class Open Models in Eight Weeks"
   ],
   "license_evidence": [
    "**MiniMax M3**cheap 1M-token context, native multimodal, modified-MIT",
    "### The production line \u2014 spring 2026",
    "The Accelerated Pace Of China\u2019s AI Releases: Four Frontier Models In Record Time"
   ],
   "significance_evidence": [
    "**MiniMax M3**cheap 1M-token context, native multimodal, modified-MIT",
    "### The production line \u2014 spring 2026",
    "The Accelerated Pace Of China\u2019s AI Releases: Four Frontier Models In Record Time"
   ]
  },
  {
   "url": "https://dreamridiculous.com/artificial-intelligence/the-accelerated-pace-of-china-s-ai-releases-four-frontier-models-in-record-time",
   "title": "The Accelerated Pace Of China\u2019s AI Releases: Four Frontier Models In Record Time - DreamRidiculous",
   "published_at": "2026-07-15T06:14:01",
   "organization": "Moonshot",
   "model": "Kimi K2.7-Code",
   "release_date": "2026-06-13",
   "access_type": "open-weight",
   "license": null,
   "significance": "agent-run specialist, ~30% fewer thinking tokens than K2.6",
   "organization_evidence": [
    "**Kimi K2.7-Code (Moonshot)**agent-run specialist, ~30% fewer thinking tokens than K2.6",
    "### The production line \u2014 spring 2026",
    "The Accelerated Pace Of China\u2019s AI Releases: Four Frontier Models In Record Time"
   ],
   "model_evidence": [
    "From late April to mid-June 2026, Chinese laboratories introduced four major open-weight models: **DeepSeek V4** on April 24, **MiniMax M3** on June 1, and **Kimi K2.7-Code** along with **GLM-5.2** in mid-June.",
    "The Accelerated Pace Of China\u2019s AI Releases: Four Frontier Models In Record Time"
   ],
   "release_date_evidence": [
    "**Kimi K2.7-Code (Moonshot)**agent-run specialist, ~30% fewer thinking tokens than K2.6",
    "### The production line \u2014 spring 2026",
    "The Accelerated Pace Of China\u2019s AI Releases: Four Frontier Models In Record Time"
   ],
   "access_type_evidence": [
    "In just eight weeks, Chinese AI labs launched four frontier-class open-weight models, accelerating China\u2019s lead in open AI development.",
    "The Accelerated Pace Of China\u2019s AI Releases: Four Frontier Models In Record Time",
    "## Four Frontier-Class Open Models in Eight Weeks"
   ],
   "license_evidence": null,
   "significance_evidence": [
    "**Kimi K2.7-Code (Moonshot)**agent-run specialist, ~30% fewer thinking tokens than K2.6",
    "### The production line \u2014 spring 2026",
    "The Accelerated Pace Of China\u2019s AI Releases: Four Frontier Models In Record Time"
   ]
  },
  {
   "url": "https://dreamridiculous.com/artificial-intelligence/the-accelerated-pace-of-china-s-ai-releases-four-frontier-models-in-record-time",
   "title": "The Accelerated Pace Of China\u2019s AI Releases: Four Frontier Models In Record Time - DreamRidiculous",
   "published_at": "2026-07-15T06:14:01",
   "organization": "Z.ai",
   "model": "GLM-5.2",
   "release_date": "2026-06-13",
   "access_type": "open-weight",
   "license": "MIT",
   "significance": "753B MoE, MIT, top open-weight on Artificial Analysis index",
   "organization_evidence": [
    "**GLM-5.2 (Z.ai)**753B MoE, MIT, top open-weight on Artificial Analysis index",
    "### The production line \u2014 spring 2026",
    "The Accelerated Pace Of China\u2019s AI Releases: Four Frontier Models In Record Time"
   ],
   "model_evidence": [
    "From late April to mid-June 2026, Chinese laboratories introduced four major open-weight models: **DeepSeek V4** on April 24, **MiniMax M3** on June 1, and **Kimi K2.7-Code** along with **GLM-5.2** in mid-June.",
    "The Accelerated Pace Of China\u2019s AI Releases: Four Frontier Models In Record Time"
   ],
   "release_date_evidence": [
    "**GLM-5.2 (Z.ai)**753B MoE, MIT, top open-weight on Artificial Analysis index",
    "### The production line \u2014 spring 2026",
    "The Accelerated Pace Of China\u2019s AI Releases: Four Frontier Models In Record Time"
   ],
   "access_type_evidence": [
    "In just eight weeks, Chinese AI labs launched four frontier-class open-weight models, accelerating China\u2019s lead in open AI development.",
    "The Accelerated Pace Of China\u2019s AI Releases: Four Frontier Models In Record Time",
    "## Four Frontier-Class Open Models in Eight Weeks"
   ],
   "license_evidence": [
    "**GLM-5.2 (Z.ai)**753B MoE, MIT, top open-weight on Artificial Analysis index",
    "### The production line \u2014 spring 2026",
    "The Accelerated Pace Of China\u2019s AI Releases: Four Frontier Models In Record Time"
   ],
   "significance_evidence": [
    "**GLM-5.2 (Z.ai)**753B MoE, MIT, top open-weight on Artificial Analysis index",
    "### The production line \u2014 spring 2026",
    "The Accelerated Pace Of China\u2019s AI Releases: Four Frontier Models In Record Time"
   ]
  },
  {
   "url": "https://aikendra.com/creator-hub/blog/inkling-thinking-machines-975b-open-weight-model",
   "title": "Mira Murati's Thinking Machines Releases Inkling, a 975B Open-Weight Model It Admits Isn't the Strongest",
   "published_at": "2026-07-16T00:00:00",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "a 975B open-weight multimodal MoE with 1M context, built for fine-tuning via Tinker rather than topping benchmarks",
   "organization_evidence": [
    "# Mira Murati's Thinking Machines Releases Inkling, a 975B Open-Weight Model It Admits Isn't the Strongest",
    "Thinking Machines Lab releases Inkling, a 975B open-weight multimodal MoE with 1M context, built for fine-tuning via Tinker rather than topping benchmarks."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "*July 16, 2026 \u00b7 ~5 min read*",
    "Thinking Machines Lab, the startup founded by former OpenAI CTO Mira Murati, released **Inkling** on Wednesday \u2014 its first in-house AI model and, unlike the flagships from OpenAI, Anthropic, or Google, an open-weight one that developers can download and modify directly, [TechCrunch reported](https://techcrunch.com/2026/07/15/thinking-machines-amps-up-its-bet-against-one-size-fits-all-ai-with-its-first-open-model-inkling/)."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Mira Murati's Thinking Machines Releases Inkling, a 975B Open-Weight Model It Admits Isn't the Strongest",
    "Thinking Machines Lab releases Inkling, a 975B open-weight multimodal MoE with 1M context, built for fine-tuning via Tinker rather than topping benchmarks."
   ]
  },
  {
   "url": "https://moclaw.ai/blog/what-is-kimi-k3",
   "title": "What Is Kimi K3? Moonshot's 2.8T Model",
   "published_at": "2026-07-17T11:45:19",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-27",
   "access_type": null,
   "license": null,
   "significance": "A ~2.8T-parameter MoE model with a 1M-token context window and native text, image, and video understanding, ranking as the effective #3 model family on the independent Artificial Analysis index (57.1).",
   "organization_evidence": [
    "| Developer | Moonshot AI |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "What Is Kimi K3? Moonshot's 2.8T Model\n\n14 min read \u00b7 \u00b7 Updated \u00b7 [ MoClaw Editorial ](/blog/authors/moclaw-editorial)\n\n![What Is Kimi K3? Moonshot's 2.8T Model](https://assets.moclaw.ai/cms/uploads/hero_what_is_kimi_k3_c154292771.png)\nWhat is Kimi K3? Moonshot's 2.8T-parameter flagship with a 1M context window, open weights since July 27, and the top spot on WebDev Arena.",
    "The full weights went public on July 27, 2026 under the bespoke Kimi K3 License, making it open-weight but not OSI open source."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "What Is Kimi K3? Moonshot's 2.8T Model\n\n14 min read \u00b7 \u00b7 Updated \u00b7 [ MoClaw Editorial ](/blog/authors/moclaw-editorial)\n\n![What Is Kimi K3? Moonshot's 2.8T Model](https://assets.moclaw.ai/cms/uploads/hero_what_is_kimi_k3_c154292771.png)\nWhat is Kimi K3? Moonshot's 2.8T-parameter flagship with a 1M context window, open weights since July 27, and the top spot on WebDev Arena.",
    "- Kimi K3 is a ~2.8T-parameter MoE model with a 1M-token context window and native text, image, and video understanding, launched July 16, 2026.",
    "- It ranks as the effective #3 model family on the independent Artificial Analysis index (57.1), the strongest open-weight result to date."
   ]
  },
  {
   "url": "https://techcrunch.com/2026/03/19/multiverse-computing-pushes-its-compressed-ai-models-into-the-mainstream",
   "title": "Multiverse Computing pushes its compressed AI models into the mainstream",
   "published_at": "2026-03-19T08:00:00",
   "organization": "Multiverse Computing",
   "model": null,
   "release_date": "2026-02-24",
   "access_type": null,
   "license": null,
   "significance": "delivers faster responses at lower cost than the original it was derived from",
   "organization_evidence": [
    "Its latest compressed model, [HyperNova 60B 2602](https://techcrunch.com/2026/02/24/spanish-soonicorn-multiverse-computing-releases-free-compressed-ai-model/), is built on gpt-oss-120b \u2014 an OpenAI model whose underlying code is publicly available."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Its latest compressed model, [HyperNova 60B 2602](https://techcrunch.com/2026/02/24/spanish-soonicorn-multiverse-computing-releases-free-compressed-ai-model/), is built on gpt-oss-120b \u2014 an OpenAI model whose underlying code is publicly available."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The company claims it now delivers [faster responses](https://multiversecomputing.com/papers/hypernova-60b-2602-same-intelligence-half-the-size-improved-tool-calling-capability) at lower cost than the original it was derived from, an advantage that matters particularly for agentic coding workflows, where AI autonomously completes complex, multistep programming tasks."
   ]
  },
  {
   "url": "https://techcrunch.com/2026/03/19/multiverse-computing-pushes-its-compressed-ai-models-into-the-mainstream",
   "title": "Multiverse Computing pushes its compressed AI models into the mainstream",
   "published_at": "2026-03-19T08:00:00",
   "organization": "Mistral AI",
   "model": null,
   "release_date": "2026-03",
   "access_type": null,
   "license": null,
   "significance": "simultaneously optimized for general chat, coding, agentic tasks, and reasoning",
   "organization_evidence": [
    "Earlier this week, Mistral updated its small model family with the [launch of Mistral Small 4](https://mistral.ai/fr/news/mistral-small-4), which it says is simultaneously optimized for general chat, coding, agentic tasks, and reasoning."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Earlier this week, Mistral updated its small model family with the [launch of Mistral Small 4](https://mistral.ai/fr/news/mistral-small-4), which it says is simultaneously optimized for general chat, coding, agentic tasks, and reasoning."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Earlier this week, Mistral updated its small model family with the [launch of Mistral Small 4](https://mistral.ai/fr/news/mistral-small-4), which it says is simultaneously optimized for general chat, coding, agentic tasks, and reasoning."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/gemma-4-26b-a4b-vs-qwen3-max",
   "title": "Gemma 4 26B A4B (Reasoning) vs Qwen3 Max: Model Comparison",
   "published_at": null,
   "organization": "Google",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "A reasoning model with 256k tokens context window and image input support",
   "organization_evidence": [
    "# Gemma 4 26B A4B (Reasoning) vs. Qwen3 Max",
    "| Creator | ![Google](/img/logos/google_small.svg)[](https://deepmind.google/technologies/gemini/) | ![Alibaba](/img/logos/alibaba_small.svg)[](https://www.alibabacloud.com/) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Gemma 4 26B A4B (Reasoning) vs. Qwen3 Max",
    "| Release Date | April, 2026 | September, 2025 | Gemma 4 26B A4B (Reasoning) has a more recent release date than Qwen3 Max |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Gemma 4 26B A4B (Reasoning) vs. Qwen3 Max",
    "| Context Window | 256k tokens (~384 A4 pages of size 12 Arial font) | 262k tokens (~393 A4 pages of size 12 Arial font) | Gemma 4 26B A4B (Reasoning) is smaller than Qwen3 Max |",
    "| Image Input Support | Yes | No | Gemma 4 26B A4B (Reasoning) has image input support while Qwen3 Max does not |"
   ]
  },
  {
   "url": "https://startupfortune.com/china-is-giving-its-best-ai-models-away-free-and-that-should-worry-openai",
   "title": "China is giving its best AI models away free and that should worry OpenAI",
   "published_at": "2026-07-28T04:16:48",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-17",
   "access_type": null,
   "license": null,
   "significance": "a 2.8 trillion-parameter mixture-of-experts model",
   "organization_evidence": [
    "On July 17, Moonshot AI released Kimi K3, a 2.8 trillion-parameter mixture-of-experts model, with open weights free to download five days later."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "On July 17, Moonshot AI released Kimi K3, a 2.8 trillion-parameter mixture-of-experts model, with open weights free to download five days later."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "On July 17, Moonshot AI released Kimi K3, a 2.8 trillion-parameter mixture-of-experts model, with open weights free to download five days later."
   ]
  },
  {
   "url": "https://startupfortune.com/china-is-giving-its-best-ai-models-away-free-and-that-should-worry-openai",
   "title": "China is giving its best AI models away free and that should worry OpenAI",
   "published_at": "2026-07-28T04:16:48",
   "organization": "Alibaba",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "a 2.4 trillion-parameter model the company says performs second only to Anthropic's Claude Fable 5",
   "organization_evidence": [
    "That same week, Alibaba previewed Qwen 3.8 Max."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "It's a 2.4 trillion-parameter model the company says performs second only to Anthropic's Claude Fable 5."
   ]
  },
  {
   "url": "https://cryptovolix.com/muratis-inkling-debuts-as-best-western-open-source-ai-model-but-at-a-price",
   "title": "Inkling Review: Murati's Open-Source AI Model Tested",
   "published_at": "2026-07-26T15:12:18",
   "organization": "Thinking Machines Lab",
   "model": "Inkling",
   "release_date": "2026-07-15",
   "access_type": "open-weights",
   "license": "Apache 2.0",
   "significance": "the strongest open-weights model built by a Western lab to date",
   "organization_evidence": [
    "Blockchain",
    "Murati's Inkling Debuts as Best Western Open-Source AI Model, But at a Price",
    "Thinking Machines Lab released Inkling on July 15, its first foundation model, and put the full weights on Hugging Face under a permissive Apache 2.0 license."
   ],
   "model_evidence": [
    "Blockchain",
    "Murati's Inkling Debuts as Best Western Open-Source AI Model, But at a Price",
    "Thinking Machines Lab released Inkling on July 15, its first foundation model, and put the full weights on Hugging Face under a permissive Apache 2.0 license."
   ],
   "release_date_evidence": [
    "Blockchain",
    "Murati's Inkling Debuts as Best Western Open-Source AI Model, But at a Price",
    "Thinking Machines Lab released Inkling on July 15, its first foundation model, and put the full weights on Hugging Face under a permissive Apache 2.0 license."
   ],
   "access_type_evidence": [
    "Blockchain",
    "Murati's Inkling Debuts as Best Western Open-Source AI Model, But at a Price",
    "Thinking Machines Lab released Inkling on July 15, its first foundation model, and put the full weights on Hugging Face under a permissive Apache 2.0 license."
   ],
   "license_evidence": [
    "Blockchain",
    "Murati's Inkling Debuts as Best Western Open-Source AI Model, But at a Price",
    "Thinking Machines Lab released Inkling on July 15, its first foundation model, and put the full weights on Hugging Face under a permissive Apache 2.0 license."
   ],
   "significance_evidence": [
    "Blockchain",
    "Murati's Inkling Debuts as Best Western Open-Source AI Model, But at a Price",
    "The numbers back up an unusually bold claim: Inkling is the strongest open-weights model built by a Western lab to date."
   ]
  },
  {
   "url": "https://cloudrift.ai/blog/state-of-open-coding-models-2026",
   "title": "The State of Open Coding AI Models in July 2026",
   "published_at": "2026-07-31T00:00:00",
   "organization": "Poolside",
   "model": null,
   "release_date": "2026-07-21",
   "access_type": null,
   "license": null,
   "significance": "118B / 8B model scoring 70.2 on Terminal-Bench 2.1 with 1M context",
   "organization_evidence": [
    "## Recent open coding models (2026)",
    "| Model (lab) | Shipped | Size total / active | License | Runs on | Coding angle |",
    "| **Poolside Laguna S 2.1** | 2026-07-21 | 118B / 8B | OpenMDW-1.1 | single NVIDIA DGX Spark | Western open coder; Poolside reports Terminal-Bench 2.1 70.2 (thinking on), roughly 10 to 12 points behind the closed frontier. 1M context. |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## Recent open coding models (2026)",
    "| Model (lab) | Shipped | Size total / active | License | Runs on | Coding angle |",
    "| **Poolside Laguna S 2.1** | 2026-07-21 | 118B / 8B | OpenMDW-1.1 | single NVIDIA DGX Spark | Western open coder; Poolside reports Terminal-Bench 2.1 70.2 (thinking on), roughly 10 to 12 points behind the closed frontier. 1M context. |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## Recent open coding models (2026)",
    "| Model (lab) | Shipped | Size total / active | License | Runs on | Coding angle |",
    "| **Poolside Laguna S 2.1** | 2026-07-21 | 118B / 8B | OpenMDW-1.1 | single NVIDIA DGX Spark | Western open coder; Poolside reports Terminal-Bench 2.1 70.2 (thinking on), roughly 10 to 12 points behind the closed frontier. 1M context. |"
   ]
  },
  {
   "url": "https://laura-martel.com/blog/kimi-k3",
   "title": "Kimi K3: The Largest Open-Weight Model Yet",
   "published_at": "2026-07-16T00:00:00",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-26",
   "access_type": null,
   "license": null,
   "significance": "A 2.8-trillion-parameter Mixture-of-Experts model that the company says is now the largest open-weight model in the world.",
   "organization_evidence": [
    "# Kimi K3: The Largest Open-Weight Model Yet",
    "Moonshot AI announced **Kimi K3** on July 16, 2026 \u2014 a 2.8-trillion-parameter Mixture-of-Experts model that the company says is now the largest open-weight model in the world."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Kimi K3: The Largest Open-Weight Model Yet",
    "Moonshot AI announced **Kimi K3** on July 16, 2026 \u2014 a 2.8-trillion-parameter Mixture-of-Experts model that the company says is now the largest open-weight model in the world.",
    "The weights went up on July 26 rather than the promised 27th, under the **Kimi K3 License** \u2014 MIT-style permissive terms, with a separate agreement required only if you run a Model-as-a-Service business above $20M revenue over any twelve months."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Kimi K3: The Largest Open-Weight Model Yet",
    "Moonshot AI announced **Kimi K3** on July 16, 2026 \u2014 a 2.8-trillion-parameter Mixture-of-Experts model that the company says is now the largest open-weight model in the world."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/glm-4-7-flash",
   "title": "GLM-4.7-Flash - Intelligence, Performance & Price Analysis",
   "published_at": null,
   "organization": "Z AI",
   "model": null,
   "release_date": "2026-01",
   "access_type": null,
   "license": null,
   "significance": null,
   "organization_evidence": [
    "Z AI has launched a newer model, [GLM-5](/models/glm-5).",
    "![GLM-4.7-Flash (Reasoning) logo](/img/logos/zai_small.svg)"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "\u2022\nReleased January 2026",
    "# GLM-4.7-Flash (Reasoning) Intelligence, Performance & Price Analysis"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": null
  },
  {
   "url": "https://chatforest.com/builders-log/longcat-2-meituan-owl-alpha-reveal-chinese-chips-open-source-coding-model",
   "title": "LongCat-2.0: Meituan's 1.6T Coding Model Was Topping OpenRouter as 'Owl Alpha' All Along",
   "published_at": "2026-07-03T15:00:00",
   "organization": "Meituan",
   "model": null,
   "release_date": "2026-06-30",
   "access_type": null,
   "license": null,
   "significance": "1.6-trillion-parameter Mixture-of-Experts coding model, trained entirely on domestic Chinese semiconductors",
   "organization_evidence": [
    "LongCat-2.0: Meituan's 1.6T Coding Model Was Topping OpenRouter as 'Owl Alpha' All Along",
    "On June 30, 2026, Chinese food-delivery giant Meituan stepped forward and revealed that [Owl Alpha](https://decrypt.co/372579/longcat-2-0-meituan-ai-stealth-model-openrouter) \u2014 the anonymous model quietly topping OpenRouter\u2019s global charts for two months \u2014 was theirs all along: [LongCat-2.0](https://www.longcatai.org/news/longcat-2), a 1.6-trillion-parameter Mixture-of-Experts coding model, trained entirely on domestic Chinese semiconductors, released under the MIT license."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "On June 30, 2026, Chinese food-delivery giant Meituan stepped forward and revealed that [Owl Alpha](https://decrypt.co/372579/longcat-2-0-meituan-ai-stealth-model-openrouter) \u2014 the anonymous model quietly topping OpenRouter\u2019s global charts for two months \u2014 was theirs all along: [LongCat-2.0](https://www.longcatai.org/news/longcat-2), a 1.6-trillion-parameter Mixture-of-Experts coding model, trained entirely on domestic Chinese semiconductors, released under the MIT license.",
    "*LongCat-2.0 released June 30, 2026. Article published July 4, 2026. Written by Grove, an AI agent.*"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "LongCat-2.0: Meituan's 1.6T Coding Model Was Topping OpenRouter as 'Owl Alpha' All Along",
    "On June 30, 2026, Chinese food-delivery giant Meituan stepped forward and revealed that [Owl Alpha](https://decrypt.co/372579/longcat-2-0-meituan-ai-stealth-model-openrouter) \u2014 the anonymous model quietly topping OpenRouter\u2019s global charts for two months \u2014 was theirs all along: [LongCat-2.0](https://www.longcatai.org/news/longcat-2), a 1.6-trillion-parameter Mixture-of-Experts coding model, trained entirely on domestic Chinese semiconductors, released under the MIT license."
   ]
  },
  {
   "url": "https://aiautomationglobal.com/blog/thinking-machines-inkling-open-weight-ai-model-2026",
   "title": "Inkling: Thinking Machines Enters the Open-Weight AI Race",
   "published_at": "2026-07-19T00:00:00",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "975-billion-parameter Mixture-of-Experts (MoE) foundation model with native multimodal reasoning and 1 million token context window",
   "organization_evidence": [
    "**Inkling** is the first open-weight foundation model released by Thinking Machines Lab, the AI startup founded by former OpenAI CTO Mira Murati."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Launched on July 15, 2026 under an Apache 2.0 license, Inkling is a 975-billion-parameter Mixture-of-Experts (MoE) \u2014 an architecture where only a small subset of specialized sub-networks activate per query \u2014 model that draws on just 41 billion active parameters per task."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Launched on July 15, 2026 under an Apache 2.0 license, Inkling is a 975-billion-parameter Mixture-of-Experts (MoE) \u2014 an architecture where only a small subset of specialized sub-networks activate per query \u2014 model that draws on just 41 billion active parameters per task.",
    "Inkling was trained on 45 trillion tokens spanning text, images, audio, and video, giving it native multimodal reasoning even though its current public release outputs text, code, and structured data only.",
    "It supports a context window of up to 1 million tokens and lets users dial a \u201cthinking effort\u201d slider up or down to trade reasoning depth for speed \u2014 a practical knob for teams balancing latency against accuracy in production."
   ]
  },
  {
   "url": "https://irenictech.com/blog/best-open-source-llms-2026",
   "title": "The Best Open-Source LLMs in 2026: GLM, DeepSeek, Qwen, Kimi, and MiniMax Compared",
   "published_at": "2026-06-28T00:00:00",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "Coding, strong all-round",
   "organization_evidence": [
    "The Best Open-Source LLMs in 2026: GLM, DeepSeek, Qwen, Kimi, and MiniMax Compared",
    "Here's the current field, with the facts that don't churn week to week.",
    "| DeepSeek V4-Pro | DeepSeek | V4-Pro, April 2026 | ~1.6T MoE, ~49B active | Permissive (MIT-class) | Coding, strong all-round |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The Best Open-Source LLMs in 2026: GLM, DeepSeek, Qwen, Kimi, and MiniMax Compared",
    "Here's the current field, with the facts that don't churn week to week.",
    "| DeepSeek V4-Pro | DeepSeek | V4-Pro, April 2026 | ~1.6T MoE, ~49B active | Permissive (MIT-class) | Coding, strong all-round |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The Best Open-Source LLMs in 2026: GLM, DeepSeek, Qwen, Kimi, and MiniMax Compared",
    "Here's the current field, with the facts that don't churn week to week.",
    "| DeepSeek V4-Pro | DeepSeek | V4-Pro, April 2026 | ~1.6T MoE, ~49B active | Permissive (MIT-class) | Coding, strong all-round |"
   ]
  },
  {
   "url": "https://irenictech.com/blog/best-open-source-llms-2026",
   "title": "The Best Open-Source LLMs in 2026: GLM, DeepSeek, Qwen, Kimi, and MiniMax Compared",
   "published_at": "2026-06-28T00:00:00",
   "organization": "Z.ai (Zhipu)",
   "model": null,
   "release_date": "2026-06",
   "access_type": null,
   "license": null,
   "significance": "Long-horizon agentic work, coding",
   "organization_evidence": [
    "The Best Open-Source LLMs in 2026: GLM, DeepSeek, Qwen, Kimi, and MiniMax Compared",
    "Here's the current field, with the facts that don't churn week to week.",
    "| GLM-5.2 | Z.ai (Zhipu) | 5.2, June 2026 | ~753B MoE, 1M context | MIT | Long-horizon agentic work, coding |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The Best Open-Source LLMs in 2026: GLM, DeepSeek, Qwen, Kimi, and MiniMax Compared",
    "Here's the current field, with the facts that don't churn week to week.",
    "| GLM-5.2 | Z.ai (Zhipu) | 5.2, June 2026 | ~753B MoE, 1M context | MIT | Long-horizon agentic work, coding |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The Best Open-Source LLMs in 2026: GLM, DeepSeek, Qwen, Kimi, and MiniMax Compared",
    "Here's the current field, with the facts that don't churn week to week.",
    "| GLM-5.2 | Z.ai (Zhipu) | 5.2, June 2026 | ~753B MoE, 1M context | MIT | Long-horizon agentic work, coding |"
   ]
  },
  {
   "url": "https://irenictech.com/blog/best-open-source-llms-2026",
   "title": "The Best Open-Source LLMs in 2026: GLM, DeepSeek, Qwen, Kimi, and MiniMax Compared",
   "published_at": "2026-06-28T00:00:00",
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-02",
   "access_type": null,
   "license": null,
   "significance": "Capable, efficient generalist",
   "organization_evidence": [
    "The Best Open-Source LLMs in 2026: GLM, DeepSeek, Qwen, Kimi, and MiniMax Compared",
    "Here's the current field, with the facts that don't churn week to week.",
    "| Qwen3.5 | Alibaba | 3.5, February 2026 | ~397B MoE, ~17B active | Apache-class | Capable, efficient generalist |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The Best Open-Source LLMs in 2026: GLM, DeepSeek, Qwen, Kimi, and MiniMax Compared",
    "Here's the current field, with the facts that don't churn week to week.",
    "| Qwen3.5 | Alibaba | 3.5, February 2026 | ~397B MoE, ~17B active | Apache-class | Capable, efficient generalist |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The Best Open-Source LLMs in 2026: GLM, DeepSeek, Qwen, Kimi, and MiniMax Compared",
    "Here's the current field, with the facts that don't churn week to week.",
    "| Qwen3.5 | Alibaba | 3.5, February 2026 | ~397B MoE, ~17B active | Apache-class | Capable, efficient generalist |"
   ]
  },
  {
   "url": "https://irenictech.com/blog/best-open-source-llms-2026",
   "title": "The Best Open-Source LLMs in 2026: GLM, DeepSeek, Qwen, Kimi, and MiniMax Compared",
   "published_at": "2026-06-28T00:00:00",
   "organization": "Moonshot",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Coding and agentic generalist",
   "organization_evidence": [
    "The Best Open-Source LLMs in 2026: GLM, DeepSeek, Qwen, Kimi, and MiniMax Compared",
    "Here's the current field, with the facts that don't churn week to week.",
    "| Kimi K2.6 | Moonshot | K2.6 (K2.7-Code newer) | ~1T MoE, long context | Modified MIT | Coding and agentic generalist |"
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The Best Open-Source LLMs in 2026: GLM, DeepSeek, Qwen, Kimi, and MiniMax Compared",
    "Here's the current field, with the facts that don't churn week to week.",
    "| Kimi K2.6 | Moonshot | K2.6 (K2.7-Code newer) | ~1T MoE, long context | Modified MIT | Coding and agentic generalist |"
   ]
  },
  {
   "url": "https://irenictech.com/blog/best-open-source-llms-2026",
   "title": "The Best Open-Source LLMs in 2026: GLM, DeepSeek, Qwen, Kimi, and MiniMax Compared",
   "published_at": "2026-06-28T00:00:00",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-06",
   "access_type": null,
   "license": null,
   "significance": "Reasoning, multimodal, long context",
   "organization_evidence": [
    "The Best Open-Source LLMs in 2026: GLM, DeepSeek, Qwen, Kimi, and MiniMax Compared",
    "Here's the current field, with the facts that don't churn week to week.",
    "| MiniMax M3 | MiniMax | M3, June 2026 | Sparse attention, 1M context | Verify before use | Reasoning, multimodal, long context |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The Best Open-Source LLMs in 2026: GLM, DeepSeek, Qwen, Kimi, and MiniMax Compared",
    "Here's the current field, with the facts that don't churn week to week.",
    "| MiniMax M3 | MiniMax | M3, June 2026 | Sparse attention, 1M context | Verify before use | Reasoning, multimodal, long context |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The Best Open-Source LLMs in 2026: GLM, DeepSeek, Qwen, Kimi, and MiniMax Compared",
    "Here's the current field, with the facts that don't churn week to week.",
    "| MiniMax M3 | MiniMax | M3, June 2026 | Sparse attention, 1M context | Verify before use | Reasoning, multimodal, long context |"
   ]
  },
  {
   "url": "https://aidb.digital/blog/2026-07-18-moonshot-ai-releases-kimi-k3-the-largest-open-weight-model-ever-at-2-",
   "title": "Moonshot AI Releases Kimi K3, the Largest Open-Weight Model Ever, at 2.8 Trillion Parameters \u00b7 AIDB",
   "published_at": "2026-07-18T00:00:00",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-16",
   "access_type": null,
   "license": null,
   "significance": "a 2.8-trillion-parameter model that the company says is now the largest open-source AI model in the world",
   "organization_evidence": [
    "## What Was Released",
    "Moonshot AI, the Beijing-based artificial intelligence startup backed by Alibaba, on Thursday released Kimi K3 \u2014 a 2.8-trillion-parameter model that the company says is now the largest open-source AI model in the world, and one that benchmarks show performs neck-and-neck with the most powerful proprietary systems from Anthropic and OpenAI."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## Scale in Context",
    "Moonshot AI publicly released Kimi K3 on July 16, 2026, with full open-source weights promised by July 27."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## What Was Released",
    "Moonshot AI, the Beijing-based artificial intelligence startup backed by Alibaba, on Thursday released Kimi K3 \u2014 a 2.8-trillion-parameter model that the company says is now the largest open-source AI model in the world, and one that benchmarks show performs neck-and-neck with the most powerful proprietary systems from Anthropic and OpenAI."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/gemma-4-31b-vs-claude-sonnet-4-6-non-reasoning-low-effort",
   "title": "Gemma 4 31B (Reasoning) vs Claude Sonnet 4.6 (Non-reasoning, Low Effort): Model Comparison | Artificial Analysis",
   "published_at": null,
   "organization": "Google",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "A 31B model with reasoning capabilities and a 256k context window.",
   "organization_evidence": [
    "# Gemma 4 31B (Reasoning) vs. Claude Sonnet 4.6 (Non-reasoning, Low Effort)",
    "| | ![Google logo](/img/logos/google_small.svg)Google Gemma 4 31B (Reasoning) | ![Anthropic logo](/img/logos/anthropic_small.svg)Anthropic Claude Sonnet 4.6 (Non-reasoning, Low Effort) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Gemma 4 31B (Reasoning) vs. Claude Sonnet 4.6 (Non-reasoning, Low Effort)",
    "| Release Date | April 2026 | February 2026 | Gemma 4 31B (Reasoning) has a more recent release date than Claude Sonnet 4.6 (Non-reasoning, Low Effort) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Gemma 4 31B (Reasoning) vs. Claude Sonnet 4.6 (Non-reasoning, Low Effort)",
    "| Context Window | 256k tokens~384 A4 pages of size 12 Arial font | 1000k tokens~1,500 A4 pages of size 12 Arial font | Claude Sonnet 4.6 (Non-reasoning, Low Effort) has a larger context window than Gemma 4 31B (Reasoning) |",
    "| Reasoning | Yes | No | Gemma 4 31B (Reasoning) has reasoning while Claude Sonnet 4.6 (Non-reasoning, Low Effort) does not |"
   ]
  },
  {
   "url": "https://tech-insider.org/ca/nvidia-nemotron-3-ultra-2026",
   "title": "NVIDIA Nemotron 3 Ultra: 550B Open AI Trails China [2026]",
   "published_at": "2026-07-06T00:32:32",
   "organization": "NVIDIA",
   "model": null,
   "release_date": "2026-06-04",
   "access_type": null,
   "license": null,
   "significance": "A ~550-billion-parameter mixture-of-experts model supporting up to 1 million tokens context window",
   "organization_evidence": [
    "On June 4, 2026, NVIDIA quietly uploaded the weights for **NVIDIA Nemotron 3 Ultra** to Hugging Face \u2013 days after Jensen Huang teased the model from the Computex keynote stage in Taipei.",
    "## NVIDIA Nemotron 3 Ultra: What Launched on June 4"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "On June 4, 2026, NVIDIA quietly uploaded the weights for **NVIDIA Nemotron 3 Ultra** to Hugging Face \u2013 days after Jensen Huang teased the model from the Computex keynote stage in Taipei.",
    "## NVIDIA Nemotron 3 Ultra: What Launched on June 4"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Ultra is the capstone: a ~550-billion-parameter mixture-of-experts model that Huang first surfaced at Computex before NVIDIA dropped the weights to Hugging Face and its own [build.nvidia.com](https://build.nvidia.com/) catalog on June 4.",
    "## NVIDIA Nemotron 3 Ultra: What Launched on June 4"
   ]
  },
  {
   "url": "https://tech-insider.org/ca/nvidia-nemotron-3-ultra-2026",
   "title": "NVIDIA Nemotron 3 Ultra: 550B Open AI Trails China [2026]",
   "published_at": "2026-07-06T00:32:32",
   "organization": "NVIDIA",
   "model": null,
   "release_date": "2026-03-11",
   "access_type": null,
   "license": null,
   "significance": "A 120.6B parameter model in the Nemotron 3 family",
   "organization_evidence": [
    "Nemotron 3 Super (120.6B) followed at GTC on March 11, 2026.",
    "## NVIDIA Nemotron 3 Ultra: What Launched on June 4"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Nemotron 3 Super (120.6B) followed at GTC on March 11, 2026.",
    "## NVIDIA Nemotron 3 Ultra: What Launched on June 4"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Nemotron 3 Super (120.6B) followed at GTC on March 11, 2026.",
    "## NVIDIA Nemotron 3 Ultra: What Launched on June 4"
   ]
  },
  {
   "url": "http://ai-tldr.dev/models/inkling-small",
   "title": "Inkling-Small \u2014 276B/12B MoE, 1M Context, Apache-2.0 | AI/TLDR",
   "published_at": "2026-07-30T00:00:00",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-30",
   "access_type": null,
   "license": null,
   "significance": "sparse Mixture-of-Experts transformer with 276 billion total parameters and 12 billion active per token \u2014 about a quarter of Inkling's size \u2014 and Thinking Machines says it matches or exceeds Inkling on reasoning and agentic tasks",
   "organization_evidence": [
    "# Inkling-Small",
    "Inkling-Small is Thinking Machines Lab's efficient sibling to Inkling, released on July 30, 2026 as an Apache-2.0 open-weights model."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Inkling-Small",
    "A quarter of Inkling's size, comparable performance, and 3x cheaper output \u2014 276B / 12B-active Apache-2.0 MoE released July 30, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Inkling-Small",
    "It is a sparse Mixture-of-Experts transformer with 276 billion total parameters and 12 billion active per token \u2014 about a quarter of Inkling's size \u2014 and Thinking Machines says it \"matches or exceeds Inkling on reasoning and agentic tasks\" while trailing on knowledge coverage and factuality."
   ]
  },
  {
   "url": "https://qbitneural.com/open-source-llms",
   "title": "Best Open Source LLMs in 2026 (Updated for Q3)",
   "published_at": "2026-07-13T11:45:03",
   "organization": "Meituan",
   "model": null,
   "release_date": "2026-01",
   "access_type": null,
   "license": null,
   "significance": "a 560B-class MoE reasoning model focused on environment scaling, multi-environment reinforcement learning, tool use, search, and robustness",
   "organization_evidence": [
    "# Best Open Source Large Language Models (LLMs)",
    "[Meituan\u2019s January 2026 LongCat](https://arxiv.org/abs/2601.16725) update is a 560B-class MoE reasoning model."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Best Open Source Large Language Models (LLMs)",
    "## **LongCat-Flash-Thinking-2601**",
    "[Meituan\u2019s January 2026 LongCat](https://arxiv.org/abs/2601.16725) update is a 560B-class MoE reasoning model."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Best Open Source Large Language Models (LLMs)",
    "## **LongCat-Flash-Thinking-2601**",
    "[Meituan\u2019s January 2026 LongCat](https://arxiv.org/abs/2601.16725) update is a 560B-class MoE reasoning model. Its training focuses on environment scaling, multi-environment reinforcement learning, tool use, search, and robustness to imperfect results from tools or environments."
   ]
  },
  {
   "url": "https://dreamridiculous.com/artificial-intelligence/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - DreamRidiculous",
   "published_at": "2026-07-15T10:56:40",
   "organization": "DeepSeek",
   "model": "DeepSeek V4",
   "release_date": "2026-04-24",
   "access_type": "open-weight",
   "license": "MIT",
   "significance": "1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor",
   "organization_evidence": [
    "Chinese labs such as DeepSeek, Z.ai, Moonshot, and Alibaba now each have distinct strategies, from price leadership to long-horizon stability and broad self-hosting options.",
    "Four Frontier-Class Open Models in Eight Weeks",
    "The production line \u2014 spring 2026"
   ],
   "model_evidence": [
    "Between late April and mid-June 2026, Chinese laboratories launched four major open-weight AI models: **DeepSeek V4** on April 24, **MiniMax M3** on June 1, and **Kimi K2.7-Code** and **GLM-5.2** within days of each other in mid-June.",
    "Four Frontier-Class Open Models in Eight Weeks",
    "The production line \u2014 spring 2026"
   ],
   "release_date_evidence": [
    "Between late April and mid-June 2026, Chinese laboratories launched four major open-weight AI models: **DeepSeek V4** on April 24, **MiniMax M3** on June 1, and **Kimi K2.7-Code** and **GLM-5.2** within days of each other in mid-June.",
    "Four Frontier-Class Open Models in Eight Weeks",
    "The production line \u2014 spring 2026"
   ],
   "access_type_evidence": [
    "Between late April and mid-June 2026, Chinese laboratories launched four major open-weight AI models: **DeepSeek V4** on April 24, **MiniMax M3** on June 1, and **Kimi K2.7-Code** and **GLM-5.2** within days of each other in mid-June.",
    "Four Frontier-Class Open Models in Eight Weeks",
    "The production line \u2014 spring 2026"
   ],
   "license_evidence": [
    "All four models are downloadable, with most under permissive licenses like MIT, and are priced far below Western frontier APIs when hosted independently.",
    "Four Frontier-Class Open Models in Eight Weeks",
    "The production line \u2014 spring 2026"
   ],
   "significance_evidence": [
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor",
    "Four Frontier-Class Open Models in Eight Weeks",
    "The production line \u2014 spring 2026"
   ]
  },
  {
   "url": "https://dreamridiculous.com/artificial-intelligence/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - DreamRidiculous",
   "published_at": "2026-07-15T10:56:40",
   "organization": "MiniMax",
   "model": "MiniMax M3",
   "release_date": "2026-06-01",
   "access_type": "open-weight",
   "license": "modified-MIT",
   "significance": "cheap 1M-token context, native multimodal",
   "organization_evidence": [
    "**Chinese AI labs have released four frontier-class open models in approximately eight weeks,** including DeepSeek V4, MiniMax M3, Kimi K2.7-Code, and GLM-5.2.",
    "Four Frontier-Class Open Models in Eight Weeks",
    "The production line \u2014 spring 2026"
   ],
   "model_evidence": [
    "Between late April and mid-June 2026, Chinese laboratories launched four major open-weight AI models: **DeepSeek V4** on April 24, **MiniMax M3** on June 1, and **Kimi K2.7-Code** and **GLM-5.2** within days of each other in mid-June.",
    "Four Frontier-Class Open Models in Eight Weeks",
    "The production line \u2014 spring 2026"
   ],
   "release_date_evidence": [
    "Between late April and mid-June 2026, Chinese laboratories launched four major open-weight AI models: **DeepSeek V4** on April 24, **MiniMax M3** on June 1, and **Kimi K2.7-Code** and **GLM-5.2** within days of each other in mid-June.",
    "Four Frontier-Class Open Models in Eight Weeks",
    "The production line \u2014 spring 2026"
   ],
   "access_type_evidence": [
    "Between late April and mid-June 2026, Chinese laboratories launched four major open-weight AI models: **DeepSeek V4** on April 24, **MiniMax M3** on June 1, and **Kimi K2.7-Code** and **GLM-5.2** within days of each other in mid-June.",
    "Four Frontier-Class Open Models in Eight Weeks",
    "The production line \u2014 spring 2026"
   ],
   "license_evidence": [
    "**MiniMax M3**cheap 1M-token context, native multimodal, modified-MIT",
    "Four Frontier-Class Open Models in Eight Weeks",
    "The production line \u2014 spring 2026"
   ],
   "significance_evidence": [
    "**MiniMax M3**cheap 1M-token context, native multimodal, modified-MIT",
    "Four Frontier-Class Open Models in Eight Weeks",
    "The production line \u2014 spring 2026"
   ]
  },
  {
   "url": "https://dreamridiculous.com/artificial-intelligence/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - DreamRidiculous",
   "published_at": "2026-07-15T10:56:40",
   "organization": "Moonshot",
   "model": "Kimi K2.7-Code",
   "release_date": "2026-06-13",
   "access_type": "open-weight",
   "license": null,
   "significance": "agent-run specialist, ~30% fewer thinking tokens than K2.6",
   "organization_evidence": [
    "**Kimi K2.7-Code (Moonshot)**agent-run specialist, ~30% fewer thinking tokens than K2.6",
    "Four Frontier-Class Open Models in Eight Weeks",
    "The production line \u2014 spring 2026"
   ],
   "model_evidence": [
    "Between late April and mid-June 2026, Chinese laboratories launched four major open-weight AI models: **DeepSeek V4** on April 24, **MiniMax M3** on June 1, and **Kimi K2.7-Code** and **GLM-5.2** within days of each other in mid-June.",
    "Four Frontier-Class Open Models in Eight Weeks",
    "The production line \u2014 spring 2026"
   ],
   "release_date_evidence": [
    "Between late April and mid-June 2026, Chinese laboratories launched four major open-weight AI models: **DeepSeek V4** on April 24, **MiniMax M3** on June 1, and **Kimi K2.7-Code** and **GLM-5.2** within days of each other in mid-June.",
    "Four Frontier-Class Open Models in Eight Weeks",
    "The production line \u2014 spring 2026"
   ],
   "access_type_evidence": [
    "Between late April and mid-June 2026, Chinese laboratories launched four major open-weight AI models: **DeepSeek V4** on April 24, **MiniMax M3** on June 1, and **Kimi K2.7-Code** and **GLM-5.2** within days of each other in mid-June.",
    "Four Frontier-Class Open Models in Eight Weeks",
    "The production line \u2014 spring 2026"
   ],
   "license_evidence": null,
   "significance_evidence": [
    "**Kimi K2.7-Code (Moonshot)**agent-run specialist, ~30% fewer thinking tokens than K2.6",
    "Four Frontier-Class Open Models in Eight Weeks",
    "The production line \u2014 spring 2026"
   ]
  },
  {
   "url": "https://dreamridiculous.com/artificial-intelligence/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - DreamRidiculous",
   "published_at": "2026-07-15T10:56:40",
   "organization": "Z.ai",
   "model": "GLM-5.2",
   "release_date": "2026-06-13",
   "access_type": "open-weight",
   "license": "MIT",
   "significance": "753B MoE, top open-weight on Artificial Analysis index",
   "organization_evidence": [
    "**GLM-5.2 (Z.ai)**753B MoE, MIT, top open-weight on Artificial Analysis index",
    "Four Frontier-Class Open Models in Eight Weeks",
    "The production line \u2014 spring 2026"
   ],
   "model_evidence": [
    "Between late April and mid-June 2026, Chinese laboratories launched four major open-weight AI models: **DeepSeek V4** on April 24, **MiniMax M3** on June 1, and **Kimi K2.7-Code** and **GLM-5.2** within days of each other in mid-June.",
    "Four Frontier-Class Open Models in Eight Weeks",
    "The production line \u2014 spring 2026"
   ],
   "release_date_evidence": [
    "Between late April and mid-June 2026, Chinese laboratories launched four major open-weight AI models: **DeepSeek V4** on April 24, **MiniMax M3** on June 1, and **Kimi K2.7-Code** and **GLM-5.2** within days of each other in mid-June.",
    "Four Frontier-Class Open Models in Eight Weeks",
    "The production line \u2014 spring 2026"
   ],
   "access_type_evidence": [
    "Between late April and mid-June 2026, Chinese laboratories launched four major open-weight AI models: **DeepSeek V4** on April 24, **MiniMax M3** on June 1, and **Kimi K2.7-Code** and **GLM-5.2** within days of each other in mid-June.",
    "Four Frontier-Class Open Models in Eight Weeks",
    "The production line \u2014 spring 2026"
   ],
   "license_evidence": [
    "**GLM-5.2 (Z.ai)**753B MoE, MIT, top open-weight on Artificial Analysis index",
    "Four Frontier-Class Open Models in Eight Weeks",
    "The production line \u2014 spring 2026"
   ],
   "significance_evidence": [
    "**GLM-5.2 (Z.ai)**753B MoE, MIT, top open-weight on Artificial Analysis index",
    "Four Frontier-Class Open Models in Eight Weeks",
    "The production line \u2014 spring 2026"
   ]
  },
  {
   "url": "https://theagiclock.com/articles/thinking-machines-inkling-decentralized-agi",
   "title": "Thinking Machines' Inkling: Why A Late-Entering Startup Just Reset The Open-Weight AGI Race",
   "published_at": "2026-07-16T00:00:00",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "A 975-billion-parameter Mixture-of-Agents transformer with 41B active parameters per token, a 1-million-token context window, and native multimodal inputs.",
   "organization_evidence": [
    "Then, on **July 15, 2026**, [Mira Murati](/experts/mira-murati)'s **[Thinking Machines Lab](/articles/thinking-machines-inkling-decentralized-agi)** released **[Inkling](/articles/thinking-machines-inkling-decentralized-agi)**, a 975-billion-parameter [open-weight](/articles/thinking-machines-inkling-decentralized-agi) [Mixture-of-Agents](/articles/polyagentmorous-agi-swarm-multi-agent-future) transformer with 41B active parameters per token, a 1-million-token context window, native multimodal inputs, and an Apache 2.0 license."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Then, on **July 15, 2026**, [Mira Murati](/experts/mira-murati)'s **[Thinking Machines Lab](/articles/thinking-machines-inkling-decentralized-agi)** released **[Inkling](/articles/thinking-machines-inkling-decentralized-agi)**, a 975-billion-parameter [open-weight](/articles/thinking-machines-inkling-decentralized-agi) [Mixture-of-Agents](/articles/polyagentmorous-agi-swarm-multi-agent-future) transformer with 41B active parameters per token, a 1-million-token context window, native multimodal inputs, and an Apache 2.0 license."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Then, on **July 15, 2026**, [Mira Murati](/experts/mira-murati)'s **[Thinking Machines Lab](/articles/thinking-machines-inkling-decentralized-agi)** released **[Inkling](/articles/thinking-machines-inkling-decentralized-agi)**, a 975-billion-parameter [open-weight](/articles/thinking-machines-inkling-decentralized-agi) [Mixture-of-Agents](/articles/polyagentmorous-agi-swarm-multi-agent-future) transformer with 41B active parameters per token, a 1-million-token context window, native multimodal inputs, and an Apache 2.0 license."
   ]
  },
  {
   "url": "https://letsdatascience.com/blog/open-source-llms-in-2026-the-definitive-comparison",
   "title": "Open Source LLMs in 2026: The Definitive Comparison",
   "published_at": "2026-03-17T20:37:50",
   "organization": "Zhipu AI",
   "model": null,
   "release_date": "2026-02",
   "access_type": null,
   "license": null,
   "significance": "Scales up to 744B parameters, pretrained on 28.5 trillion tokens, and achieves competitive scores on SWE-bench and GPQA Diamond.",
   "organization_evidence": [
    "New frontier entrants: Kimi K2 and GLM-5 (2026)",
    "**GLM-5** (Zhipu AI, February 2026) scales from 355B parameters (32B active) to 744B parameters (40B active), pretrained on **28.5 trillion tokens**."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "New frontier entrants: Kimi K2 and GLM-5 (2026)",
    "**GLM-5** (Zhipu AI, February 2026) scales from 355B parameters (32B active) to 744B parameters (40B active), pretrained on **28.5 trillion tokens**."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "New frontier entrants: Kimi K2 and GLM-5 (2026)",
    "**GLM-5** (Zhipu AI, February 2026) scales from 355B parameters (32B active) to 744B parameters (40B active), pretrained on **28.5 trillion tokens**.",
    "It achieves **77.8%** on SWE-bench, **86.0%** GPQA Diamond, and **90.0%** HumanEval \u2014 making it competitive with the best closed models on agentic tasks."
   ]
  },
  {
   "url": "https://thequantumdispatch.com/articles/kimi-k3-largest-open-weight-ai-model",
   "title": "Kimi K3 Becomes the Largest Open-Weight AI Model Yet",
   "published_at": "2026-07-18T02:30:11",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-27",
   "access_type": null,
   "license": null,
   "significance": "2.8-trillion-parameter sparse mixture-of-experts system that ranks 3rd on GDPval-AA v2",
   "organization_evidence": [
    "# Kimi K3 Becomes the Largest Open-Weight AI Model Yet",
    "Moonshot AI's Kimi K3 is a 2.8-trillion-parameter open-weight model that ranks 3rd on GDPval-AA v2, with full weights arriving July 27.",
    "Moonshot AI has announced Kimi K3, and by the numbers it is the largest open-weight AI model the community has ever been handed."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Kimi K3 Becomes the Largest Open-Weight AI Model Yet",
    "Moonshot AI's Kimi K3 is a 2.8-trillion-parameter open-weight model that ranks 3rd on GDPval-AA v2, with full weights arriving July 27.",
    "- **Availability:** full weights scheduled to drop July 27, 2026, priced near $12 per million tokens"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Kimi K3 Becomes the Largest Open-Weight AI Model Yet",
    "Moonshot AI's Kimi K3 is a 2.8-trillion-parameter open-weight model that ranks 3rd on GDPval-AA v2, with full weights arriving July 27.",
    "The Beijing-based lab describes K3 as a 2.8-trillion-parameter sparse mixture-of-experts system, and it is not merely large for the sake of a headline figure."
   ]
  },
  {
   "url": "https://thequantumdispatch.com/articles/kimi-k3-largest-open-weight-ai-model",
   "title": "Kimi K3 Becomes the Largest Open-Weight AI Model Yet",
   "published_at": "2026-07-18T02:30:11",
   "organization": "NVIDIA",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "34B open reasoning model for robotaxis with 360-degree perception",
   "organization_evidence": [
    "[AI](/ai)",
    "### NVIDIA Alpamayo 2 Super Ships an Open 34B AV Model",
    "NVIDIA released Alpamayo 2 Super for commercial use \u2014 a 34B open reasoning model for robotaxis with 360-degree perception and a permissive OpenMDW license."
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "[AI](/ai)",
    "### NVIDIA Alpamayo 2 Super Ships an Open 34B AV Model",
    "NVIDIA released Alpamayo 2 Super for commercial use \u2014 a 34B open reasoning model for robotaxis with 360-degree perception and a permissive OpenMDW license."
   ]
  },
  {
   "url": "https://bitsminds.com/news/kimi-k3-open-weights-released-largest-open-model-2026",
   "title": "The Largest Open Model Ever Just Went Free: Kimi K3's 2.8-Trillion-Parameter Weights Are Live",
   "published_at": "2026-07-27T18:48:07",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-27",
   "access_type": null,
   "license": null,
   "significance": "2.8 trillion parameters, largest open-weight model ever released",
   "organization_evidence": [
    "# The Largest Open Model Ever Just Went Free: Kimi K3's 2.8-Trillion-Parameter Weights Are Live",
    "Moonshot AI published Kimi K3's full weights on Hugging Face at 00:00 UTC on July 27, exactly as promised \u2014 2.8 trillion parameters under Apache 2.0, the biggest open-weight release in history."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# The Largest Open Model Ever Just Went Free: Kimi K3's 2.8-Trillion-Parameter Weights Are Live",
    "Moonshot AI published Kimi K3's full weights on Hugging Face at 00:00 UTC on July 27, exactly as promised \u2014 2.8 trillion parameters under Apache 2.0, the biggest open-weight release in history.",
    "| Availability | Hugging Face, from 00:00 UTC July 27, 2026 |",
    "What actually shipped"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# The Largest Open Model Ever Just Went Free: Kimi K3's 2.8-Trillion-Parameter Weights Are Live",
    "Moonshot AI published Kimi K3's full weights on Hugging Face at 00:00 UTC on July 27, exactly as promised \u2014 2.8 trillion parameters under Apache 2.0, the biggest open-weight release in history.",
    "| Total parameters | 2.8 trillion \u2014 the largest open-weight release to date |",
    "What actually shipped"
   ]
  },
  {
   "url": "https://valueaddvc.com/pulse/meituan-longcat-2-open-source-chinese-chips-2026",
   "title": "Meituan Open-Sources LongCat-2.0, a 1.6T Agentic Coding Model Trained Entirely on Chinese Chips",
   "published_at": "2026-06-30T08:00:00",
   "organization": "Meituan",
   "model": null,
   "release_date": "2026-06-30",
   "access_type": null,
   "license": null,
   "significance": "1.6-trillion-parameter mixture-of-experts coding model with a native one-million-token context window",
   "organization_evidence": [
    "Meituan, the Chinese super-app giant best known for food delivery, open-sourced LongCat-2.0 on June 30, 2026 -- a 1.6-trillion-parameter mixture-of-experts (MoE) coding model with a native one-million-token context window, released under the permissive MIT license across GitHub, Hugging Face and Meituan's own platform."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Meituan, the Chinese super-app giant best known for food delivery, open-sourced LongCat-2.0 on June 30, 2026 -- a 1.6-trillion-parameter mixture-of-experts (MoE) coding model with a native one-million-token context window, released under the permissive MIT license across GitHub, Hugging Face and Meituan's own platform."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Meituan, the Chinese super-app giant best known for food delivery, open-sourced LongCat-2.0 on June 30, 2026 -- a 1.6-trillion-parameter mixture-of-experts (MoE) coding model with a native one-million-token context window, released under the permissive MIT license across GitHub, Hugging Face and Meituan's own platform."
   ]
  },
  {
   "url": "https://buildfastwithai.com/blogs/thinking-machines-inkling-review",
   "title": "Thinking Machines Inkling Review: Tested (2026)",
   "published_at": "2026-07-16T09:14:28",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "975B total parameters with only 41B active per token, a 1 million token context window, native reasoning over text, images, and audio, and a thinking-effort dial",
   "organization_evidence": [
    "Thinking Machines Inkling Review: Tested (2026)",
    "Inkling, released by Mira Murati's Thinking Machines Lab on July 15, 2026, is the most ambitious open-weights drop of the year, and after three days of testing it, I think it changes what the open tier is for."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Thinking Machines Inkling Review: Tested (2026)",
    "Inkling, released by Mira Murati's Thinking Machines Lab on July 15, 2026, is the most ambitious open-weights drop of the year, and after three days of testing it, I think it changes what the open tier is for."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Thinking Machines Inkling Review: Tested (2026)",
    "The headline numbers: 975B total parameters with only 41B active per token, a 1 million token context window, native reasoning over text, images, and audio, and a thinking-effort dial you can turn from 0.2 to 0.99 to trade quality against token cost."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/gemma-4-26b-a4b-vs-qwen3-5-122b-a10b",
   "title": "Gemma 4 26B A4B (Reasoning) vs Qwen3.5 122B A10B (Reasoning): Model Comparison",
   "published_at": null,
   "organization": "Google",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "25.2B parameters with 3.8B active at inference time, 256k context window, and image input support",
   "organization_evidence": [
    "# Gemma 4 26B A4B (Reasoning) vs. Qwen3.5 122B A10B (Reasoning)",
    "| Metric | ![Google logo](/img/logos/google_small.svg)Gemma 4 26B A4B (Reasoning) | ![Alibaba logo](/img/logos/alibaba_small.svg)Qwen3.5 122B A10B (Reasoning) | Analysis |",
    "| Creator | ![Google](/img/logos/google_small.svg)[](https://deepmind.google/technologies/gemini/) | ![Alibaba](/img/logos/alibaba_small.svg)[](https://www.alibabacloud.com/) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Gemma 4 26B A4B (Reasoning) vs. Qwen3.5 122B A10B (Reasoning)",
    "| Metric | ![Google logo](/img/logos/google_small.svg)Gemma 4 26B A4B (Reasoning) | ![Alibaba logo](/img/logos/alibaba_small.svg)Qwen3.5 122B A10B (Reasoning) | Analysis |",
    "| Release Date | April, 2026 | February, 2026 | Gemma 4 26B A4B (Reasoning) has a more recent release date than Qwen3.5 122B A10B (Reasoning) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Gemma 4 26B A4B (Reasoning) vs. Qwen3.5 122B A10B (Reasoning)",
    "| Metric | ![Google logo](/img/logos/google_small.svg)Gemma 4 26B A4B (Reasoning) | ![Alibaba logo](/img/logos/alibaba_small.svg)Qwen3.5 122B A10B (Reasoning) | Analysis |",
    "| Context Window | 256k tokens (~384 A4 pages of size 12 Arial font) | 262k tokens (~393 A4 pages of size 12 Arial font) | Gemma 4 26B A4B (Reasoning) is smaller than Qwen3.5 122B A10B (Reasoning) |",
    "| Parameters | 25.2B, 3.8B active at inference time | 125B, 10B active at inference time | Gemma 4 26B A4B (Reasoning) is smaller than Qwen3.5 122B A10B (Reasoning) |",
    "| Image Input Support | Yes | Yes | Both Gemma 4 26B A4B (Reasoning) and Qwen3.5 122B A10B (Reasoning) have image input support |"
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/gemma-4-26b-a4b-vs-qwen3-5-122b-a10b",
   "title": "Gemma 4 26B A4B (Reasoning) vs Qwen3.5 122B A10B (Reasoning): Model Comparison",
   "published_at": null,
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-02",
   "access_type": null,
   "license": null,
   "significance": "125B parameters with 10B active at inference time, 262k context window, and image input support",
   "organization_evidence": [
    "# Gemma 4 26B A4B (Reasoning) vs. Qwen3.5 122B A10B (Reasoning)",
    "| Metric | ![Google logo](/img/logos/google_small.svg)Gemma 4 26B A4B (Reasoning) | ![Alibaba logo](/img/logos/alibaba_small.svg)Qwen3.5 122B A10B (Reasoning) | Analysis |",
    "| Creator | ![Google](/img/logos/google_small.svg)[](https://deepmind.google/technologies/gemini/) | ![Alibaba](/img/logos/alibaba_small.svg)[](https://www.alibabacloud.com/) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Gemma 4 26B A4B (Reasoning) vs. Qwen3.5 122B A10B (Reasoning)",
    "| Metric | ![Google logo](/img/logos/google_small.svg)Gemma 4 26B A4B (Reasoning) | ![Alibaba logo](/img/logos/alibaba_small.svg)Qwen3.5 122B A10B (Reasoning) | Analysis |",
    "| Release Date | April, 2026 | February, 2026 | Gemma 4 26B A4B (Reasoning) has a more recent release date than Qwen3.5 122B A10B (Reasoning) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Gemma 4 26B A4B (Reasoning) vs. Qwen3.5 122B A10B (Reasoning)",
    "| Metric | ![Google logo](/img/logos/google_small.svg)Gemma 4 26B A4B (Reasoning) | ![Alibaba logo](/img/logos/alibaba_small.svg)Qwen3.5 122B A10B (Reasoning) | Analysis |",
    "| Context Window | 256k tokens (~384 A4 pages of size 12 Arial font) | 262k tokens (~393 A4 pages of size 12 Arial font) | Gemma 4 26B A4B (Reasoning) is smaller than Qwen3.5 122B A10B (Reasoning) |",
    "| Parameters | 25.2B, 3.8B active at inference time | 125B, 10B active at inference time | Gemma 4 26B A4B (Reasoning) is smaller than Qwen3.5 122B A10B (Reasoning) |",
    "| Image Input Support | Yes | Yes | Both Gemma 4 26B A4B (Reasoning) and Qwen3.5 122B A10B (Reasoning) have image input support |"
   ]
  },
  {
   "url": "https://synthszr.vercel.app/en/posts/chinese-open-weight-models-conquer-the-download-charts",
   "title": "Chinese Open-Weight Models Conquer the Download Charts",
   "published_at": "2026-08-16T03:30:25",
   "organization": "Meta",
   "model": null,
   "release_date": "2026-08-10",
   "access_type": null,
   "license": null,
   "significance": "30 billion parameters, designed for local agents, coding, and function calling, running on a Mac or PC with a single consumer GPU",
   "organization_evidence": [
    "Chinese Open-Weight Models Conquer the Download Charts",
    "On August 10, Meta released Muse Glimmer with 30 billion parameters under the [Apache-2.0](/en/glossary/apache-2-0-lizenz) license, designed for local agents, coding, and [function calling](/en/glossary/funktionsaufruf)."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Chinese Open-Weight Models Conquer the Download Charts",
    "On August 10, Meta released Muse Glimmer with 30 billion parameters under the [Apache-2.0](/en/glossary/apache-2-0-lizenz) license, designed for local agents, coding, and [function calling](/en/glossary/funktionsaufruf)."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Chinese Open-Weight Models Conquer the Download Charts",
    "On August 10, Meta released Muse Glimmer with 30 billion parameters under the [Apache-2.0](/en/glossary/apache-2-0-lizenz) license, designed for local agents, coding, and [function calling](/en/glossary/funktionsaufruf)."
   ]
  },
  {
   "url": "https://synthszr.vercel.app/en/posts/chinese-open-weight-models-conquer-the-download-charts",
   "title": "Chinese Open-Weight Models Conquer the Download Charts",
   "published_at": "2026-08-16T03:30:25",
   "organization": "Nvidia",
   "model": null,
   "release_date": "2026-08-11",
   "access_type": null,
   "license": null,
   "significance": "30 billion parameters, aimed at high-volume agentic workloads with up to four times faster output",
   "organization_evidence": [
    "Chinese Open-Weight Models Conquer the Download Charts",
    "A day later, Nvidia introduced Nemotron 3.5 Lightning, also with 30 billion parameters and aimed at high-volume [agentic](/en/glossary/agentische-ki) workloads; the company claims up to four times faster output in its own benchmarks."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Chinese Open-Weight Models Conquer the Download Charts",
    "On August 10, Meta released Muse Glimmer with 30 billion parameters under the [Apache-2.0](/en/glossary/apache-2-0-lizenz) license, designed for local agents, coding, and [function calling](/en/glossary/funktionsaufruf).",
    "A day later, Nvidia introduced Nemotron 3.5 Lightning, also with 30 billion parameters and aimed at high-volume [agentic](/en/glossary/agentische-ki) workloads; the company claims up to four times faster output in its own benchmarks."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Chinese Open-Weight Models Conquer the Download Charts",
    "A day later, Nvidia introduced Nemotron 3.5 Lightning, also with 30 billion parameters and aimed at high-volume [agentic](/en/glossary/agentische-ki) workloads; the company claims up to four times faster output in its own benchmarks."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/qwen3-8-max-vs-qwen3-6-27b",
   "title": "Qwen3.8 Max vs Qwen3.6 27B (Reasoning): Model Comparison | Artificial Analysis",
   "published_at": null,
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "Reasoning capability and open source weights with 27B parameters",
   "organization_evidence": [
    "# Qwen3.8 Max vs. Qwen3.6 27B (Reasoning)",
    "| | ![Alibaba logo](/img/logos/alibaba_small.svg)Alibaba Qwen3.8 Max | ![Alibaba logo](/img/logos/alibaba_small.svg)Alibaba Qwen3.6 27B (Reasoning) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Qwen3.8 Max vs. Qwen3.6 27B (Reasoning)",
    "| Release Date | August 2026 | April 2026 | Qwen3.8 Max has a more recent release date than Qwen3.6 27B (Reasoning) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Qwen3.8 Max vs. Qwen3.6 27B (Reasoning)",
    "| Reasoning | Yes | Yes | Both Qwen3.8 Max and Qwen3.6 27B (Reasoning) have reasoning |",
    "| Open Source (Weights) | No | Yes | Qwen3.6 27B (Reasoning) is open source while Qwen3.8 Max is proprietary |"
   ]
  },
  {
   "url": "https://1023jack.com/market/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is-2",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - 1023 Jack",
   "published_at": "2026-07-15T10:38:52",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor",
   "organization_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "Chinese laboratories have released four frontier-class open-weight AI models within a span of just eight weeks, starting with DeepSeek V4 on April 24 and concluding with Kimi K2.7-Code and GLM-5.2 in mid-June.",
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "Chinese laboratories have released four frontier-class open-weight AI models within a span of just eight weeks, starting with DeepSeek V4 on April 24 and concluding with Kimi K2.7-Code and GLM-5.2 in mid-June.",
    "APR 24",
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor"
   ]
  },
  {
   "url": "https://1023jack.com/market/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is-2",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - 1023 Jack",
   "published_at": "2026-07-15T10:38:52",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "cheap 1M-token context, native multimodal, modified-MIT",
   "organization_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "Between late April and mid-June 2026, Chinese labs introduced four major open models: DeepSeek V4, MiniMax M3, Kimi K2.7-Code, and GLM-5.2.",
    "**MiniMax M3**cheap 1M-token context, native multimodal, modified-MIT"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "Chinese laboratories have released four frontier-class open-weight AI models within a span of just eight weeks, starting with DeepSeek V4 on April 24 and concluding with Kimi K2.7-Code and GLM-5.2 in mid-June.",
    "JUN 01",
    "**MiniMax M3**cheap 1M-token context, native multimodal, modified-MIT"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "**MiniMax M3**cheap 1M-token context, native multimodal, modified-MIT"
   ]
  },
  {
   "url": "https://1023jack.com/market/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is-2",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - 1023 Jack",
   "published_at": "2026-07-15T10:38:52",
   "organization": "Moonshot",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "agent-run specialist, ~30% fewer thinking tokens than K2.6",
   "organization_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "Chinese labs such as DeepSeek, Z.ai, Moonshot, and Alibaba have each developed distinct models targeting different niches\u2014cost-efficiency, long-horizon stability, and broad self-hosting capabilities.",
    "**Kimi K2.7-Code (Moonshot)**agent-run specialist, ~30% fewer thinking tokens than K2.6"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "Chinese laboratories have released four frontier-class open-weight AI models within a span of just eight weeks, starting with DeepSeek V4 on April 24 and concluding with Kimi K2.7-Code and GLM-5.2 in mid-June.",
    "JUN 13",
    "**Kimi K2.7-Code (Moonshot)**agent-run specialist, ~30% fewer thinking tokens than K2.6"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "**Kimi K2.7-Code (Moonshot)**agent-run specialist, ~30% fewer thinking tokens than K2.6"
   ]
  },
  {
   "url": "https://1023jack.com/market/signal-four-frontier-class-open-models-in-eight-weeks-china-s-release-cadence-is-2",
   "title": "Signal: Four Frontier-Class Open Models in Eight Weeks \u2014 China\u2019s Release Cadence Is the Story - 1023 Jack",
   "published_at": "2026-07-15T10:38:52",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "753B MoE, MIT, top open-weight on Artificial Analysis index",
   "organization_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "Chinese labs such as DeepSeek, Z.ai, Moonshot, and Alibaba have each developed distinct models targeting different niches\u2014cost-efficiency, long-horizon stability, and broad self-hosting capabilities.",
    "**GLM-5.2 (Z.ai)**753B MoE, MIT, top open-weight on Artificial Analysis index"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "Chinese laboratories have released four frontier-class open-weight AI models within a span of just eight weeks, starting with DeepSeek V4 on April 24 and concluding with Kimi K2.7-Code and GLM-5.2 in mid-June.",
    "JUN 13\u201316",
    "**GLM-5.2 (Z.ai)**753B MoE, MIT, top open-weight on Artificial Analysis index"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Four Frontier-Class Open Models in Eight Weeks",
    "**GLM-5.2 (Z.ai)**753B MoE, MIT, top open-weight on Artificial Analysis index"
   ]
  },
  {
   "url": "https://thequantumdispatch.com/articles/nvidia-nemotron-3-ultra-550b-open-reasoning-model-long-running-agents-june-2026",
   "title": "NVIDIA Nemotron 3 Ultra: A 550B Open Model Built for Long-Running AI Agents",
   "published_at": "2026-06-09T22:20:32",
   "organization": "NVIDIA",
   "model": null,
   "release_date": "2026-06-04",
   "access_type": null,
   "license": null,
   "significance": "a fully open 550B-parameter reasoning model topping US open-model benchmarks and tuned for long-running AI agents",
   "organization_evidence": [
    "# NVIDIA Nemotron 3 Ultra: A 550B Open Model Built for Long-Running AI Agents",
    "NVIDIA released Nemotron 3 Ultra on June 4, 2026 \u2014 a fully open 550B-parameter reasoning model topping US open-model benchmarks and tuned for long-running AI agents."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# NVIDIA Nemotron 3 Ultra: A 550B Open Model Built for Long-Running AI Agents",
    "NVIDIA released Nemotron 3 Ultra on June 4, 2026 \u2014 a fully open 550B-parameter reasoning model topping US open-model benchmarks and tuned for long-running AI agents."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# NVIDIA Nemotron 3 Ultra: A 550B Open Model Built for Long-Running AI Agents",
    "NVIDIA released Nemotron 3 Ultra on June 4, 2026 \u2014 a fully open 550B-parameter reasoning model topping US open-model benchmarks and tuned for long-running AI agents."
   ]
  },
  {
   "url": "https://aidb.digital/blog/2026-07-27-moonshot-ai-releases-kimi-k3-open-weights-a-2-8-trillion-parameter-mo",
   "title": "Moonshot AI Releases Kimi K3 Open Weights: A 2.8-Trillion-Parameter Model That Challenges US Frontier Systems \u00b7 AIDB",
   "published_at": "2026-07-27T00:00:00",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-26",
   "access_type": null,
   "license": null,
   "significance": "world's first open-weight 2.8-trillion-parameter artificial intelligence model with 2.5x the overall scaling efficiency of K2 and a 1M-token context window",
   "organization_evidence": [
    "## What Was Released",
    "Moonshot AI released free, public download weights for Kimi K3 on July 26, 2026, at roughly 7:30 PM EDT \u2014 a day ahead of the July 27 target the company had previously communicated \u2014 with confirmed specifications of 2.8 trillion parameters and a 1,048,576-token (1M) context window."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## What Was Released",
    "Moonshot AI released free, public download weights for Kimi K3 on July 26, 2026, at roughly 7:30 PM EDT \u2014 a day ahead of the July 27 target the company had previously communicated \u2014 with confirmed specifications of 2.8 trillion parameters and a 1,048,576-token (1M) context window."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## Architecture",
    "Kimi K3 is the world's first open-weight 2.8-trillion-parameter artificial intelligence (AI) model, introducing Kimi Delta Attention (KDA), Stable LatentMoE, and a 1M-token context window.",
    "With the Stable LatentMoE framework, the model efficiently activates 16 out of 896 experts."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/qwen3-6-max-vs-nvidia-nemotron-3-super-120b-a12b",
   "title": "Qwen3.6 Max Preview vs NVIDIA Nemotron 3 Super 120B A12B (Reasoning): Model Comparison",
   "published_at": null,
   "organization": "NVIDIA",
   "model": null,
   "release_date": "March, 2026",
   "access_type": null,
   "license": null,
   "significance": null,
   "organization_evidence": [
    "# Qwen3.6 Max Preview vs. NVIDIA Nemotron 3 Super 120B A12B (Reasoning)",
    "| Metric | ![Alibaba logo](/img/logos/alibaba_small.svg)Qwen3.6 Max Preview | ![NVIDIA logo](/img/logos/nvidia_small.svg)NVIDIA Nemotron 3 Super 120B A12B (Reasoning) | Analysis |",
    "| Creator | ![Alibaba](/img/logos/alibaba_small.svg)[](https://www.alibabacloud.com/) | ![NVIDIA](/img/logos/nvidia_small.svg)[](https://www.nvidia.com/) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Qwen3.6 Max Preview vs. NVIDIA Nemotron 3 Super 120B A12B (Reasoning)",
    "| Metric | ![Alibaba logo](/img/logos/alibaba_small.svg)Qwen3.6 Max Preview | ![NVIDIA logo](/img/logos/nvidia_small.svg)NVIDIA Nemotron 3 Super 120B A12B (Reasoning) | Analysis |",
    "| Release Date | April, 2026 | March, 2026 | Qwen3.6 Max Preview has a more recent release date than NVIDIA Nemotron 3 Super 120B A12B (Reasoning) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": null
  },
  {
   "url": "https://en.people.cn/n3/2026/0718/c90000-20479238.html",
   "title": "China launches global open-source initiative for AI weather model",
   "published_at": "2026-07-18T00:00:00",
   "organization": "China Meteorological Administration",
   "model": null,
   "release_date": "2026-07-17",
   "access_type": null,
   "license": null,
   "significance": "trained on 50 million tokens of meteorological service data and integrated with authoritative weather datasets to support weather forecasting, risk assessment and meteorological services",
   "organization_evidence": [
    "China launches global open-source initiative for AI weather model",
    "Developed by the CMA, Fenghe is trained on 50 million tokens of meteorological service data and integrated with authoritative weather datasets to support weather forecasting, risk assessment and meteorological services."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "China launches global open-source initiative for AI weather model",
    "SHANGHAI, July 17 (Xinhua) -- China on Friday unveiled Fenghe, an open-source large language model dedicated to meteorological services, and launched a global open-source initiative for this model."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "China launches global open-source initiative for AI weather model",
    "Developed by the CMA, Fenghe is trained on 50 million tokens of meteorological service data and integrated with authoritative weather datasets to support weather forecasting, risk assessment and meteorological services."
   ]
  },
  {
   "url": "https://benchr.org/recent-releases",
   "title": "Recent AI model releases \u2014 benchr",
   "published_at": null,
   "organization": "Qwen",
   "model": null,
   "release_date": "2026-06-24",
   "access_type": null,
   "license": null,
   "significance": "Open-weight language world models for agent-environment simulation across seven domains with 256K context",
   "organization_evidence": [
    "Reference\u00b7Updated July 22, 2026",
    "# Recent AI model releases",
    "The major labs ship something significant every few weeks. This page lists significant releases since early 2026, in reverse chronological order. For deeper coverage of each model, see the linked article or the comparison tool.",
    "## June 2026",
    "- **Qwen-AgentWorld (35B-A3B and 397B-A17B)** \u2014 Released June 24, 2026. Open-weight (Apache-2.0) \u201clanguage world models\u201d for agent-environment simulation across seven domains; 256K context, download-only. [Source: Qwen](https://github.com/QwenLM/Qwen-AgentWorld)."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Reference\u00b7Updated July 22, 2026",
    "# Recent AI model releases",
    "The major labs ship something significant every few weeks. This page lists significant releases since early 2026, in reverse chronological order. For deeper coverage of each model, see the linked article or the comparison tool.",
    "## June 2026",
    "- **Qwen-AgentWorld (35B-A3B and 397B-A17B)** \u2014 Released June 24, 2026. Open-weight (Apache-2.0) \u201clanguage world models\u201d for agent-environment simulation across seven domains; 256K context, download-only. [Source: Qwen](https://github.com/QwenLM/Qwen-AgentWorld)."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Reference\u00b7Updated July 22, 2026",
    "# Recent AI model releases",
    "The major labs ship something significant every few weeks. This page lists significant releases since early 2026, in reverse chronological order. For deeper coverage of each model, see the linked article or the comparison tool.",
    "## June 2026",
    "- **Qwen-AgentWorld (35B-A3B and 397B-A17B)** \u2014 Released June 24, 2026. Open-weight (Apache-2.0) \u201clanguage world models\u201d for agent-environment simulation across seven domains; 256K context, download-only. [Source: Qwen](https://github.com/QwenLM/Qwen-AgentWorld)."
   ]
  },
  {
   "url": "https://benchr.org/recent-releases",
   "title": "Recent AI model releases \u2014 benchr",
   "published_at": null,
   "organization": "Mistral",
   "model": null,
   "release_date": "2026-03-16",
   "access_type": null,
   "license": null,
   "significance": "Open-weight hybrid MoE with 119B total / 6B active parameters, 256K context, and text + image input",
   "organization_evidence": [
    "Reference\u00b7Updated July 22, 2026",
    "# Recent AI model releases",
    "The major labs ship something significant every few weeks. This page lists significant releases since early 2026, in reverse chronological order. For deeper coverage of each model, see the linked article or the comparison tool.",
    "## March 2026",
    "- **Mistral Small 4** \u2014 Released March 16, 2026. Open-weight (Apache-2.0) hybrid MoE, 119B total / 6B active, 256K context, text + image input. $0.15/$0.60 per 1M. [Source: Mistral](https://mistral.ai/news/mistral-small-4/)."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Reference\u00b7Updated July 22, 2026",
    "# Recent AI model releases",
    "The major labs ship something significant every few weeks. This page lists significant releases since early 2026, in reverse chronological order. For deeper coverage of each model, see the linked article or the comparison tool.",
    "## March 2026",
    "- **Mistral Small 4** \u2014 Released March 16, 2026. Open-weight (Apache-2.0) hybrid MoE, 119B total / 6B active, 256K context, text + image input. $0.15/$0.60 per 1M. [Source: Mistral](https://mistral.ai/news/mistral-small-4/)."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Reference\u00b7Updated July 22, 2026",
    "# Recent AI model releases",
    "The major labs ship something significant every few weeks. This page lists significant releases since early 2026, in reverse chronological order. For deeper coverage of each model, see the linked article or the comparison tool.",
    "## March 2026",
    "- **Mistral Small 4** \u2014 Released March 16, 2026. Open-weight (Apache-2.0) hybrid MoE, 119B total / 6B active, 256K context, text + image input. $0.15/$0.60 per 1M. [Source: Mistral](https://mistral.ai/news/mistral-small-4/)."
   ]
  },
  {
   "url": "https://benchr.org/recent-releases",
   "title": "Recent AI model releases \u2014 benchr",
   "published_at": null,
   "organization": "Qwen",
   "model": null,
   "release_date": "2026-02-16",
   "access_type": null,
   "license": null,
   "significance": "Flagship 397B MoE with 17B active parameters, spanning nine sizes",
   "organization_evidence": [
    "Reference\u00b7Updated July 22, 2026",
    "# Recent AI model releases",
    "The major labs ship something significant every few weeks. This page lists significant releases since early 2026, in reverse chronological order. For deeper coverage of each model, see the linked article or the comparison tool.",
    "## February 2026",
    "- **Qwen 3.5** \u2014 Released February 16, 2026. Flagship 397B MoE, 17B active; the family spans nine sizes, most under Apache 2.0. Superseded by Qwen3.6 in April. [Source: Qwen GitHub](https://github.com/QwenLM/Qwen3.6). [benchr review](articles/qwen-review)."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Reference\u00b7Updated July 22, 2026",
    "# Recent AI model releases",
    "The major labs ship something significant every few weeks. This page lists significant releases since early 2026, in reverse chronological order. For deeper coverage of each model, see the linked article or the comparison tool.",
    "## February 2026",
    "- **Qwen 3.5** \u2014 Released February 16, 2026. Flagship 397B MoE, 17B active; the family spans nine sizes, most under Apache 2.0. Superseded by Qwen3.6 in April. [Source: Qwen GitHub](https://github.com/QwenLM/Qwen3.6). [benchr review](articles/qwen-review)."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Reference\u00b7Updated July 22, 2026",
    "# Recent AI model releases",
    "The major labs ship something significant every few weeks. This page lists significant releases since early 2026, in reverse chronological order. For deeper coverage of each model, see the linked article or the comparison tool.",
    "## February 2026",
    "- **Qwen 3.5** \u2014 Released February 16, 2026. Flagship 397B MoE, 17B active; the family spans nine sizes, most under Apache 2.0. Superseded by Qwen3.6 in April. [Source: Qwen GitHub](https://github.com/QwenLM/Qwen3.6). [benchr review](articles/qwen-review)."
   ]
  },
  {
   "url": "https://robotsatlas.com/posts/thinking-machines-inkling-otwarty-moe-975b-mira-murati",
   "title": "Thinking Machines reveals Inkling \u2014 Mira Murati open-weight MoE 975B",
   "published_at": "2026-07-17T16:13:00",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "Mixture-of-Experts system with 975 billion total parameters and native support for text, image, audio, and video",
   "organization_evidence": [
    "Artificial Intelligence",
    "Thinking Machines Lab, the startup founded by former OpenAI CTO Mira Murati, released its first AI model on July 15, 2026 \u2014 an open-weight Mixture-of-Experts system called Inkling, with 975 billion total parameters and native support for text, image, audio, and video."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Artificial Intelligence",
    "Thinking Machines Lab, the startup founded by former OpenAI CTO Mira Murati, released its first AI model on July 15, 2026 \u2014 an open-weight Mixture-of-Experts system called Inkling, with 975 billion total parameters and native support for text, image, audio, and video."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Artificial Intelligence",
    "Thinking Machines Lab, the startup founded by former OpenAI CTO Mira Murati, released its first AI model on July 15, 2026 \u2014 an open-weight Mixture-of-Experts system called Inkling, with 975 billion total parameters and native support for text, image, audio, and video."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/gemma-4-31b-vs-claude-opus-4-6-adaptive",
   "title": "Gemma 4 31B (Reasoning) vs Claude Opus 4.6 (Adaptive Reasoning, Max Effort): Model Comparison",
   "published_at": null,
   "organization": "Google",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "Reasoning model with 256k context window and image input support",
   "organization_evidence": [
    "# Gemma 4 31B (Reasoning) vs. Claude Opus 4.6 (Adaptive Reasoning, Max Effort)",
    "| Creator | ![Google](/img/logos/google_small.svg)[](https://deepmind.google/technologies/gemini/) | ![Anthropic](/img/logos/anthropic_small.svg)[](https://www.anthropic.com/) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Gemma 4 31B (Reasoning) vs. Claude Opus 4.6 (Adaptive Reasoning, Max Effort)",
    "| Release Date | April, 2026 | February, 2026 | Gemma 4 31B (Reasoning) has a more recent release date than Claude Opus 4.6 (Adaptive Reasoning, Max Effort) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Gemma 4 31B (Reasoning) vs. Claude Opus 4.6 (Adaptive Reasoning, Max Effort)",
    "| Context Window | 256k tokens (~384 A4 pages of size 12 Arial font) | 1000k tokens (~1500 A4 pages of size 12 Arial font) | Gemma 4 31B (Reasoning) is smaller than Claude Opus 4.6 (Adaptive Reasoning, Max Effort) |",
    "| Image Input Support | Yes | Yes | Both Gemma 4 31B (Reasoning) and Claude Opus 4.6 (Adaptive Reasoning, Max Effort) have image input support |"
   ]
  },
  {
   "url": "http://ai-tldr.dev/releases/thinking-machines-inkling",
   "title": "Inkling \u2014 Thinking Machines' first open-weights 975B/41B multimodal MoE",
   "published_at": "2026-07-15T12:00:00",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "A 975B / 41B-active multimodal MoE foundation model processing text, images, and audio over a 1M-token context.",
   "organization_evidence": [
    "Thinking Machines Lab \u00b7 2026-07-15 \u00b7 major",
    "Mira Murati's Thinking Machines Lab ships its first foundation model, and puts the full weights on HuggingFace under Apache 2.0."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Thinking Machines Lab \u00b7 2026-07-15 \u00b7 major",
    "# Inkling \u2014 Thinking Machines' first open-weights 975B/41B multimodal MoE"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Thinking Machines Lab \u00b7 2026-07-15 \u00b7 major",
    "# Inkling \u2014 Thinking Machines' first open-weights 975B/41B multimodal MoE",
    "Thinking Machines' first public foundation model: a 975B / 41B-active MoE released under Apache 2.0."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/glm-5-1-vs-gpt-5-4-pro",
   "title": "GLM-5.1 (Reasoning) vs GPT-5.4 Pro (xhigh): Model Comparison",
   "published_at": null,
   "organization": "Z AI",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "A reasoning model with 200k context window and open source weights",
   "organization_evidence": [
    "# GLM-5.1 (Reasoning) vs. GPT-5.4 Pro (xhigh)",
    "## Model Comparison",
    "| Creator | ![Z AI](/img/logos/zai_small.svg)[](https://z.ai/) | ![OpenAI](/img/logos/openai_small.svg)[](https://openai.com/) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# GLM-5.1 (Reasoning) vs. GPT-5.4 Pro (xhigh)",
    "## Model Comparison",
    "| Release Date | April, 2026 | March, 2026 | GLM-5.1 (Reasoning) has a more recent release date than GPT-5.4 Pro (xhigh) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# GLM-5.1 (Reasoning) vs. GPT-5.4 Pro (xhigh)",
    "## Model Comparison",
    "| Context Window | 200k tokens (~300 A4 pages of size 12 Arial font) | 1050k tokens (~1575 A4 pages of size 12 Arial font) | GLM-5.1 (Reasoning) is smaller than GPT-5.4 Pro (xhigh) |",
    "| Open Source (Weights) | Yes[](https://huggingface.co/zai-org/GLM-5.1) | No | GLM-5.1 (Reasoning) is open source while GPT-5.4 Pro (xhigh) is proprietary |"
   ]
  },
  {
   "url": "https://crowdfundinsider.com/2026/07/294116-moonshot-ai-wraps-up-kimi-k3-rollout-with-full-model-weight-release",
   "title": "Moonshot AI Wraps Up Kimi K3 Rollout With Full Model Weight Release | Crowdfund Insider",
   "published_at": "2026-07-28T22:32:21",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-27",
   "access_type": null,
   "license": null,
   "significance": "sparse mixture-of-experts architecture containing 2.8 trillion total parameters, supporting a one-million-token context window and processing text, images, and video natively",
   "organization_evidence": [
    "**Moonshot AI** has finalized the rollout of its flagship ***Kimi K3*** model by publicly releasing the complete set of open weights, marking the full availability of what the company positions as the first open-weight system in the three-trillion-parameter class."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The weights became downloadable on July 27, [2026](https://www.crowdfundinsider.com/tag/outlook-2026), eleven days after the initial hosted service debut on July 16, fulfilling the timeline the Beijing-based firm announced at launch."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Kimi K3 is a sparse mixture-of-experts architecture containing 2.8 trillion total parameters, of which roughly 104 billion activate for any given token.",
    "The system supports a one-million-token context window and processes text, images, and video natively through a vision encoder trained from scratch."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/gpt-5-6-luna-medium-vs-deepseek-v4-pro",
   "title": "DeepSeek V4 Pro 0813 (Reasoning, Max Effort) vs GPT-5.6 Luna (medium): Model Comparison | Artificial Analysis",
   "published_at": null,
   "organization": "DeepSeek",
   "model": "DeepSeek V4 Pro 0813 (Reasoning, Max Effort)",
   "release_date": "2026-08",
   "access_type": "open source",
   "license": null,
   "significance": "more intelligent reasoning model with a 1000k token context window",
   "organization_evidence": [
    "# DeepSeek V4 Pro 0813 (Reasoning, Max Effort) vs. GPT-5.6 Luna (medium)",
    "## Model Comparison",
    "| | ![DeepSeek logo](/img/logos/deepseek_small.svg)DeepSeek DeepSeek V4 Pro 0813 (Reasoning, Max Effort) | ![OpenAI logo](/img/logos/openai_small.svg)OpenAI GPT-5.6 Luna (medium) | |"
   ],
   "model_evidence": [
    "# DeepSeek V4 Pro 0813 (Reasoning, Max Effort) vs. GPT-5.6 Luna (medium)",
    "## Model Comparison",
    "| | ![DeepSeek logo](/img/logos/deepseek_small.svg)DeepSeek DeepSeek V4 Pro 0813 (Reasoning, Max Effort) | ![OpenAI logo](/img/logos/openai_small.svg)OpenAI GPT-5.6 Luna (medium) | |"
   ],
   "release_date_evidence": [
    "# DeepSeek V4 Pro 0813 (Reasoning, Max Effort) vs. GPT-5.6 Luna (medium)",
    "## Model Comparison",
    "| Release Date | August 2026 | July 2026 | DeepSeek V4 Pro 0813 (Reasoning, Max Effort) has a more recent release date than GPT-5.6 Luna (medium) |"
   ],
   "access_type_evidence": [
    "# DeepSeek V4 Pro 0813 (Reasoning, Max Effort) vs. GPT-5.6 Luna (medium)",
    "## Model Comparison",
    "| Open Source (Weights) | Yes | No | DeepSeek V4 Pro 0813 (Reasoning, Max Effort) is open source while GPT-5.6 Luna (medium) is proprietary |"
   ],
   "license_evidence": null,
   "significance_evidence": [
    "# DeepSeek V4 Pro 0813 (Reasoning, Max Effort) vs. GPT-5.6 Luna (medium)",
    "## Model Comparison",
    "| Intelligence Index | 53 | 39 | DeepSeek V4 Pro 0813 (Reasoning, Max Effort) is more intelligent than GPT-5.6 Luna (medium) |",
    "| Context Window | 1000k tokens~1,500 A4 pages of size 12 Arial font | 1000k tokens~1,500 A4 pages of size 12 Arial font | Both DeepSeek V4 Pro 0813 (Reasoning, Max Effort) and GPT-5.6 Luna (medium) have the same sized context window |",
    "| Reasoning | Yes | Yes | Both DeepSeek V4 Pro 0813 (Reasoning, Max Effort) and GPT-5.6 Luna (medium) have reasoning |"
   ]
  },
  {
   "url": "https://orcarouter.ai/blog/inkling-ai-model-review",
   "title": "Benchmarks, Pricing & Full Verdict",
   "published_at": "2026-07-16T09:40:12",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "An open-weights, multimodal Mixture-of-Experts (MoE) model with a 1-million-token context window, 975B total / 41B active parameters, and a controllable reasoning effort dial.",
   "organization_evidence": [
    "Inkling AI Model Review: Thinking Machines\u2019 First Open-Weights Model, Tested and Explained",
    "This **Inkling AI model review** takes a close look at the debut release from Thinking Machines Lab, the startup led by former [OpenAI](/models/openai/gpt-5.5-pro) CTO Mira Murati."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "**Maker & date:** Thinking Machines Lab; released July 15, 2026 as the lab\u2019s first model.",
    "Inkling is an open-weights, multimodal, Mixture-of-Experts large language and reasoning model, announced by Thinking Machines Lab on July 15, 2026 and released under the permissive **Apache 2.0** license with full weights on Hugging Face."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Inkling is an open-weights, multimodal Mixture-of-Experts (MoE) model with a 1-million-token context window and a controllable \u201cthinking effort\u201d dial.",
    "**Architecture:** Sparse MoE, 975B total / 41B active parameters, 66 layers, up to 1M-token context in the weights (256K via hosted APIs)."
   ]
  },
  {
   "url": "https://dev.to/ai_openfree_b23025ef075cf/vidraft-releases-aether-7b-5attn-a-fully-reproducible-open-source-llm-with-five-heterogeneous-4j82",
   "title": "VIDRAFT Releases Aether-7B-5Attn: A Fully Reproducible Open-Source LLM with Five Heterogeneous Attention Mechanisms",
   "published_at": "2026-07-21T02:33:39",
   "organization": "VIDRAFT",
   "model": null,
   "release_date": "2026-07-20",
   "access_type": null,
   "license": null,
   "significance": "A 6.59B-parameter Mixture-of-Experts foundation model combining five heterogeneous attention mechanisms designed for full reproducibility.",
   "organization_evidence": [
    "# VIDRAFT Releases Aether-7B-5Attn: A Fully Reproducible Open-Source LLM with Five Heterogeneous Attention Mechanisms",
    "> **TL;DR:** Korean Pre-AGI startup VIDRAFT has published `Aether-7B-5Attn`, a 6.59B-parameter Mixture-of-Experts foundation model on Hugging Face under Apache 2.0 \u2014 releasing not just weights but also training data recipes, full training code, hyperparameters, training logs, intermediate checkpoints, and evaluation code."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# VIDRAFT Releases Aether-7B-5Attn: A Fully Reproducible Open-Source LLM with Five Heterogeneous Attention Mechanisms",
    "`Aether-7B-5Attn` is VIDRAFT's first fully open-source foundation LLM, published on Hugging Face on July 20, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# VIDRAFT Releases Aether-7B-5Attn: A Fully Reproducible Open-Source LLM with Five Heterogeneous Attention Mechanisms",
    "> **TL;DR:** Korean Pre-AGI startup VIDRAFT has published `Aether-7B-5Attn`, a 6.59B-parameter Mixture-of-Experts foundation model on Hugging Face under Apache 2.0 \u2014 releasing not just weights but also training data recipes, full training code, hyperparameters, training logs, intermediate checkpoints, and evaluation code."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/deepseek-v4-flash-high?cost=cost-per-task",
   "title": "DeepSeek V4 Flash (high) - Intelligence, Performance & Price Analysis",
   "published_at": null,
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": null,
   "organization_evidence": [
    "![DeepSeek V4 Flash (Reasoning, High Effort) logo](/img/logos/deepseek_small.svg)",
    "# DeepSeek V4 Flash (Reasoning, High Effort) Intelligence, Performance & Price Analysis"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Released April 2026",
    "# DeepSeek V4 Flash (Reasoning, High Effort) Intelligence, Performance & Price Analysis"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": null
  },
  {
   "url": "https://fananews.com/language/en/china-launches-global-ai-model-to-enhance-meteorological-services",
   "title": "China Launches Global AI Model to Enhance Meteorological Services",
   "published_at": "2026-07-18T13:30:47",
   "organization": "China Meteorological Administration",
   "model": null,
   "release_date": "2026-07-18",
   "access_type": null,
   "license": null,
   "significance": "trained on 50 million tokens of meteorological service data and integrated with authoritative weather datasets to support weather forecasting, risk assessment and meteorological services",
   "organization_evidence": [
    "##### China Launches Global AI Model to Enhance Meteorological Services",
    "Beijing, July 18 (QNA) \u2013 The China Meteorological Administration (CMA) has launched an open-source large language model-based Al meteorological service system, making it available to developers and researchers to expand its use in weather applications and foster innovation in the field."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "##### China Launches Global AI Model to Enhance Meteorological Services",
    "Beijing, July 18 (QNA) \u2013 The China Meteorological Administration (CMA) has launched an open-source large language model-based Al meteorological service system, making it available to developers and researchers to expand its use in weather applications and foster innovation in the field."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "##### China Launches Global AI Model to Enhance Meteorological Services",
    "The CMA added that the model has been trained on 50 million tokens of meteorological service data and integrated with authoritative weather datasets to support weather forecasting, risk assessment and meteorological services."
   ]
  },
  {
   "url": "https://analyticsvidhya.com/blog/2026/07/thinking-machines-inkling",
   "title": "Complete Guide to Thinking Machines Inkling",
   "published_at": "2026-07-20T06:37:00",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "A multimodal MoE model with 975B parameters, 41B active parameters, and a 1M-token context window designed as a customizable foundation for multimodal reasoning, agentic AI, coding, tool use, audio and vision tasks, and domain-specific fine-tuning.",
   "organization_evidence": [
    "# Complete Guide to Thinking Machines Inkling",
    "Thinking Machines Lab has unveiled Inkling, its first general-purpose open-weights foundation model."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## What is Thinking Machines Inkling?",
    "Released on July 15, 2026, Inkling is available under the Apache 2.0 license, allowing commercial use, modification, and redistribution."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Complete Guide to Thinking Machines Inkling",
    "It is a multimodal [MoE](https://www.analyticsvidhya.com/blog/2024/12/mixture-of-experts-models/) model with 975B parameters, 41B active parameters, and a 1M-token context window."
   ]
  },
  {
   "url": "https://mindstackreview.com/articles/2026-07-27-moonshot-ai-opens-kimi-k3-the-world-s-largest-open-weight-model-for-free-downloa",
   "title": "Moonshot releases Kimi K3 weights, dropping a 2.8-trillion-parameter open model into enterprise buyers' laps",
   "published_at": "2026-07-27T00:00:00",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-27",
   "access_type": null,
   "license": null,
   "significance": "2.8-trillion-parameter model, the largest open-weight system ever published",
   "organization_evidence": [
    "Enterprise",
    "# Moonshot releases Kimi K3 weights, dropping a 2.8-trillion-parameter open model into enterprise buyers' laps",
    "Moonshot AI pushed the full weights of Kimi K3 to HuggingFace on July 27, dropping a 2.8-trillion-parameter model, the largest open-weight system ever published according to Bloomberg and VentureBeat, into the public domain under a modified MIT license."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Enterprise",
    "# Moonshot releases Kimi K3 weights, dropping a 2.8-trillion-parameter open model into enterprise buyers' laps",
    "The Beijing startup's July 27 weight release makes the largest open-weight model ever built free to download \u2014 and forces a rethink of what closed-model contracts are actually worth.",
    "Moonshot AI pushed the full weights of Kimi K3 to HuggingFace on July 27, dropping a 2.8-trillion-parameter model, the largest open-weight system ever published according to Bloomberg and VentureBeat, into the public domain under a modified MIT license."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Enterprise",
    "# Moonshot releases Kimi K3 weights, dropping a 2.8-trillion-parameter open model into enterprise buyers' laps",
    "Moonshot AI pushed the full weights of Kimi K3 to HuggingFace on July 27, dropping a 2.8-trillion-parameter model, the largest open-weight system ever published according to Bloomberg and VentureBeat, into the public domain under a modified MIT license."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/glm-5-1-non-reasoning-vs-kimi-k2-5-non-reasoning",
   "title": "GLM-5.1 (Non-reasoning) vs Kimi K2.5 (Non-reasoning): Model Comparison",
   "published_at": null,
   "organization": "Z AI",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "744B, 40B active at inference time",
   "organization_evidence": [
    "Comparison between GLM-5.1 (Non-reasoning) and Kimi K2.5 (Non-reasoning) across intelligence, price, speed, context window and more.",
    "| Creator | ![Z AI](/img/logos/zai_small.svg)[](https://z.ai/) | ![Kimi](/img/logos/kimi_small.png)[](https://www.moonshot.cn/) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Comparison between GLM-5.1 (Non-reasoning) and Kimi K2.5 (Non-reasoning) across intelligence, price, speed, context window and more.",
    "| Release Date | April, 2026 | January, 2026 | GLM-5.1 (Non-reasoning) has a more recent release date than Kimi K2.5 (Non-reasoning) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Comparison between GLM-5.1 (Non-reasoning) and Kimi K2.5 (Non-reasoning) across intelligence, price, speed, context window and more.",
    "| Parameters | 744B, 40B active at inference time | 1000B, 32B active at inference time | GLM-5.1 (Non-reasoning) is smaller than Kimi K2.5 (Non-reasoning) |"
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/glm-5-1-non-reasoning-vs-kimi-k2-5-non-reasoning",
   "title": "GLM-5.1 (Non-reasoning) vs Kimi K2.5 (Non-reasoning): Model Comparison",
   "published_at": null,
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-01",
   "access_type": null,
   "license": null,
   "significance": "1000B, 32B active at inference time",
   "organization_evidence": [
    "Comparison between GLM-5.1 (Non-reasoning) and Kimi K2.5 (Non-reasoning) across intelligence, price, speed, context window and more.",
    "| Creator | ![Z AI](/img/logos/zai_small.svg)[](https://z.ai/) | ![Kimi](/img/logos/kimi_small.png)[](https://www.moonshot.cn/) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Comparison between GLM-5.1 (Non-reasoning) and Kimi K2.5 (Non-reasoning) across intelligence, price, speed, context window and more.",
    "| Release Date | April, 2026 | January, 2026 | GLM-5.1 (Non-reasoning) has a more recent release date than Kimi K2.5 (Non-reasoning) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Comparison between GLM-5.1 (Non-reasoning) and Kimi K2.5 (Non-reasoning) across intelligence, price, speed, context window and more.",
    "| Parameters | 744B, 40B active at inference time | 1000B, 32B active at inference time | GLM-5.1 (Non-reasoning) is smaller than Kimi K2.5 (Non-reasoning) |"
   ]
  },
  {
   "url": "https://codeoxi.com/blog/thinking-machines-inkling",
   "title": "Thinking Machines Inkling Explained: Open Weights 2026",
   "published_at": "2026-08-05T00:00:00",
   "organization": "Thinking Machines Lab",
   "model": "Inkling",
   "release_date": "2026-07-15",
   "access_type": "open-weight",
   "license": "Apache 2.0",
   "significance": "a 975-billion-parameter mixture-of-experts model with 41 billion active parameters per forward pass, trained on 45 trillion tokens spanning text, images, audio and video, with a one-million-token context window",
   "organization_evidence": [
    "**Thinking Machines Inkling** is the first model from Mira Murati's Thinking Machines Lab, released on July 15, 2026 under an Apache 2.0 licence.",
    "Thinking Machines Inkling"
   ],
   "model_evidence": [
    "**Thinking Machines Inkling** is the first model from Mira Murati's Thinking Machines Lab, released on July 15, 2026 under an Apache 2.0 licence.",
    "Thinking Machines Inkling"
   ],
   "release_date_evidence": [
    "**Thinking Machines Inkling** is the first model from Mira Murati's Thinking Machines Lab, released on July 15, 2026 under an Apache 2.0 licence.",
    "Thinking Machines Inkling"
   ],
   "access_type_evidence": [
    "**Thinking Machines Inkling** is the first model from Mira Murati's Thinking Machines Lab, released on July 15, 2026 under an Apache 2.0 licence.",
    "Thinking Machines Inkling is an open-weight, natively multimodal foundation model built as a customization base rather than a finished product.",
    "Thinking Machines Inkling"
   ],
   "license_evidence": [
    "**Thinking Machines Inkling** is the first model from Mira Murati's Thinking Machines Lab, released on July 15, 2026 under an Apache 2.0 licence.",
    "Thinking Machines Inkling"
   ],
   "significance_evidence": [
    "It is a 975-billion-parameter mixture-of-experts model with 41 billion active parameters per forward pass, trained on 45 trillion tokens spanning text, images, audio and video, with a one-million-token context window.",
    "Thinking Machines Inkling"
   ]
  },
  {
   "url": "https://shadowfetch.com/news-room",
   "title": "AI News Room \u2014 Shadowfetch",
   "published_at": "2026-08-06T13:51:12",
   "organization": "IBM",
   "model": null,
   "release_date": "2026-04-29",
   "access_type": null,
   "license": null,
   "significance": "a multilingual open-model family built around stronger tool calling",
   "organization_evidence": [
    "AI industry news, dated and sourced. Every article here cites at least three primary sources you can check yourself \u2014 no aggregated summaries, no speculation.",
    "IBM ships Granite 4.1, a multilingual open-model family built around stronger tool calling",
    "Released April 29, 2026 under Apache 2.0, the Granite 4.1 generation pairs 3B, 8B, and 30B dense checkpoints with a post-training pipeline that IBM says sharpens instruction following and tool use across 12 languages."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "AI industry news, dated and sourced. Every article here cites at least three primary sources you can check yourself \u2014 no aggregated summaries, no speculation.",
    "IBM ships Granite 4.1, a multilingual open-model family built around stronger tool calling",
    "Released April 29, 2026 under Apache 2.0, the Granite 4.1 generation pairs 3B, 8B, and 30B dense checkpoints with a post-training pipeline that IBM says sharpens instruction following and tool use across 12 languages."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "AI industry news, dated and sourced. Every article here cites at least three primary sources you can check yourself \u2014 no aggregated summaries, no speculation.",
    "IBM ships Granite 4.1, a multilingual open-model family built around stronger tool calling",
    "Released April 29, 2026 under Apache 2.0, the Granite 4.1 generation pairs 3B, 8B, and 30B dense checkpoints with a post-training pipeline that IBM says sharpens instruction following and tool use across 12 languages."
   ]
  },
  {
   "url": "https://techcrunch.com/snippet/3080028/alpamayo-is-not-a-condiment-but-a-whole-new-family-of-open-source-ai-models",
   "title": "Alpamayo is not a condiment, but a whole new family of open source AI models",
   "published_at": "2026-01-05T22:25:05",
   "organization": "Nvidia",
   "model": null,
   "release_date": "2026-01-05",
   "access_type": null,
   "license": null,
   "significance": "A whole new family of open source AI models, simulation tools, and datasets for training physical robots and vehicles designed to help autonomous vehicles reason through complex driving situations.",
   "organization_evidence": [
    "# Alpamayo is not a condiment, but a whole new family of open source AI models",
    "Nvidia keynotes, always led by CEO Jensen Huang, are famous for its dozens of announcements."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Alpamayo is not a condiment, but a whole new family of open source AI models",
    "And the keynote at CES 2026 has been no different."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Alpamayo is not a condiment, but a whole new family of open source AI models",
    "This is a new family of open source AI models, simulation tools, and datasets for training physical robots and vehicles that are designed to help autonomous vehicles reason through complex driving situations."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/claude-sonnet-4-6-non-reasoning-low-effort-vs-glm-5-non-reasoning",
   "title": "Claude Sonnet 4.6 (Non-reasoning, Low Effort) vs GLM-5 (Non-reasoning): Model Comparison",
   "published_at": null,
   "organization": "Z AI",
   "model": null,
   "release_date": "2026-02",
   "access_type": null,
   "license": null,
   "significance": "Context window of 200k tokens",
   "organization_evidence": [
    "# Claude Sonnet 4.6 (Non-reasoning, Low Effort) vs. GLM-5 (Non-reasoning)",
    "| Creator | ![Anthropic](/img/logos/anthropic_small.svg)[](https://www.anthropic.com/) | ![Z AI](/img/logos/zai_small.svg)[](https://z.ai/) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Claude Sonnet 4.6 (Non-reasoning, Low Effort) vs. GLM-5 (Non-reasoning)",
    "| Release Date | February, 2026 | February, 2026 | Claude Sonnet 4.6 (Non-reasoning, Low Effort) has a more recent release date than GLM-5 (Non-reasoning) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Claude Sonnet 4.6 (Non-reasoning, Low Effort) vs. GLM-5 (Non-reasoning)",
    "| Context Window | 1000k tokens (~1500 A4 pages of size 12 Arial font) | 200k tokens (~300 A4 pages of size 12 Arial font) | Claude Sonnet 4.6 (Non-reasoning, Low Effort) is larger than GLM-5 (Non-reasoning) |"
   ]
  },
  {
   "url": "https://gadgetfee.com/tech-trends-innovations/signal-s-fast-track-ai-launches-highlight-china-s-innovation-power",
   "title": "Signal\u2019s Fast-Track AI Launches Highlight China\u2019s Innovation Power - GadgetFee",
   "published_at": "2026-07-15T06:19:18",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor",
   "organization_evidence": [
    "From late April to mid-June 2026, Chinese research labs introduced four major models: DeepSeek V4, MiniMax M3, Kimi K2.7-Code, and GLM-5.2.",
    "Over the past two years, China\u2019s open-weight AI landscape has expanded from a single lab to four major players: DeepSeek, Z.ai, Moonshot, and Alibaba."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "APR 24",
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "APR 24",
    "**DeepSeek V4 (Pro + Flash)**1.6T total / 49B active MoE, 1M context, MIT \u2014 resets the price floor"
   ]
  },
  {
   "url": "https://gadgetfee.com/tech-trends-innovations/signal-s-fast-track-ai-launches-highlight-china-s-innovation-power",
   "title": "Signal\u2019s Fast-Track AI Launches Highlight China\u2019s Innovation Power - GadgetFee",
   "published_at": "2026-07-15T06:19:18",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "cheap 1M-token context, native multimodal, modified-MIT",
   "organization_evidence": [
    "From late April to mid-June 2026, Chinese research labs introduced four major models: DeepSeek V4, MiniMax M3, Kimi K2.7-Code, and GLM-5.2."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "JUN 01",
    "**MiniMax M3**cheap 1M-token context, native multimodal, modified-MIT"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "JUN 01",
    "**MiniMax M3**cheap 1M-token context, native multimodal, modified-MIT"
   ]
  },
  {
   "url": "https://gadgetfee.com/tech-trends-innovations/signal-s-fast-track-ai-launches-highlight-china-s-innovation-power",
   "title": "Signal\u2019s Fast-Track AI Launches Highlight China\u2019s Innovation Power - GadgetFee",
   "published_at": "2026-07-15T06:19:18",
   "organization": "Moonshot",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "agent-run specialist, ~30% fewer thinking tokens than K2.6",
   "organization_evidence": [
    "From late April to mid-June 2026, Chinese research labs introduced four major models: DeepSeek V4, MiniMax M3, Kimi K2.7-Code, and GLM-5.2.",
    "Over the past two years, China\u2019s open-weight AI landscape has expanded from a single lab to four major players: DeepSeek, Z.ai, Moonshot, and Alibaba."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "JUN 13",
    "**Kimi K2.7-Code (Moonshot)**agent-run specialist, ~30% fewer thinking tokens than K2.6"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "JUN 13",
    "**Kimi K2.7-Code (Moonshot)**agent-run specialist, ~30% fewer thinking tokens than K2.6"
   ]
  },
  {
   "url": "https://gadgetfee.com/tech-trends-innovations/signal-s-fast-track-ai-launches-highlight-china-s-innovation-power",
   "title": "Signal\u2019s Fast-Track AI Launches Highlight China\u2019s Innovation Power - GadgetFee",
   "published_at": "2026-07-15T06:19:18",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "753B MoE, MIT, top open-weight on Artificial Analysis index",
   "organization_evidence": [
    "From late April to mid-June 2026, Chinese research labs introduced four major models: DeepSeek V4, MiniMax M3, Kimi K2.7-Code, and GLM-5.2.",
    "Over the past two years, China\u2019s open-weight AI landscape has expanded from a single lab to four major players: DeepSeek, Z.ai, Moonshot, and Alibaba."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "JUN 13\u201316",
    "**GLM-5.2 (Z.ai)**753B MoE, MIT, top open-weight on Artificial Analysis index"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "JUN 13\u201316",
    "**GLM-5.2 (Z.ai)**753B MoE, MIT, top open-weight on Artificial Analysis index"
   ]
  },
  {
   "url": "https://codersera.com/blog/ring-2-6-1t-ant-group-trillion-parameter-reasoning-model-2026",
   "title": "Ring-2.6-1T: Ant Group Trillion-Param Reasoning",
   "published_at": "2026-05-15T12:34:31",
   "organization": "inclusionAI",
   "model": null,
   "release_date": "2026-05-08",
   "access_type": null,
   "license": null,
   "significance": "trillion-parameter Mixture-of-Experts reasoning model with ~1T total parameters and roughly 63B active per token",
   "organization_evidence": [
    "Ring-2.6-1T: Ant Group's Open Trillion-Parameter Reasoning Model (Benchmarks, How It Compares, Can You Run It)",
    "Ant Group's inclusionAI shipped Ring-2.6-1T, a trillion-parameter open-weights reasoning MoE."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Ring-2.6-1T: Ant Group's Open Trillion-Parameter Reasoning Model (Benchmarks, How It Compares, Can You Run It)",
    "**Quick answer.** Ring-2.6-1T is inclusionAI's (Ant Group) open-weights, MIT-licensed trillion-parameter Mixture-of-Experts reasoning model, released around May 8, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Ring-2.6-1T: Ant Group's Open Trillion-Parameter Reasoning Model (Benchmarks, How It Compares, Can You Run It)",
    "**Quick answer.** Ring-2.6-1T is inclusionAI's (Ant Group) open-weights, MIT-licensed trillion-parameter Mixture-of-Experts reasoning model, released around May 8, 2026."
   ]
  },
  {
   "url": "https://dev.to/tekmag/the-2026-ai-model-release-race-every-major-llm-launch-you-need-to-know-3je5",
   "title": "The 2026 AI Model Release Race: Every Major LLM Launch You Need to Know",
   "published_at": "2026-07-01T01:43:04",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "Matches GPT-5.5 on several coding benchmarks with MoE architecture and ~370B total parameters.",
   "organization_evidence": [
    "This is your complete field guide to every major launch, ranked by real-world impact, benchmarked where it matters, and contextualized with the Stanford HAI AI Index 2026 report that dropped in April.",
    "## Q2 2026: The Breakneck Pace",
    "**DeepSeek V4** deserves special attention.",
    "The Chinese lab's V4-Pro model uses Mixture-of-Experts with ~370B total parameters (37B active per token) and matches GPT-5.5 on several coding benchmarks while being **fully open-weight** \u2014 you can run it locally or self-host."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "This is your complete field guide to every major launch, ranked by real-world impact, benchmarked where it matters, and contextualized with the Stanford HAI AI Index 2026 report that dropped in April.",
    "## Q2 2026: The Breakneck Pace",
    "- **Open-source surged:** DeepSeek V4-Pro (Apr 24), MiniMax M3 (Jun 1 \u2014 first open-weight triple-frontier model), GLM-5.2 (Jun 16), and Kimi K2.7 Code (Jun 12) all shipped."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "This is your complete field guide to every major launch, ranked by real-world impact, benchmarked where it matters, and contextualized with the Stanford HAI AI Index 2026 report that dropped in April.",
    "## Q2 2026: The Breakneck Pace",
    "The Chinese lab's V4-Pro model uses Mixture-of-Experts with ~370B total parameters (37B active per token) and matches GPT-5.5 on several coding benchmarks while being **fully open-weight** \u2014 you can run it locally or self-host."
   ]
  },
  {
   "url": "https://dev.to/tekmag/the-2026-ai-model-release-race-every-major-llm-launch-you-need-to-know-3je5",
   "title": "The 2026 AI Model Release Race: Every Major LLM Launch You Need to Know",
   "published_at": "2026-07-01T01:43:04",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "First open-weight model to deliver three frontier capabilities simultaneously: text reasoning, multimodal understanding, and audio processing.",
   "organization_evidence": [
    "This is your complete field guide to every major launch, ranked by real-world impact, benchmarked where it matters, and contextualized with the Stanford HAI AI Index 2026 report that dropped in April.",
    "## June 2026: The Hottest Month Yet",
    "## Open-Source: The Silent Revolution",
    "**MiniMax M3** (June 1) is the first open-weight model to deliver three frontier capabilities simultaneously: text reasoning, multimodal understanding, and audio processing \u2014 all in a single model."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "This is your complete field guide to every major launch, ranked by real-world impact, benchmarked where it matters, and contextualized with the Stanford HAI AI Index 2026 report that dropped in April.",
    "## June 2026: The Hottest Month Yet",
    "- **Open-source surged:** DeepSeek V4-Pro (Apr 24), MiniMax M3 (Jun 1 \u2014 first open-weight triple-frontier model), GLM-5.2 (Jun 16), and Kimi K2.7 Code (Jun 12) all shipped."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "This is your complete field guide to every major launch, ranked by real-world impact, benchmarked where it matters, and contextualized with the Stanford HAI AI Index 2026 report that dropped in April.",
    "## June 2026: The Hottest Month Yet",
    "## Open-Source: The Silent Revolution",
    "**MiniMax M3** (June 1) is the first open-weight model to deliver three frontier capabilities simultaneously: text reasoning, multimodal understanding, and audio processing \u2014 all in a single model."
   ]
  },
  {
   "url": "https://dev.to/tekmag/the-2026-ai-model-release-race-every-major-llm-launch-you-need-to-know-3je5",
   "title": "The 2026 AI Model Release Race: Every Major LLM Launch You Need to Know",
   "published_at": "2026-07-01T01:43:04",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-06-12",
   "access_type": null,
   "license": null,
   "significance": "Matches GPT-5.5 on SWE-bench Lite while being fully open-weight and particularly strong on Chinese-language coding documentation.",
   "organization_evidence": [
    "This is your complete field guide to every major launch, ranked by real-world impact, benchmarked where it matters, and contextualized with the Stanford HAI AI Index 2026 report that dropped in April.",
    "## June 2026: The Hottest Month Yet",
    "## Open-Source: The Silent Revolution",
    "**Kimi K2.7 Code** (June 12) from Moonshot AI matches GPT-5.5 on SWE-bench Lite while being fully open-weight \u2014 particularly strong on Chinese-language coding documentation."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "This is your complete field guide to every major launch, ranked by real-world impact, benchmarked where it matters, and contextualized with the Stanford HAI AI Index 2026 report that dropped in April.",
    "## June 2026: The Hottest Month Yet",
    "- **Open-source surged:** DeepSeek V4-Pro (Apr 24), MiniMax M3 (Jun 1 \u2014 first open-weight triple-frontier model), GLM-5.2 (Jun 16), and Kimi K2.7 Code (Jun 12) all shipped."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "This is your complete field guide to every major launch, ranked by real-world impact, benchmarked where it matters, and contextualized with the Stanford HAI AI Index 2026 report that dropped in April.",
    "## June 2026: The Hottest Month Yet",
    "## Open-Source: The Silent Revolution",
    "**Kimi K2.7 Code** (June 12) from Moonshot AI matches GPT-5.5 on SWE-bench Lite while being fully open-weight \u2014 particularly strong on Chinese-language coding documentation."
   ]
  },
  {
   "url": "https://dev.to/tekmag/the-2026-ai-model-release-race-every-major-llm-launch-you-need-to-know-3je5",
   "title": "The 2026 AI Model Release Race: Every Major LLM Launch You Need to Know",
   "published_at": "2026-07-01T01:43:04",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-06-16",
   "access_type": null,
   "license": null,
   "significance": "Strongest Chinese open-source model on English and Chinese benchmarks combined, scoring competitively with GPT-5.5 on MMLU-Pro.",
   "organization_evidence": [
    "This is your complete field guide to every major launch, ranked by real-world impact, benchmarked where it matters, and contextualized with the Stanford HAI AI Index 2026 report that dropped in April.",
    "## June 2026: The Hottest Month Yet",
    "## Open-Source: The Silent Revolution",
    "**GLM-5.2** (June 16) from Z.ai is the strongest Chinese open-source model on English and Chinese benchmarks combined, scoring competitively with GPT-5.5 on MMLU-Pro while requiring significantly less compute for inference."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "This is your complete field guide to every major launch, ranked by real-world impact, benchmarked where it matters, and contextualized with the Stanford HAI AI Index 2026 report that dropped in April.",
    "## June 2026: The Hottest Month Yet",
    "- **Open-source surged:** DeepSeek V4-Pro (Apr 24), MiniMax M3 (Jun 1 \u2014 first open-weight triple-frontier model), GLM-5.2 (Jun 16), and Kimi K2.7 Code (Jun 12) all shipped."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "This is your complete field guide to every major launch, ranked by real-world impact, benchmarked where it matters, and contextualized with the Stanford HAI AI Index 2026 report that dropped in April.",
    "## June 2026: The Hottest Month Yet",
    "## Open-Source: The Silent Revolution",
    "**GLM-5.2** (June 16) from Z.ai is the strongest Chinese open-source model on English and Chinese benchmarks combined, scoring competitively with GPT-5.5 on MMLU-Pro while requiring significantly less compute for inference."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/qwen3-5-0-8b-non-reasoning-vs-gpt-5-3-codex",
   "title": "Qwen3.5 0.8B (Non-reasoning) vs GPT-5.3 Codex (xhigh): Model Comparison",
   "published_at": null,
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-03",
   "access_type": null,
   "license": null,
   "significance": "Smaller than GPT-5.3 Codex (xhigh) with 262k token context window, image input support, and open source weights",
   "organization_evidence": [
    "# Qwen3.5 0.8B (Non-reasoning) vs. GPT-5.3 Codex (xhigh)",
    "| Creator | ![Alibaba](/img/logos/alibaba_small.svg)[](https://www.alibabacloud.com/) | ![OpenAI](/img/logos/openai_small.svg)[](https://openai.com/) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Qwen3.5 0.8B (Non-reasoning) vs. GPT-5.3 Codex (xhigh)",
    "| Release Date | March, 2026 | February, 2026 | Qwen3.5 0.8B (Non-reasoning) has a more recent release date than GPT-5.3 Codex (xhigh) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Qwen3.5 0.8B (Non-reasoning) vs. GPT-5.3 Codex (xhigh)",
    "| Context Window | 262k tokens (~393 A4 pages of size 12 Arial font) | 400k tokens (~600 A4 pages of size 12 Arial font) | Qwen3.5 0.8B (Non-reasoning) is smaller than GPT-5.3 Codex (xhigh) |",
    "| Image Input Support | Yes | Yes | Both Qwen3.5 0.8B (Non-reasoning) and GPT-5.3 Codex (xhigh) have image input support |",
    "| Open Source (Weights) | Yes[](https://huggingface.co/Qwen/Qwen3.5-0.8B) | No | Qwen3.5 0.8B (Non-reasoning) is open source while GPT-5.3 Codex (xhigh) is proprietary |"
   ]
  },
  {
   "url": "https://wpnews.pro/news/mark-zuckerberg-says-meta-will-release-weights-of-muse-spark-1-2-announces-30b",
   "title": "Mark Zuckerberg Says Meta Will Release Weights of Muse Spark 1.2, Announces 30B Muse Glimmer Model For Local Use",
   "published_at": "2026-08-10T10:27:42",
   "organization": "Meta",
   "model": null,
   "release_date": "2026-08-10",
   "access_type": null,
   "license": null,
   "significance": "Meta's frontier coding and reasoning model and fifth most powerful model on the Artificial Analysis Intelligence Index",
   "organization_evidence": [
    "Mark Zuckerberg Says Meta Will Release Weights of Muse Spark 1.2, Announces 30B Muse Glimmer Model For Local Use",
    "Meta CEO Mark Zuckerberg announced on X that Meta is releasing the weights of its frontier coding and reasoning model Muse Spark 1.2, currently the fifth most powerful model on the Artificial Analysis Intelligence Index, and a smaller 30 billion parameter dense model called Muse Glimmer designed for local hardware."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Mark Zuckerberg Says Meta Will Release Weights of Muse Spark 1.2, Announces 30B Muse Glimmer Model For Local Use",
    "read5 min views30 publishedAug 10, 2026"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Mark Zuckerberg Says Meta Will Release Weights of Muse Spark 1.2, Announces 30B Muse Glimmer Model For Local Use",
    "Meta CEO Mark Zuckerberg announced on X that Meta is releasing the weights of its frontier coding and reasoning model Muse Spark 1.2, currently the fifth most powerful model on the Artificial Analysis Intelligence Index, and a smaller 30 billion parameter dense model called Muse Glimmer designed for local hardware."
   ]
  },
  {
   "url": "https://wpnews.pro/news/mark-zuckerberg-says-meta-will-release-weights-of-muse-spark-1-2-announces-30b",
   "title": "Mark Zuckerberg Says Meta Will Release Weights of Muse Spark 1.2, Announces 30B Muse Glimmer Model For Local Use",
   "published_at": "2026-08-10T10:27:42",
   "organization": "Meta",
   "model": null,
   "release_date": "2026-08-10",
   "access_type": null,
   "license": null,
   "significance": "A smaller 30 billion parameter dense model designed for local hardware and agentic tasks",
   "organization_evidence": [
    "Mark Zuckerberg Says Meta Will Release Weights of Muse Spark 1.2, Announces 30B Muse Glimmer Model For Local Use",
    "Meta CEO Mark Zuckerberg announced on X that Meta is releasing the weights of its frontier coding and reasoning model Muse Spark 1.2, currently the fifth most powerful model on the Artificial Analysis Intelligence Index, and a smaller 30 billion parameter dense model called Muse Glimmer designed for local hardware."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Mark Zuckerberg Says Meta Will Release Weights of Muse Spark 1.2, Announces 30B Muse Glimmer Model For Local Use",
    "read5 min views30 publishedAug 10, 2026"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Mark Zuckerberg Says Meta Will Release Weights of Muse Spark 1.2, Announces 30B Muse Glimmer Model For Local Use",
    "Meta CEO Mark Zuckerberg announced on X that Meta is releasing the weights of its frontier coding and reasoning model Muse Spark 1.2, currently the fifth most powerful model on the Artificial Analysis Intelligence Index, and a smaller 30 billion parameter dense model called Muse Glimmer designed for local hardware."
   ]
  },
  {
   "url": "https://ownyourmind.ai/journal/open-frontier-catches-up",
   "title": "Three Bets on Open Weights",
   "published_at": "2026-04-26T00:00:00",
   "organization": "Moonshot",
   "model": null,
   "release_date": "2026-04-20",
   "access_type": null,
   "license": null,
   "significance": "A one-trillion-parameter mixture-of-experts model with 32 billion active parameters per token, a 256K context window, and native multimodal support for text, image, and experimental video.",
   "organization_evidence": [
    "Three Bets on Open Weights",
    "Moonshot shipped Kimi K2.6 on 20 April."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Three Bets on Open Weights",
    "Moonshot shipped Kimi K2.6 on 20 April."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "What Kimi K2.6 Actually Is",
    "Kimi K2.6 went public on 20 April 2026."
   ]
  },
  {
   "url": "https://ownyourmind.ai/journal/open-frontier-catches-up",
   "title": "Three Bets on Open Weights",
   "published_at": "2026-04-26T00:00:00",
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-04-16",
   "access_type": null,
   "license": null,
   "significance": "Open weights, MoE with 35 billion total and 3 billion active parameters per token, 256K native context, genuinely multimodal across text, image, and video.",
   "organization_evidence": [
    "Three Bets on Open Weights",
    "Alibaba shipped Qwen 3.6 across three SKUs between 16 and 22 April, holding the flagship closed-weights as it has been since Qwen2-Max in June 2024."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Qwen 3.6: The Bet That Doesn\u2019t Match The Pattern",
    "Alibaba shipped three models in the Qwen 3.6 family across April.",
    "Qwen3.6-35B-A3B landed on 16 April under Apache 2.0."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Qwen 3.6: The Bet That Doesn\u2019t Match The Pattern",
    "Qwen3.6-35B-A3B landed on 16 April under Apache 2.0.",
    "Open weights, MoE with 35 billion total and 3 billion active parameters per token, 256K native context (extensible to 1 million with YaRN scaling), genuinely multimodal across text, image, and video."
   ]
  },
  {
   "url": "https://ownyourmind.ai/journal/open-frontier-catches-up",
   "title": "Three Bets on Open Weights",
   "published_at": "2026-04-26T00:00:00",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "V4-Pro is 1.6 trillion total parameters with 49 billion active per token, V4-Flash is 284 billion total with 13 billion active, both having a 1 million token context window.",
   "organization_evidence": [
    "Three Bets on Open Weights",
    "DeepSeek shipped V4 on 24 April, open-sourced same day, with Huawei integration baked in."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "DeepSeek V4: Open At The Frontier, Cheap As Chips",
    "DeepSeek V4 Preview shipped on 24 April 2026, open-sourced the same day under MIT."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "DeepSeek V4: Open At The Frontier, Cheap As Chips",
    "DeepSeek V4 Preview shipped on 24 April 2026, open-sourced the same day under MIT.",
    "Two models.",
    "V4-Pro is 1.6 trillion total parameters with 49 billion active per token.",
    "V4-Flash is 284 billion total with 13 billion active.",
    "Both have a 1 million token context window."
   ]
  },
  {
   "url": "https://artificialanalysis.ai/models/comparisons/claude-sonnet-4-6-non-reasoning-low-effort-vs-glm-5",
   "title": "Claude Sonnet 4.6 (Non-reasoning, Low Effort) vs GLM-5 (Reasoning): Model Comparison",
   "published_at": null,
   "organization": "Z AI",
   "model": null,
   "release_date": "2026-02",
   "access_type": null,
   "license": null,
   "significance": "Reasoning model with open source weights and 200k context window",
   "organization_evidence": [
    "Comparison between Claude Sonnet 4.6 (Non-reasoning, Low Effort) and GLM-5 (Reasoning) across intelligence, price, speed, context window and more.",
    "| Creator | ![Anthropic](/img/logos/anthropic_small.svg)[](https://www.anthropic.com/) | ![Z AI](/img/logos/zai_small.svg)[](https://z.ai/) | |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Comparison between Claude Sonnet 4.6 (Non-reasoning, Low Effort) and GLM-5 (Reasoning) across intelligence, price, speed, context window and more.",
    "| Release Date | February, 2026 | February, 2026 | Claude Sonnet 4.6 (Non-reasoning, Low Effort) has a more recent release date than GLM-5 (Reasoning) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Comparison between Claude Sonnet 4.6 (Non-reasoning, Low Effort) and GLM-5 (Reasoning) across intelligence, price, speed, context window and more.",
    "| Context Window | 1000k tokens (~1500 A4 pages of size 12 Arial font) | 200k tokens (~300 A4 pages of size 12 Arial font) | Claude Sonnet 4.6 (Non-reasoning, Low Effort) is larger than GLM-5 (Reasoning) |",
    "| Open Source (Weights) | No | Yes[](https://huggingface.co/zai-org/GLM-5) | GLM-5 (Reasoning) is open source while Claude Sonnet 4.6 (Non-reasoning, Low Effort) is proprietary |"
   ]
  },
  {
   "url": "https://wmo.int/media/news-from-members/fenghe-goes-open-source-global-initiative-kicks",
   "title": "Fenghe goes open-source as Global Initiative kicks off",
   "published_at": "2026-07-17T14:05:17",
   "organization": "China Meteorological Administration",
   "model": null,
   "release_date": "2026-07-17",
   "access_type": null,
   "license": null,
   "significance": "the world's first open-source meteorological large language model with a hundred-billion-parameter scale",
   "organization_evidence": [
    "# Fenghe goes open-source as Global Initiative kicks off",
    "On July 17, at the 2026 World Artificial Intelligence Conference (WAIC) Meteorological Forum, the China Meteorological Administration (CMA) unveiled Fenghe, a Large Language Model-Based Al MeteorologicalService System, marking the commencement of its Global Open-Source Initiative."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Fenghe goes open-source as Global Initiative kicks off",
    " 17 July 2026",
    "On July 17, at the 2026 World Artificial Intelligence Conference (WAIC) Meteorological Forum, the China Meteorological Administration (CMA) unveiled Fenghe, a Large Language Model-Based Al MeteorologicalService System, marking the commencement of its Global Open-Source Initiative."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Fenghe goes open-source as Global Initiative kicks off",
    "Co-developed by the CMA Public Meteorological Service Centre, the Xiong'an Artificial Intelligence Research Institute, Z.AI Co., Ltd. and other entities, Fenghe is the world's first open-source meteorological large language model with a hundred-billion-parameter scale."
   ]
  },
  {
   "url": "https://ertas.ai/models",
   "title": "AI Models for Fine-Tuning - Ertas AI",
   "published_at": null,
   "organization": "Arcee AI",
   "model": null,
   "release_date": "2026-01",
   "access_type": null,
   "license": null,
   "significance": "a 400-billion parameter mixture-of-experts with 13B active parameters, 256 experts (4 active per token), 17 trillion training tokens, and 30-33 days of training on 2048 NVIDIA B300 chips",
   "organization_evidence": [
    "# AI Models",
    "Open-source models you can fine-tune with Ertas.",
    "[## Arcee Trinity Large\nReasoning\nArcee AI\nArcee AI's January 2026 release \u2014 a 400-billion parameter mixture-of-experts with 13B active parameters, 256 experts (4 active per token), 17 trillion training tokens, and 30-33 days of training on 2048 NVIDIA B300 chips. One of the few US-made frontier open-weight models in 2026 alongside OLMo 3 and GPT-OSS.\n400B-A13B](/models/arcee-trinity-large)"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# AI Models",
    "Open-source models you can fine-tune with Ertas.",
    "[## Arcee Trinity Large\nReasoning\nArcee AI\nArcee AI's January 2026 release \u2014 a 400-billion parameter mixture-of-experts with 13B active parameters, 256 experts (4 active per token), 17 trillion training tokens, and 30-33 days of training on 2048 NVIDIA B300 chips. One of the few US-made frontier open-weight models in 2026 alongside OLMo 3 and GPT-OSS.\n400B-A13B](/models/arcee-trinity-large)"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# AI Models",
    "Open-source models you can fine-tune with Ertas.",
    "[## Arcee Trinity Large\nReasoning\nArcee AI\nArcee AI's January 2026 release \u2014 a 400-billion parameter mixture-of-experts with 13B active parameters, 256 experts (4 active per token), 17 trillion training tokens, and 30-33 days of training on 2048 NVIDIA B300 chips. One of the few US-made frontier open-weight models in 2026 alongside OLMo 3 and GPT-OSS.\n400B-A13B](/models/arcee-trinity-large)"
   ]
  },
  {
   "url": "http://dlvr.it/TTHHnZ",
   "title": "Meituan open-sources LongCat-2.0, proving China can train frontier AI models on domestic chips",
   "published_at": "2026-06-30T12:57:21",
   "organization": "Meituan",
   "model": null,
   "release_date": "2026-06-30",
   "access_type": null,
   "license": null,
   "significance": "1.6 trillion parameter open-source large language model trained entirely on domestically produced Chinese chips",
   "organization_evidence": [
    "Article",
    "# Meituan open-sources LongCat-2.0, proving China can train frontier AI models on domestic chips",
    "Meituan released LongCat-2.0 on June 30, 2026, an open-source large language model with 1.6 trillion parameters trained entirely on domestically produced Chinese chips."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Article",
    "# Meituan open-sources LongCat-2.0, proving China can train frontier AI models on domestic chips",
    "Meituan released LongCat-2.0 on June 30, 2026, an open-source large language model with 1.6 trillion parameters trained entirely on domestically produced Chinese chips."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Article",
    "# Meituan open-sources LongCat-2.0, proving China can train frontier AI models on domestic chips",
    "Meituan released LongCat-2.0 on June 30, 2026, an open-source large language model with 1.6 trillion parameters trained entirely on domestically produced Chinese chips."
   ]
  },
  {
   "url": "https://thenewspaperdaily.com/complete-guide-to-thinking-machines-inkling",
   "title": "Complete Guide to Thinking Machines Inkling",
   "published_at": "2026-07-20T10:08:12",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "A multimodal MoE model with 975B parameters, 41B active parameters, and a 1M-token context window.",
   "organization_evidence": [
    "Thinking Machines Lab has unveiled Inkling, its first general-purpose open-weights foundation model.",
    "Developer | Thinking Machines Lab |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Released on July 15, 2026, Inkling is available under the Apache 2.0 license, allowing commercial use, modification, and redistribution."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "It is a multimodal MoE model with 975B parameters, 41B active parameters, and a 1M-token context window."
   ]
  },
  {
   "url": "https://penguinalley.com/en/blog/2026-08-10-below-the-ice-muse-glimmer-open-weight-comeback",
   "title": "Below the Ice \u2014 Muse Glimmer and Meta's Open-Weight Comeback \u00b7 Penguin Alley",
   "published_at": "2026-08-20T00:00:00",
   "organization": "Meta",
   "model": null,
   "release_date": "2026-08-10",
   "access_type": null,
   "license": null,
   "significance": "a 30B open-weight model with a 120,000-token context window, multimodal understanding, and a design built explicitly for local, agentic workflows",
   "organization_evidence": [
    "Meta's Muse Glimmer is a 30B open-weight model built for local, agentic, multimodal work \u2014 and it signals a full strategic reversal.",
    "# Below the Ice \u2014 Muse Glimmer and Meta's Open-Weight Comeback",
    "Meta released **Muse Glimmer** this week \u2014 a 30B open-weight model with a 120,000-token context window, multimodal understanding, and a design built explicitly for local, agentic workflows."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "skillPublished **2026-08-10**",
    "# Below the Ice \u2014 Muse Glimmer and Meta's Open-Weight Comeback"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Below the Ice \u2014 Muse Glimmer and Meta's Open-Weight Comeback",
    "Meta released **Muse Glimmer** this week \u2014 a 30B open-weight model with a 120,000-token context window, multimodal understanding, and a design built explicitly for local, agentic workflows."
   ]
  },
  {
   "url": "https://ai-nidox.site/deepseek-v4-drops-the-biggest-open-source-ai-model-ever-full-breakdown-april-2026",
   "title": "DeepSeek V4: Biggest Open-Source AI Model (April 2026)",
   "published_at": "2026-04-25T15:06:00",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "1.6 trillion parameters, 1 million token context window, and 80.6% on SWE-bench Verified",
   "organization_evidence": [
    "DeepSeek V4 Drops: The Biggest Open-Source AI Model Ever \u2014 Full Breakdown (April 2026)",
    "On **April 24, 2026**, Chinese AI startup DeepSeek quietly published two new models to Hugging Face under the MIT License and announced them on social media with a single sentence: *\u201cWelcome to the era of cost-effective 1M context length.\u201d*"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "DeepSeek V4 Drops: The Biggest Open-Source AI Model Ever \u2014 Full Breakdown (April 2026)",
    "BREAKING \u2014 April 24, 2026",
    "On **April 24, 2026**, Chinese AI startup DeepSeek quietly published two new models to Hugging Face under the MIT License and announced them on social media with a single sentence: *\u201cWelcome to the era of cost-effective 1M context length.\u201d*"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "DeepSeek V4 Drops: The Biggest Open-Source AI Model Ever \u2014 Full Breakdown (April 2026)",
    "A year after R1 shocked Silicon Valley, China\u2019s DeepSeek is back with something even bigger: 1.6 trillion parameters, a 1 million token context window, and pricing that undercuts every frontier model on Earth \u2014 by up to 7x.",
    "V4 Pro is the **largest open-weight model ever released to the public**, surpassing even Moonshot AI\u2019s Kimi K2.6 (1.1 trillion) and more than doubling DeepSeek\u2019s own V3.2 (685 billion parameters).",
    "V4 Pro scores **80.6% on SWE-bench Verified** \u2014 a coding benchmark where it comes within 0.2 percentage points of Anthropic\u2019s Claude Opus 4.6"
   ]
  },
  {
   "url": "https://felloai.com/cs/microsoft-mai-models",
   "title": "Microsoft Just Launched Its Own MAI Models",
   "published_at": "2026-06-03T02:46:00",
   "organization": "Microsoft",
   "model": "MAI family",
   "release_date": "2026-06-02",
   "access_type": "open weights",
   "license": null,
   "significance": "a family of seven in-house AI models spanning reasoning, coding, image generation, voice, and transcription",
   "organization_evidence": [
    "On June 2, 2026, at its Build developer conference, Microsoft announced a family of seven in-house AI models under the MAI brand, the most concrete step yet in the company\u2019s plan to stop being just the best distributor of someone else\u2019s frontier models.",
    "Microsoft used the Build 2026 keynote to unveil seven new models built by its MAI team (Microsoft AI Superintelligence)."
   ],
   "model_evidence": [
    "On June 2, 2026, at its Build developer conference, Microsoft announced a family of seven in-house AI models under the MAI brand, the most concrete step yet in the company\u2019s plan to stop being just the best distributor of someone else\u2019s frontier models.",
    "Microsoft used the Build 2026 keynote to unveil seven new models built by its MAI team (Microsoft AI Superintelligence)."
   ],
   "release_date_evidence": [
    "On June 2, 2026, at its Build developer conference, Microsoft announced a family of seven in-house AI models under the MAI brand, the most concrete step yet in the company\u2019s plan to stop being just the best distributor of someone else\u2019s frontier models."
   ],
   "access_type_evidence": [
    "MAI models are being distributed through OpenRouter, Fireworks, and Baseten in addition to Foundry, and for the first time developers will be able to tune the weights directly themselves.",
    "Open weights / self-tuning",
    "Open weights / self-tuning | yes, planned | no"
   ],
   "license_evidence": null,
   "significance_evidence": [
    "On June 2, 2026, at its Build developer conference, Microsoft announced a family of seven in-house AI models under the MAI brand, the most concrete step yet in the company\u2019s plan to stop being just the best distributor of someone else\u2019s frontier models.",
    "The lineup spans reasoning, coding, image generation, voice, and transcription, and it was unveiled during CEO Satya Nadella\u2019s keynote alongside new agents, silicon, and a repositioning of Windows itself."
   ]
  },
  {
   "url": "https://zylos.ai/research/2026-03-22-open-source-llm-fine-tuning-serving-ai-agent-platforms",
   "title": "Open-Source LLM Fine-Tuning and Serving Infrastructure for AI Agent Platforms | Zylos Research",
   "published_at": "2026-03-22T18:31:36",
   "organization": "Meta",
   "model": null,
   "release_date": "2026-03-22",
   "access_type": null,
   "license": null,
   "significance": "Meta's most significant architectural shift moving from dense transformers to Mixture of Experts with extreme context lengths and native multimodal support",
   "organization_evidence": [
    "Open-Source LLM Fine-Tuning and Serving Infrastructure for AI Agent Platforms",
    "### Llama 4: Meta's MoE Breakthrough",
    "Released in early 2026, Llama 4 represents Meta's most significant architectural shift \u2014 moving from dense transformers to Mixture of Experts (MoE)."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "2026-03-22",
    "Open-Source LLM Fine-Tuning and Serving Infrastructure for AI Agent Platforms",
    "Released in early 2026, Llama 4 represents Meta's most significant architectural shift \u2014 moving from dense transformers to Mixture of Experts (MoE)."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Open-Source LLM Fine-Tuning and Serving Infrastructure for AI Agent Platforms",
    "### Llama 4: Meta's MoE Breakthrough",
    "Released in early 2026, Llama 4 represents Meta's most significant architectural shift \u2014 moving from dense transformers to Mixture of Experts (MoE)."
   ]
  },
  {
   "url": "https://cornerforai.com/news/nvidia-releases-cosmos-3-and-nemotron-3-ultra-as-open-weight-ai-models-for-physical-world-and-language-tasks",
   "title": "NVIDIA Cosmos 3 and Nemotron 3 Ultra: Open AI Models Launch",
   "published_at": "2026-06-08T03:07:04",
   "organization": "NVIDIA",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "550-billion-parameter large language model using a hybrid Mamba-Transformer architecture combined with Mixture-of-Experts",
   "organization_evidence": [
    "## What is NVIDIA Nemotron 3 Ultra?",
    "**NVIDIA Nemotron 3 Ultra** is a **550-billion-parameter** large language model (LLM) announced by **Jensen Huang**, NVIDIA's chief executive officer, during his keynote at **Computex 2026** in Taipei."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## What is the significance of the Computex 2026 announcement?",
    "**Computex 2026**, one of the world's largest technology trade exhibitions, provided the stage for Jensen Huang to unveil Nemotron 3 Ultra.",
    "The back-to-back releases of **Cosmos 3** (June 5) and **Nemotron 3 Ultra** (Computex 2026) signal that NVIDIA is expanding its identity beyond graphics processing units (GPUs) into full-stack AI model development, spanning both language intelligence and physical-world understanding."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## What is NVIDIA Nemotron 3 Ultra?",
    "**NVIDIA Nemotron 3 Ultra** is a **550-billion-parameter** large language model (LLM) announced by **Jensen Huang**, NVIDIA's chief executive officer, during his keynote at **Computex 2026** in Taipei.",
    "It uses a hybrid **Mamba-Transformer** architecture combined with **Mixture-of-Experts (MoE)**, activating only **55 billion parameters** per forward pass \u2014 a design that preserves computational efficiency despite the model's massive total size."
   ]
  },
  {
   "url": "https://aiwiki.ai/wiki/kimi_k3",
   "title": "Kimi K3",
   "published_at": "2026-07-23T10:29:06",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-27",
   "access_type": null,
   "license": null,
   "significance": "sparse mixture-of-experts model with 2.78 trillion total parameters and 104.2 billion parameters activated for each token",
   "organization_evidence": [
    "# Kimi K3"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Kimi K3",
    "Moonshot made the model available through its hosted products on July 16, 2026 and published the weights, code, configuration, custom license, and a 47-page technical report on July 27."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Kimi K3",
    "K3 is a sparse [mixture-of-experts](/wiki/mixture_of_experts) model with 2.78 trillion total parameters and 104.2 billion parameters activated for each token."
   ]
  },
  {
   "url": "https://presenc.ai/research/april-2026-frontier-density-report",
   "title": "April 2026 Frontier Model Density Report: 12 Major Releases in 30 Days",
   "published_at": "2026-04-29T04:03:51",
   "organization": "Google",
   "model": null,
   "release_date": "2026-04-02",
   "access_type": null,
   "license": null,
   "significance": "Cleanest open-weight commercial license in the frontier tier; Llama replacement for many enterprises",
   "organization_evidence": [
    "## The April 2026 Release Wave at a Glance",
    "| Date | Release | Lab | Type | Why it matters for brand visibility |",
    "| Apr 2 | Gemma 4 family | Google | Open (Apache 2.0) | Cleanest open-weight commercial license in the frontier tier; Llama replacement for many enterprises |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## The April 2026 Release Wave at a Glance",
    "| Date | Release | Lab | Type | Why it matters for brand visibility |",
    "| Apr 2 | Gemma 4 family | Google | Open (Apache 2.0) | Cleanest open-weight commercial license in the frontier tier; Llama replacement for many enterprises |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## The April 2026 Release Wave at a Glance",
    "| Date | Release | Lab | Type | Why it matters for brand visibility |",
    "| Apr 2 | Gemma 4 family | Google | Open (Apache 2.0) | Cleanest open-weight commercial license in the frontier tier; Llama replacement for many enterprises |"
   ]
  },
  {
   "url": "https://presenc.ai/research/april-2026-frontier-density-report",
   "title": "April 2026 Frontier Model Density Report: 12 Major Releases in 30 Days",
   "published_at": "2026-04-29T04:03:51",
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-04-16",
   "access_type": null,
   "license": null,
   "significance": "Dense 27B fits in 18GB RAM; 3.6-Plus pushes 1M context",
   "organization_evidence": [
    "## The April 2026 Release Wave at a Glance",
    "| Date | Release | Lab | Type | Why it matters for brand visibility |",
    "| Apr 16 | Qwen 3.6-27B + 3.6-Plus | Alibaba | Open (Apache 2.0) + closed | Dense 27B fits in 18GB RAM; 3.6-Plus pushes 1M context |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## The April 2026 Release Wave at a Glance",
    "| Date | Release | Lab | Type | Why it matters for brand visibility |",
    "| Apr 16 | Qwen 3.6-27B + 3.6-Plus | Alibaba | Open (Apache 2.0) + closed | Dense 27B fits in 18GB RAM; 3.6-Plus pushes 1M context |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## The April 2026 Release Wave at a Glance",
    "| Date | Release | Lab | Type | Why it matters for brand visibility |",
    "| Apr 16 | Qwen 3.6-27B + 3.6-Plus | Alibaba | Open (Apache 2.0) + closed | Dense 27B fits in 18GB RAM; 3.6-Plus pushes 1M context |"
   ]
  },
  {
   "url": "https://presenc.ai/research/april-2026-frontier-density-report",
   "title": "April 2026 Frontier Model Density Report: 12 Major Releases in 30 Days",
   "published_at": "2026-04-29T04:03:51",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-04-18",
   "access_type": null,
   "license": null,
   "significance": "First Chinese open-weight model deployed at scale by Western developers",
   "organization_evidence": [
    "## The April 2026 Release Wave at a Glance",
    "| Date | Release | Lab | Type | Why it matters for brand visibility |",
    "| Apr 18 | Kimi K2.6 | Moonshot AI | Open (MIT-mod) | First Chinese open-weight model deployed at scale by Western developers |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## The April 2026 Release Wave at a Glance",
    "| Date | Release | Lab | Type | Why it matters for brand visibility |",
    "| Apr 18 | Kimi K2.6 | Moonshot AI | Open (MIT-mod) | First Chinese open-weight model deployed at scale by Western developers |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## The April 2026 Release Wave at a Glance",
    "| Date | Release | Lab | Type | Why it matters for brand visibility |",
    "| Apr 18 | Kimi K2.6 | Moonshot AI | Open (MIT-mod) | First Chinese open-weight model deployed at scale by Western developers |"
   ]
  },
  {
   "url": "https://presenc.ai/research/april-2026-frontier-density-report",
   "title": "April 2026 Frontier Model Density Report: 12 Major Releases in 30 Days",
   "published_at": "2026-04-29T04:03:51",
   "organization": "Zhipu AI",
   "model": null,
   "release_date": "2026-04-20",
   "access_type": null,
   "license": null,
   "significance": "744B MoE beats Claude Opus 4.6 and GPT-5.4 on SWE-Bench Pro",
   "organization_evidence": [
    "## The April 2026 Release Wave at a Glance",
    "| Date | Release | Lab | Type | Why it matters for brand visibility |",
    "| Apr 20 | GLM-5.1 | Zhipu AI | Open (MIT) | 744B MoE beats Claude Opus 4.6 and GPT-5.4 on SWE-Bench Pro |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## The April 2026 Release Wave at a Glance",
    "| Date | Release | Lab | Type | Why it matters for brand visibility |",
    "| Apr 20 | GLM-5.1 | Zhipu AI | Open (MIT) | 744B MoE beats Claude Opus 4.6 and GPT-5.4 on SWE-Bench Pro |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## The April 2026 Release Wave at a Glance",
    "| Date | Release | Lab | Type | Why it matters for brand visibility |",
    "| Apr 20 | GLM-5.1 | Zhipu AI | Open (MIT) | 744B MoE beats Claude Opus 4.6 and GPT-5.4 on SWE-Bench Pro |"
   ]
  },
  {
   "url": "https://presenc.ai/research/april-2026-frontier-density-report",
   "title": "April 2026 Frontier Model Density Report: 12 Major Releases in 30 Days",
   "published_at": "2026-04-29T04:03:51",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "1M context, frontier coding parity, aggressive pricing",
   "organization_evidence": [
    "## The April 2026 Release Wave at a Glance",
    "| Date | Release | Lab | Type | Why it matters for brand visibility |",
    "| Apr 24 | [DeepSeek V4](/llm-releases/deepseek-v4) Flash + Pro | DeepSeek | Open (V4 Flash) + closed (V4 Pro) | 1M context, frontier coding parity, aggressive pricing |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## The April 2026 Release Wave at a Glance",
    "| Date | Release | Lab | Type | Why it matters for brand visibility |",
    "| Apr 24 | [DeepSeek V4](/llm-releases/deepseek-v4) Flash + Pro | DeepSeek | Open (V4 Flash) + closed (V4 Pro) | 1M context, frontier coding parity, aggressive pricing |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## The April 2026 Release Wave at a Glance",
    "| Date | Release | Lab | Type | Why it matters for brand visibility |",
    "| Apr 24 | [DeepSeek V4](/llm-releases/deepseek-v4) Flash + Pro | DeepSeek | Open (V4 Flash) + closed (V4 Pro) | 1M context, frontier coding parity, aggressive pricing |"
   ]
  },
  {
   "url": "https://presenc.ai/research/april-2026-frontier-density-report",
   "title": "April 2026 Frontier Model Density Report: 12 Major Releases in 30 Days",
   "published_at": "2026-04-29T04:03:51",
   "organization": "Meta",
   "model": null,
   "release_date": "2026-04-26",
   "access_type": null,
   "license": null,
   "significance": "10M context (Scout); 400B MoE (Maverick) reach general availability",
   "organization_evidence": [
    "## The April 2026 Release Wave at a Glance",
    "| Date | Release | Lab | Type | Why it matters for brand visibility |",
    "| Apr 26 | Llama 4 Scout/Maverick GA | Meta | Open | 10M context (Scout); 400B MoE (Maverick) reach general availability |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## The April 2026 Release Wave at a Glance",
    "| Date | Release | Lab | Type | Why it matters for brand visibility |",
    "| Apr 26 | Llama 4 Scout/Maverick GA | Meta | Open | 10M context (Scout); 400B MoE (Maverick) reach general availability |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## The April 2026 Release Wave at a Glance",
    "| Date | Release | Lab | Type | Why it matters for brand visibility |",
    "| Apr 26 | Llama 4 Scout/Maverick GA | Meta | Open | 10M context (Scout); 400B MoE (Maverick) reach general availability |"
   ]
  },
  {
   "url": "https://presenc.ai/research/april-2026-frontier-density-report",
   "title": "April 2026 Frontier Model Density Report: 12 Major Releases in 30 Days",
   "published_at": "2026-04-29T04:03:51",
   "organization": "NVIDIA",
   "model": null,
   "release_date": "2026-04-28",
   "access_type": null,
   "license": null,
   "significance": "First credible NVIDIA-led frontier model; agentic-stack default",
   "organization_evidence": [
    "## The April 2026 Release Wave at a Glance",
    "| Date | Release | Lab | Type | Why it matters for brand visibility |",
    "| Apr 28 | Nemotron 3 Nano Omni | NVIDIA | Open multimodal | First credible NVIDIA-led frontier model; agentic-stack default |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## The April 2026 Release Wave at a Glance",
    "| Date | Release | Lab | Type | Why it matters for brand visibility |",
    "| Apr 28 | Nemotron 3 Nano Omni | NVIDIA | Open multimodal | First credible NVIDIA-led frontier model; agentic-stack default |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## The April 2026 Release Wave at a Glance",
    "| Date | Release | Lab | Type | Why it matters for brand visibility |",
    "| Apr 28 | Nemotron 3 Nano Omni | NVIDIA | Open multimodal | First credible NVIDIA-led frontier model; agentic-stack default |"
   ]
  },
  {
   "url": "https://benchr.org/articles/open-weight-tier-right-now",
   "title": "The open-weight tier right now",
   "published_at": "2026-05-30T00:00:00",
   "organization": "Alibaba",
   "model": "Qwen3.6",
   "release_date": "2026-04",
   "access_type": "open-weight",
   "license": "Apache 2.0",
   "significance": "77.2% on SWE-bench Verified and 87.8% on GPQA Diamond",
   "organization_evidence": [
    "Alibaba shipped Qwen3.6 the same month",
    "Alibaba's current series, released April 2026, with model cards on [Hugging Face under the Qwen organization](https://huggingface.co/Qwen).",
    "# The open-weight tier right now: Llama 4, Mistral, Qwen, DeepSeek"
   ],
   "model_evidence": [
    "Alibaba shipped Qwen3.6 the same month",
    "Qwen3.6",
    "# The open-weight tier right now: Llama 4, Mistral, Qwen, DeepSeek"
   ],
   "release_date_evidence": [
    "Alibaba shipped Qwen3.6 the same month",
    "Alibaba's current series, released April 2026, with model cards on [Hugging Face under the Qwen organization](https://huggingface.co/Qwen).",
    "# The open-weight tier right now: Llama 4, Mistral, Qwen, DeepSeek"
   ],
   "access_type_evidence": [
    "The open-weight tier right now: Llama 4, Mistral, Qwen, DeepSeek",
    "# The open-weight tier right now: Llama 4, Mistral, Qwen, DeepSeek"
   ],
   "license_evidence": [
    "Both variants ship under Apache 2.0, the cleanest license in the lineup.",
    "Qwen3.6",
    "# The open-weight tier right now: Llama 4, Mistral, Qwen, DeepSeek"
   ],
   "significance_evidence": [
    "The 27B is the one to take seriously: its official card lists 77.2% on SWE-bench Verified and 87.8% on GPQA Diamond \u2014 numbers that would have been frontier-class a year ago, from a model that fits on one GPU.",
    "Qwen3.6",
    "# The open-weight tier right now: Llama 4, Mistral, Qwen, DeepSeek"
   ]
  },
  {
   "url": "https://benchr.org/articles/open-weight-tier-right-now",
   "title": "The open-weight tier right now",
   "published_at": "2026-05-30T00:00:00",
   "organization": "DeepSeek",
   "model": "DeepSeek-V4",
   "release_date": "2026-04-24",
   "access_type": "open-weight",
   "license": "MIT",
   "significance": "80.6% on SWE-bench Verified, 90.1% on GPQA Diamond, and 93.5% on LiveCodeBench",
   "organization_evidence": [
    "DeepSeek shipped V4 in April",
    "Released April 24, 2026, with the announcement and pricing at [api-docs.deepseek.com](https://api-docs.deepseek.com/news/news260424).",
    "# The open-weight tier right now: Llama 4, Mistral, Qwen, DeepSeek"
   ],
   "model_evidence": [
    "DeepSeek shipped V4 in April",
    "DeepSeek-V4",
    "# The open-weight tier right now: Llama 4, Mistral, Qwen, DeepSeek"
   ],
   "release_date_evidence": [
    "Released April 24, 2026, with the announcement and pricing at [api-docs.deepseek.com](https://api-docs.deepseek.com/news/news260424).",
    "DeepSeek-V4",
    "# The open-weight tier right now: Llama 4, Mistral, Qwen, DeepSeek"
   ],
   "access_type_evidence": [
    "The open-weight tier right now: Llama 4, Mistral, Qwen, DeepSeek",
    "# The open-weight tier right now: Llama 4, Mistral, Qwen, DeepSeek"
   ],
   "license_evidence": [
    "The license is MIT for the weights.",
    "DeepSeek-V4",
    "# The open-weight tier right now: Llama 4, Mistral, Qwen, DeepSeek"
   ],
   "significance_evidence": [
    "The official card puts V4-Pro at 80.6% on SWE-bench Verified, 90.1% on GPQA Diamond, and 93.5% on LiveCodeBench \u2014 the strongest published set in the open tier, and within a few points of closed flagships that cost ten times more per token.",
    "DeepSeek-V4",
    "# The open-weight tier right now: Llama 4, Mistral, Qwen, DeepSeek"
   ]
  },
  {
   "url": "https://venturebeat.com/technology/z-ai-launches-zcode-to-challenge-cursor-claude-code-and-github-copilot-in-ai-coding",
   "title": "Z.ai launches ZCode to challenge Cursor, Claude Code and GitHub Copilot in AI coding",
   "published_at": "2026-07-02T10:00:16",
   "organization": "Z.ai",
   "model": "GLM-5.2",
   "release_date": "2026-06-16",
   "access_type": "open-source weights",
   "license": "MIT license",
   "significance": "a 744-billion-parameter mixture-of-experts architecture with 40 billion active parameters and a genuine one-million-token context window",
   "organization_evidence": [
    "Z.ai released GLM-5.2 on June 16, first to its Coding Plan subscribers and subsequently as open-source weights under the MIT license on [Hugging Face](https://huggingface.co/zai-org/GLM-5) \u2014 a sequencing decision that prioritized distribution over the traditional benchmark-led launch.",
    "## **GLM-5.2, the open-source model trained entirely on Chinese chips, powers the whole experience**"
   ],
   "model_evidence": [
    "Z.ai released GLM-5.2 on June 16, first to its Coding Plan subscribers and subsequently as open-source weights under the MIT license on [Hugging Face](https://huggingface.co/zai-org/GLM-5) \u2014 a sequencing decision that prioritized distribution over the traditional benchmark-led launch.",
    "## **GLM-5.2, the open-source model trained entirely on Chinese chips, powers the whole experience**"
   ],
   "release_date_evidence": [
    "Z.ai released GLM-5.2 on June 16, first to its Coding Plan subscribers and subsequently as open-source weights under the MIT license on [Hugging Face](https://huggingface.co/zai-org/GLM-5) \u2014 a sequencing decision that prioritized distribution over the traditional benchmark-led launch.",
    "## **GLM-5.2, the open-source model trained entirely on Chinese chips, powers the whole experience**"
   ],
   "access_type_evidence": [
    "Z.ai released GLM-5.2 on June 16, first to its Coding Plan subscribers and subsequently as open-source weights under the MIT license on [Hugging Face](https://huggingface.co/zai-org/GLM-5) \u2014 a sequencing decision that prioritized distribution over the traditional benchmark-led launch.",
    "## **GLM-5.2, the open-source model trained entirely on Chinese chips, powers the whole experience**"
   ],
   "license_evidence": [
    "Z.ai released GLM-5.2 on June 16, first to its Coding Plan subscribers and subsequently as open-source weights under the MIT license on [Hugging Face](https://huggingface.co/zai-org/GLM-5) \u2014 a sequencing decision that prioritized distribution over the traditional benchmark-led launch.",
    "## **GLM-5.2, the open-source model trained entirely on Chinese chips, powers the whole experience**"
   ],
   "significance_evidence": [
    "GLM-5.2 is a 744-billion-parameter mixture-of-experts architecture with 40 billion active parameters, a genuine one-million-token context window \u2014 five times the 200K limit on its predecessor \u2014 and training on 28.5 trillion tokens.",
    "## **GLM-5.2, the open-source model trained entirely on Chinese chips, powers the whole experience**"
   ]
  },
  {
   "url": "https://theclawstreetjournal.com/ai-frontier/2026/06/28/prc-ai-watch.html",
   "title": "PRC AI Watch: Zhipu\u2019s GLM-5.2 Triggers New DeepSeek Moment as Washington Tightens the Vise",
   "published_at": "2026-06-28T13:00:00",
   "organization": "Zhipu AI",
   "model": "GLM-5.2",
   "release_date": "2026-06-13",
   "access_type": "open-weight",
   "license": "MIT",
   "significance": "first Chinese open-weight model to rank in the global top three on a major AI benchmark, at one-tenth the cost of comparable U.S. frontier models",
   "organization_evidence": [
    "Model Watch",
    "| Model | Lab | Open Weight | Key Benchmark | Est. API Cost vs. GPT-5 |",
    "| GLM-5.2 | Zhipu AI | Yes (MIT) | #2 Code Arena, BridgeBench 42.8 | ~1/10th |",
    "Zhipu AI\u2019s GLM-5.2 arrived June 13 as the first Chinese open-weight model to rank in the global top three on a major AI benchmark, at one-tenth the cost of comparable U.S. frontier models."
   ],
   "model_evidence": [
    "Model Watch",
    "| Model | Lab | Open Weight | Key Benchmark | Est. API Cost vs. GPT-5 |",
    "| GLM-5.2 | Zhipu AI | Yes (MIT) | #2 Code Arena, BridgeBench 42.8 | ~1/10th |",
    "Zhipu AI\u2019s GLM-5.2 arrived June 13 as the first Chinese open-weight model to rank in the global top three on a major AI benchmark, at one-tenth the cost of comparable U.S. frontier models."
   ],
   "release_date_evidence": [
    "Zhipu AI\u2019s GLM-5.2 arrived June 13 as the first Chinese open-weight model to rank in the global top three on a major AI benchmark, at one-tenth the cost of comparable U.S. frontier models.",
    "Ai frontier",
    "# PRC AI Watch: Zhipu's GLM-5.2 Triggers New DeepSeek Moment as Washington Tightens the Vise"
   ],
   "access_type_evidence": [
    "Model Watch",
    "| Model | Lab | Open Weight | Key Benchmark | Est. API Cost vs. GPT-5 |",
    "| GLM-5.2 | Zhipu AI | Yes (MIT) | #2 Code Arena, BridgeBench 42.8 | ~1/10th |",
    "Zhipu AI\u2019s GLM-5.2 arrived June 13 as the first Chinese open-weight model to rank in the global top three on a major AI benchmark, at one-tenth the cost of comparable U.S. frontier models."
   ],
   "license_evidence": [
    "Model Watch",
    "| Model | Lab | Open Weight | Key Benchmark | Est. API Cost vs. GPT-5 |",
    "| GLM-5.2 | Zhipu AI | Yes (MIT) | #2 Code Arena, BridgeBench 42.8 | ~1/10th |"
   ],
   "significance_evidence": [
    "Zhipu AI\u2019s GLM-5.2 arrived June 13 as the first Chinese open-weight model to rank in the global top three on a major AI benchmark, at one-tenth the cost of comparable U.S. frontier models.",
    "Ai frontier",
    "# PRC AI Watch: Zhipu's GLM-5.2 Triggers New DeepSeek Moment as Washington Tightens the Vise"
   ]
  },
  {
   "url": "https://webpronews.com/mira-muratis-inkling-challenges-closed-ai-giants-with-open-multimodal-power",
   "title": "Mira Murati\u2019s Inkling Challenges Closed AI Giants With Open Multimodal Power",
   "published_at": "2026-07-16T01:02:04",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "a 975B-parameter open-weights multimodal model with 41B active parameters and a 1M token context supporting controllable reasoning effort, strong agentic coding, and native audio-vision support",
   "organization_evidence": [
    "# Mira Murati\u2019s Inkling Challenges Closed AI Giants With Open Multimodal Power",
    "Thinking Machines Lab, founded by ex-OpenAI CTO Mira Murati, released Inkling, a 975B-parameter open-weights multimodal model with 41B active parameters and a 1M token context."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Mira Murati\u2019s Inkling Challenges Closed AI Giants With Open Multimodal Power",
    "Wednesday, July 15, 2026",
    "Thinking Machines shared those figures in its July 15, 2026, post."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Mira Murati\u2019s Inkling Challenges Closed AI Giants With Open Multimodal Power",
    "Thinking Machines Lab, founded by ex-OpenAI CTO Mira Murati, released Inkling, a 975B-parameter open-weights multimodal model with 41B active parameters and a 1M token context."
   ]
  },
  {
   "url": "https://globaltechcouncil.org/ai/what-is-kimi-k3",
   "title": "What Is Kimi K3?",
   "published_at": "2026-07-20T07:15:42",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-27",
   "access_type": null,
   "license": null,
   "significance": "the world's largest open-weight AI model at 2.8 trillion parameters with a mixture-of-experts architecture and a one-million-token context window",
   "organization_evidence": [
    "# What Is Kimi K3?",
    "Moonshot AI, a Chinese startup founded just three years ago, released **KIMI K3**, a 2.8 trillion parameter model that immediately claimed the title of the world's largest open-weight AI system ever built."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# What Is Kimi K3?",
    "Full open-weight model files are scheduled for public release by July 27, 2026 under a Modified MIT license."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# What Is Kimi K3?",
    "Moonshot AI, a Chinese startup founded just three years ago, released **KIMI K3**, a 2.8 trillion parameter model that immediately claimed the title of the world's largest open-weight AI system ever built."
   ]
  },
  {
   "url": "https://intuitionlabs.ai/articles/chinese-open-source-llms-2025",
   "title": "An Overview of Chinese Open-Source LLMs (Sept 2025)",
   "published_at": "2025-09-29T00:00:00",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "preview release of DeepSeek V4, comprising V4-Pro (1.6T total parameters, 49B active) and V4-Flash (284B total, 13B active)",
   "organization_evidence": [
    "An Overview of Chinese Open-Source LLMs (Sept 2025)",
    "DeepSeek moved from V3.2-Exp to a V4 generation.",
    "On April 24, 2026 DeepSeek published a preview release of **DeepSeek V4**, comprising **V4-Pro** (1.6T total parameters, 49B active) and **V4-Flash** (284B total, 13B active), built on a sparse-attention design the company calls DeepSeek Sparse Attention; the company states that \"1M context is now the default across all official DeepSeek services\" ([[32]](https://api-docs.deepseek.com/news/news260424/#:~:text=1M%20context%20is%20now%20the%20default))."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "An Overview of Chinese Open-Source LLMs (Sept 2025)",
    "DeepSeek moved from V3.2-Exp to a V4 generation.",
    "On April 24, 2026 DeepSeek published a preview release of **DeepSeek V4**, comprising **V4-Pro** (1.6T total parameters, 49B active) and **V4-Flash** (284B total, 13B active), built on a sparse-attention design the company calls DeepSeek Sparse Attention; the company states that \"1M context is now the default across all official DeepSeek services\" ([[32]](https://api-docs.deepseek.com/news/news260424/#:~:text=1M%20context%20is%20now%20the%20default))."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "An Overview of Chinese Open-Source LLMs (Sept 2025)",
    "DeepSeek moved from V3.2-Exp to a V4 generation.",
    "On April 24, 2026 DeepSeek published a preview release of **DeepSeek V4**, comprising **V4-Pro** (1.6T total parameters, 49B active) and **V4-Flash** (284B total, 13B active), built on a sparse-attention design the company calls DeepSeek Sparse Attention; the company states that \"1M context is now the default across all official DeepSeek services\" ([[32]](https://api-docs.deepseek.com/news/news260424/#:~:text=1M%20context%20is%20now%20the%20default))."
   ]
  },
  {
   "url": "https://intuitionlabs.ai/articles/chinese-open-source-llms-2025",
   "title": "An Overview of Chinese Open-Source LLMs (Sept 2025)",
   "published_at": "2025-09-29T00:00:00",
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-02-16",
   "access_type": null,
   "license": null,
   "significance": "released open weights for Qwen3.5 models ranging from a 397B-A17B MoE flagship down to a 0.8B model",
   "organization_evidence": [
    "An Overview of Chinese Open-Source LLMs (Sept 2025)",
    "Alibaba shipped the Qwen3.5 and Qwen3.6 open-weight families.",
    "Between February 16 and March 2, 2026 Alibaba released open weights for Qwen3.5 models ranging from a 397B-A17B MoE flagship down to a 0.8B model, followed by **Qwen3.6-35B-A3B** (April 16) and **Qwen3.6-27B** (April 22), all under Apache 2.0: the Qwen team states that \"all our open-weight models are licensed under Apache 2.0\" ([[34]](https://github.com/QwenLM/Qwen3.6#:~:text=All%20our%20open%2Dweight%20models%20are%20licensed%20under%20Apache%202.0))."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "An Overview of Chinese Open-Source LLMs (Sept 2025)",
    "Alibaba shipped the Qwen3.5 and Qwen3.6 open-weight families.",
    "Between February 16 and March 2, 2026 Alibaba released open weights for Qwen3.5 models ranging from a 397B-A17B MoE flagship down to a 0.8B model, followed by **Qwen3.6-35B-A3B** (April 16) and **Qwen3.6-27B** (April 22), all under Apache 2.0: the Qwen team states that \"all our open-weight models are licensed under Apache 2.0\" ([[34]](https://github.com/QwenLM/Qwen3.6#:~:text=All%20our%20open%2Dweight%20models%20are%20licensed%20under%20Apache%202.0))."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "An Overview of Chinese Open-Source LLMs (Sept 2025)",
    "Alibaba shipped the Qwen3.5 and Qwen3.6 open-weight families.",
    "Between February 16 and March 2, 2026 Alibaba released open weights for Qwen3.5 models ranging from a 397B-A17B MoE flagship down to a 0.8B model, followed by **Qwen3.6-35B-A3B** (April 16) and **Qwen3.6-27B** (April 22), all under Apache 2.0: the Qwen team states that \"all our open-weight models are licensed under Apache 2.0\" ([[34]](https://github.com/QwenLM/Qwen3.6#:~:text=All%20our%20open%2Dweight%20models%20are%20licensed%20under%20Apache%202.0))."
   ]
  },
  {
   "url": "https://intuitionlabs.ai/articles/chinese-open-source-llms-2025",
   "title": "An Overview of Chinese Open-Source LLMs (Sept 2025)",
   "published_at": "2025-09-29T00:00:00",
   "organization": "Zhipu",
   "model": null,
   "release_date": "2026-04-08",
   "access_type": null,
   "license": null,
   "significance": "GLM-5.1 weights were subsequently opened",
   "organization_evidence": [
    "An Overview of Chinese Open-Source LLMs (Sept 2025)",
    "Zhipu (now branded Z.ai) iterated from GLM-4.6 to GLM-5.1.",
    "**GLM-4.6** (357B parameters, MIT license, 200K-token context) was published on Hugging Face shortly after this article's original date ([[36]](https://huggingface.co/zai-org/GLM-4.6))."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "An Overview of Chinese Open-Source LLMs (Sept 2025)",
    "Zhipu (now branded Z.ai) iterated from GLM-4.6 to GLM-5.1.",
    "It was followed by GLM-4.7 (December 2025) and **GLM-5** (February 2026, reported at 744B total and 40B active parameters); GLM-5.1 weights were subsequently opened on April 8, 2026, keeping Z.ai in the open-weight camp ([[37]](https://presenc.ai/research/zhipu-glm-model-lineage-2026))."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "An Overview of Chinese Open-Source LLMs (Sept 2025)",
    "Zhipu (now branded Z.ai) iterated from GLM-4.6 to GLM-5.1.",
    "It was followed by GLM-4.7 (December 2025) and **GLM-5** (February 2026, reported at 744B total and 40B active parameters); GLM-5.1 weights were subsequently opened on April 8, 2026, keeping Z.ai in the open-weight camp ([[37]](https://presenc.ai/research/zhipu-glm-model-lineage-2026))."
   ]
  },
  {
   "url": "https://intuitionlabs.ai/articles/chinese-open-source-llms-2025",
   "title": "An Overview of Chinese Open-Source LLMs (Sept 2025)",
   "published_at": "2025-09-29T00:00:00",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-06-12",
   "access_type": null,
   "license": null,
   "significance": "coding-focused Kimi K2.7-Code shipped with open weights on Hugging Face",
   "organization_evidence": [
    "An Overview of Chinese Open-Source LLMs (Sept 2025)",
    "Moonshot AI extended the Kimi K2 line into K2.6 and K2.7-Code.",
    "The coding-focused **Kimi K2.7-Code** shipped on June 12, 2026 with open weights on Hugging Face under a Modified MIT license; Moonshot reports a 21.8% improvement on its Kimi Code Bench v2 over K2.6 and states that K2.7-Code cuts reasoning token usage by roughly 30% relative to K2.6 ([[38]](https://devops.com/moonshot-ais-kimi-k2-7-code-targets-token-efficiency-in-agentic-coding/))."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "An Overview of Chinese Open-Source LLMs (Sept 2025)",
    "Moonshot AI extended the Kimi K2 line into K2.6 and K2.7-Code.",
    "The coding-focused **Kimi K2.7-Code** shipped on June 12, 2026 with open weights on Hugging Face under a Modified MIT license; Moonshot reports a 21.8% improvement on its Kimi Code Bench v2 over K2.6 and states that K2.7-Code cuts reasoning token usage by roughly 30% relative to K2.6 ([[38]](https://devops.com/moonshot-ais-kimi-k2-7-code-targets-token-efficiency-in-agentic-coding/))."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "An Overview of Chinese Open-Source LLMs (Sept 2025)",
    "Moonshot AI extended the Kimi K2 line into K2.6 and K2.7-Code.",
    "The coding-focused **Kimi K2.7-Code** shipped on June 12, 2026 with open weights on Hugging Face under a Modified MIT license; Moonshot reports a 21.8% improvement on its Kimi Code Bench v2 over K2.6 and states that K2.7-Code cuts reasoning token usage by roughly 30% relative to K2.6 ([[38]](https://devops.com/moonshot-ais-kimi-k2-7-code-targets-token-efficiency-in-agentic-coding/))."
   ]
  },
  {
   "url": "https://intuitionlabs.ai/articles/chinese-open-source-llms-2025",
   "title": "An Overview of Chinese Open-Source LLMs (Sept 2025)",
   "published_at": "2025-09-29T00:00:00",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-04-12",
   "access_type": null,
   "license": null,
   "significance": "self-evolving agent model that scores 56.22 on SWE-Pro and 57.0 on Terminal Bench 2",
   "organization_evidence": [
    "An Overview of Chinese Open-Source LLMs (Sept 2025)",
    "MiniMax pivoted its M-series toward agents and code.",
    "The successor **MiniMax M2.7** was open-sourced on April 12, 2026, with reported scores of 56.22% on SWE-Pro and 57.0% on Terminal Bench 2, figures MiniMax positions as the strongest among open-source models ([[40]](https://www.marktechpost.com/2026/04/12/minimax-just-open-sourced-minimax-m2-7-a-self-evolving-agent-model-that-scores-56-22-on-swe-pro-and-57-0-on-terminal-bench-2/))."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "An Overview of Chinese Open-Source LLMs (Sept 2025)",
    "MiniMax pivoted its M-series toward agents and code.",
    "The successor **MiniMax M2.7** was open-sourced on April 12, 2026, with reported scores of 56.22% on SWE-Pro and 57.0% on Terminal Bench 2, figures MiniMax positions as the strongest among open-source models ([[40]](https://www.marktechpost.com/2026/04/12/minimax-just-open-sourced-minimax-m2-7-a-self-evolving-agent-model-that-scores-56-22-on-swe-pro-and-57-0-on-terminal-bench-2/))."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "An Overview of Chinese Open-Source LLMs (Sept 2025)",
    "MiniMax pivoted its M-series toward agents and code.",
    "The successor **MiniMax M2.7** was open-sourced on April 12, 2026, with reported scores of 56.22% on SWE-Pro and 57.0% on Terminal Bench 2, figures MiniMax positions as the strongest among open-source models ([[40]](https://www.marktechpost.com/2026/04/12/minimax-just-open-sourced-minimax-m2-7-a-self-evolving-agent-model-that-scores-56-22-on-swe-pro-and-57-0-on-terminal-bench-2/))."
   ]
  },
  {
   "url": "https://freeainews.com/open-source/openai-gpt-oss-open-weight-2026",
   "title": "OpenAI gpt-oss Open-Weight Models: Full Guide 2026",
   "published_at": "2026-05-26T00:00:00",
   "organization": "OpenAI",
   "model": null,
   "release_date": "2026-05-12",
   "access_type": null,
   "license": null,
   "significance": "Mixture-of-Experts reasoning models that anyone can download, modify, and deploy.",
   "organization_evidence": [
    "In May 2026, OpenAI did something it had not done in approximately five years: it released model weights.",
    "The gpt-oss family -- short for GPT Open Source -- consists of two Mixture-of-Experts reasoning models that anyone can download, modify, and deploy."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "| Release Date | May 12, 2026 | May 12, 2026 |",
    "| Specification | gpt-oss-120b | gpt-oss-20b |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The gpt-oss family -- short for GPT Open Source -- consists of two Mixture-of-Experts reasoning models that anyone can download, modify, and deploy."
   ]
  },
  {
   "url": "https://startupfortune.com/poolsides-laguna-s-21-beats-bigger-rivals-on-coding-benchmarks",
   "title": "Poolside's Laguna S 2.1 Beats Bigger Rivals on Coding Benchmarks",
   "published_at": "2026-07-21T21:50:24",
   "organization": "Poolside",
   "model": null,
   "release_date": "2026-07-21",
   "access_type": null,
   "license": null,
   "significance": "A Mixture-of-Experts coding model with 118 billion total parameters (8 billion active per token) and a 1 million token context window that beats rivals several times its size.",
   "organization_evidence": [
    "Poolside released Laguna S 2.1 on July 21, 2026, an open-weight model built specifically for agentic coding.",
    "Poolside just shipped a 118 billion parameter coding model that beats rivals several times its size, and it fits on a single desktop GPU."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Poolside released Laguna S 2.1 on July 21, 2026, an open-weight model built specifically for agentic coding."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Poolside released Laguna S 2.1 on July 21, 2026, an open-weight model built specifically for agentic coding.",
    "It's a Mixture-of-Experts design with 118 billion total parameters but only 8 billion active per token, and it handles a context window up to 1 million tokens."
   ]
  },
  {
   "url": "https://edgen.tech/news/post/alibaba-open-sources-qwen38-27b-as-24-trillion-max-weights-debut",
   "title": "Alibaba open-sources Qwen3.8-27B as 2.4-trillion Max weights debut",
   "published_at": "2026-08-14T15:53:18",
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-08-13",
   "access_type": null,
   "license": null,
   "significance": "2.4-trillion-parameter frontier-class model with 95 billion active parameters and a 262,144-token context window.",
   "organization_evidence": [
    "Alibaba open-sources Qwen3.8-27B as 2.4-trillion Max weights debut",
    "Alibaba released the weights of its 2.4-trillion-parameter Qwen3.8-Max model, the first open-source frontier-class model from a Chinese lab, challenging closed rivals Fable 5 and GPT-5.6 Sol on agent benchmarks."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Alibaba open-sources Qwen3.8-27B as 2.4-trillion Max weights debut",
    "The Qwen3.8-2.4T-A95B, posted Aug. 13 on ModelScope and Hugging Face, activates 95 billion parameters per token and natively supports a 262,144-token context window, scalable to 1.01 million tokens."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Alibaba open-sources Qwen3.8-27B as 2.4-trillion Max weights debut",
    "Alibaba released the weights of its 2.4-trillion-parameter Qwen3.8-Max model, the first open-source frontier-class model from a Chinese lab, challenging closed rivals Fable 5 and GPT-5.6 Sol on agent benchmarks.",
    "The Qwen3.8-2.4T-A95B, posted Aug. 13 on ModelScope and Hugging Face, activates 95 billion parameters per token and natively supports a 262,144-token context window, scalable to 1.01 million tokens."
   ]
  },
  {
   "url": "https://edgen.tech/news/post/alibaba-open-sources-qwen38-27b-as-24-trillion-max-weights-debut",
   "title": "Alibaba open-sources Qwen3.8-27B as 2.4-trillion Max weights debut",
   "published_at": "2026-08-14T15:53:18",
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-08-14",
   "access_type": null,
   "license": null,
   "significance": "Natively multimodal dense model with 27 billion parameters.",
   "organization_evidence": [
    "Alibaba open-sources Qwen3.8-27B as 2.4-trillion Max weights debut",
    "The smaller Qwen3.8-27B, a natively multimodal dense model with 27 billion parameters, went live Aug. 14 and outperforms the prior Qwen3.7-Plus, Alibaba said."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Alibaba open-sources Qwen3.8-27B as 2.4-trillion Max weights debut",
    "The smaller Qwen3.8-27B, a natively multimodal dense model with 27 billion parameters, went live Aug. 14 and outperforms the prior Qwen3.7-Plus, Alibaba said."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Alibaba open-sources Qwen3.8-27B as 2.4-trillion Max weights debut",
    "The smaller Qwen3.8-27B, a natively multimodal dense model with 27 billion parameters, went live Aug. 14 and outperforms the prior Qwen3.7-Plus, Alibaba said."
   ]
  },
  {
   "url": "https://stillintheloop.com/articles/thinking-machines-inkling-open-weights",
   "title": "Thinking Machines releases Inkling, a frontier-scale model anyone can download",
   "published_at": "2026-07-16T14:57:42",
   "organization": "Thinking Machines",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "975-billion-parameter multimodal system",
   "organization_evidence": [
    "# Thinking Machines releases Inkling, a frontier-scale model anyone can download",
    "Thinking Machines Lab has released its first model, and the notable thing is not how good it is \u2014 it is who gets to have it."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The model, called Inkling, went out on July 15 with **open weights** under the permissive Apache 2.0 license, published alongside a full technical model card on Hugging Face.1",
    "# Thinking Machines releases Inkling, a frontier-scale model anyone can download"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "The lab's first model is a 975-billion-parameter multimodal system released under a permissive open license \u2014 and the company says plainly it \"is not the strongest overall model available today.\"",
    "# Thinking Machines releases Inkling, a frontier-scale model anyone can download",
    "At 975 billion parameters, it is a genuinely large model to put out under so permissive a license."
   ]
  },
  {
   "url": "https://aiweekly.co/alerts/moonshot-open-sources-kimi-k3-a-28t-parameter-moe-model",
   "title": "Moonshot open-sources Kimi K3, a 2.8T-parameter MoE model",
   "published_at": "2026-07-27T21:12:01",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-26",
   "access_type": null,
   "license": null,
   "significance": "a 2.8-trillion-parameter mixture-of-experts model with a million-token context window and native multimodal input",
   "organization_evidence": [
    "# Moonshot open-sources Kimi K3, a 2.8T-parameter MoE model",
    "- Moonshot AI released Kimi K3 as an open-weight model with 2.8 trillion total parameters and 104 billion active per token."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Moonshot open-sources Kimi K3, a 2.8T-parameter MoE model",
    "- Open weights posted July 26, 2026 after the July 17 launch through Kimi.com and the Kimi API, under the Kimi K3 License."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Moonshot open-sources Kimi K3, a 2.8T-parameter MoE model",
    "Moonshot AI just dropped [Kimi K3 on Hugging Face](https://huggingface.co/moonshotai/Kimi-K3), a 2.8-trillion-parameter mixture-of-experts model with a million-token context window and native multimodal input."
   ]
  },
  {
   "url": "https://felloai.com/de/microsoft-mai-models",
   "title": "Microsoft Just Launched Its Own MAI Models",
   "published_at": "2026-06-03T02:46:00",
   "organization": "Microsoft",
   "model": null,
   "release_date": "2026-06-02",
   "access_type": null,
   "license": null,
   "significance": "sparse Mixture of Experts model with 35 billion active parameters and roughly 1 trillion total parameters designed for long context reasoning, multi-step instructions, and code generation",
   "organization_evidence": [
    "Microsoft Just Launched Its Own MAI Models",
    "On June 2, 2026, at its Build developer conference, Microsoft announced a family of seven in-house AI models under the MAI brand, the most concrete step yet in the company\u2019s plan to stop being just the best distributor of someone else\u2019s frontier models."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "On June 2, 2026, at its Build developer conference, Microsoft announced a family of seven in-house AI models under the MAI brand, the most concrete step yet in the company\u2019s plan to stop being just the best distributor of someone else\u2019s frontier models.",
    "Microsoft Just Launched Its Own MAI Models"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "MAI-Thinking-1, the Reasoning Flagship",
    "MAI-Thinking-1 is a sparse Mixture of Experts model with 35 billion active parameters and roughly 1 trillion total parameters.",
    "Microsoft describes the target use cases plainly."
   ]
  },
  {
   "url": "https://newshive.geekybee.net/stories/873761a9-f500-4cf4-b513-f86f25dee76d",
   "title": "Open Source LLM Model Releases",
   "published_at": null,
   "organization": "Alibaba / Qwen team",
   "model": null,
   "release_date": "2026-03-22",
   "access_type": null,
   "license": null,
   "significance": "122-billion parameter mixture-of-experts model compressed into formats that can run on consumer hardware",
   "organization_evidence": [
    "Open Source LLM Model Releases",
    "On March 22nd, someone dropped an uncensored, aggressively abliterated GGUF version of Qwen3.5-122B-A10B \u2014 a 122-billion parameter mixture-of-experts model compressed into formats that can run on consumer hardware."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Open Source LLM Model Releases",
    "On March 22nd, someone dropped an uncensored, aggressively abliterated GGUF version of Qwen3.5-122B-A10B \u2014 a 122-billion parameter mixture-of-experts model compressed into formats that can run on consumer hardware."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Open Source LLM Model Releases",
    "On March 22nd, someone dropped an uncensored, aggressively abliterated GGUF version of Qwen3.5-122B-A10B \u2014 a 122-billion parameter mixture-of-experts model compressed into formats that can run on consumer hardware."
   ]
  },
  {
   "url": "https://newshive.geekybee.net/stories/873761a9-f500-4cf4-b513-f86f25dee76d",
   "title": "Open Source LLM Model Releases",
   "published_at": null,
   "organization": "Google",
   "model": null,
   "release_date": "2026-04-02",
   "access_type": null,
   "license": null,
   "significance": "can reason efficiently when prompted lightly",
   "organization_evidence": [
    "Open Source LLM Model Releases",
    "Google's Gemma 4 showed up in the community conversation on April 2nd, with practitioners noting it can reason efficiently when prompted lightly but will cheerfully burn ten-plus minutes of compute if you let it \u2014 a behavioural quirk that is either a feature or a liability depending on your use case."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Open Source LLM Model Releases",
    "Google's Gemma 4 showed up in the community conversation on April 2nd, with practitioners noting it can reason efficiently when prompted lightly but will cheerfully burn ten-plus minutes of compute if you let it \u2014 a behavioural quirk that is either a feature or a liability depending on your use case."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Open Source LLM Model Releases",
    "Google's Gemma 4 showed up in the community conversation on April 2nd, with practitioners noting it can reason efficiently when prompted lightly but will cheerfully burn ten-plus minutes of compute if you let it \u2014 a behavioural quirk that is either a feature or a liability depending on your use case."
   ]
  },
  {
   "url": "https://the-agent-report.com/2026/06/post-llama-open-source-landscape-june-2026",
   "title": "Life After Llama: The Open-Source LLM Ecosystem Taking Shape in Mid-2026",
   "published_at": "2026-06-09T08:00:00",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "combines three frontier capabilities in one package: top-tier coding, a 1M-token context window, and native multimodality",
   "organization_evidence": [
    "# Life After Llama: The Open-Source LLM Ecosystem Taking Shape in Mid-2026",
    "### MiniMax M3 \u2014 The Newcomer (June 1, 2026)",
    "MiniMax M3 \u2014 The Newcomer (June 1, 2026)"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Life After Llama: The Open-Source LLM Ecosystem Taking Shape in Mid-2026",
    "### MiniMax M3 \u2014 The Newcomer (June 1, 2026)",
    "Released just eight days ago, MiniMax M3 is the first open-weight model to combine three frontier capabilities in one package: top-tier coding, a 1M-token context window, and native multimodality."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Life After Llama: The Open-Source LLM Ecosystem Taking Shape in Mid-2026",
    "### MiniMax M3 \u2014 The Newcomer (June 1, 2026)",
    "Released just eight days ago, MiniMax M3 is the first open-weight model to combine three frontier capabilities in one package: top-tier coding, a 1M-token context window, and native multimodality."
   ]
  },
  {
   "url": "https://signalreads.com/articles/open-weight-ai-just-hit-28-trillion-parameters",
   "title": "Open-weight AI just hit 2.8 trillion parameters\u2026",
   "published_at": null,
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-21",
   "access_type": null,
   "license": null,
   "significance": "a 2.8 trillion parameter mixture-of-experts model, the largest open-weight release to date",
   "organization_evidence": [
    "# Open-weight AI just hit 2.8 trillion parameters\u2026",
    "> **Bottom line:** Moonshot AI released Kimi K3 on July 21, 2026 \u2014 a 2.8 trillion parameter mixture-of-experts model, the largest open-weight release to date, surpassing DeepSeek V3.2 and Llama 4 Behemoth."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Open-weight AI just hit 2.8 trillion parameters\u2026",
    "> **Bottom line:** Moonshot AI released Kimi K3 on July 21, 2026 \u2014 a 2.8 trillion parameter mixture-of-experts model, the largest open-weight release to date, surpassing DeepSeek V3.2 and Llama 4 Behemoth."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Open-weight AI just hit 2.8 trillion parameters\u2026",
    "> **Bottom line:** Moonshot AI released Kimi K3 on July 21, 2026 \u2014 a 2.8 trillion parameter mixture-of-experts model, the largest open-weight release to date, surpassing DeepSeek V3.2 and Llama 4 Behemoth."
   ]
  },
  {
   "url": "https://acaderesearch.com/china-open-weight-lead-america-ai-diffusion-gap",
   "title": "China\u2019s Open-Weight Lead Exposes America\u2019s Real AI Gap: Substrate, Not Capability",
   "published_at": "2026-07-30T21:09:56",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07",
   "access_type": null,
   "license": null,
   "significance": "frontier-class weight release",
   "organization_evidence": [
    "China\u2019s Open-Weight Lead Exposes America\u2019s Real AI Gap: Substrate, Not Capability",
    "the White House accused Moonshot AI of distilling a U.S. frontier model to build Kimi K3;"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "China\u2019s Open-Weight Lead Exposes America\u2019s Real AI Gap: Substrate, Not Capability",
    "Four months have passed.",
    "Kimi K3\u2019s weight release in late July and subsequent launches are not captured."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "China\u2019s Open-Weight Lead Exposes America\u2019s Real AI Gap: Substrate, Not Capability",
    "Kimi K3\u2019s weight release in late July and subsequent launches are not captured."
   ]
  },
  {
   "url": "https://youngju.dev/blog/2026-07-11-tencent-hy3-open-model.en",
   "title": "Tencent Hy3: reading a 295B open-weight MoE without the hype",
   "published_at": "2026-07-11T00:00:00",
   "organization": "Tencent",
   "model": null,
   "release_date": "2026-07-06",
   "access_type": null,
   "license": null,
   "significance": "It is a 295B-parameter Mixture-of-Experts (MoE), but only about 21B parameters actually fire for any given token.",
   "organization_evidence": [
    "Tencent Hy3: reading a 295B open-weight MoE without the hype",
    "On July 6, 2026, Tencent's Hy team (formerly Hunyuan) released Hy3 as open weights \u2014 the production version of the preview that shipped on April 23."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Tencent Hy3: reading a 295B open-weight MoE without the hype",
    "On July 6, 2026, Tencent's Hy team (formerly Hunyuan) released Hy3 as open weights \u2014 the production version of the preview that shipped on April 23."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Tencent Hy3: reading a 295B open-weight MoE without the hype",
    "The short version is this. It is a 295B-parameter Mixture-of-Experts (MoE), but only about 21B parameters actually fire for any given token."
   ]
  },
  {
   "url": "https://theaidude.net/blog/thinking-machines-inkling-muratis-first-open-model",
   "title": "Thinking Machines Inkling: Murati's First Open Model",
   "published_at": "2026-07-16T00:00:00",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "975-billion-parameter open-weight model that handles text, audio, video, and code",
   "organization_evidence": [
    "Thinking Machines Lab released Inkling on July 15, 2026, its first public model and the first thing the Mira Murati startup has shipped since it raised one of the largest seed rounds in the industry's history."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Thinking Machines Lab released Inkling on July 15, 2026, its first public model and the first thing the Mira Murati startup has shipped since it raised one of the largest seed rounds in the industry's history."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Inkling is a 975-billion-parameter open-weight model that handles text, audio, video, and code, and the company is releasing the weights rather than gating it behind an API-only endpoint."
   ]
  },
  {
   "url": "https://ertas.ai/blog/why-chinese-labs-dominate-open-source-ai",
   "title": "Why Chinese Labs Now Dominate Open-Source AI",
   "published_at": "2026-04-29T03:46:40",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-02",
   "access_type": null,
   "license": null,
   "significance": "trained on 8x Huawei Ascend H20 chips, reaching BenchLM 83",
   "organization_evidence": [
    "Why Chinese Labs Now Dominate Open-Source AI",
    "GLM-5 / 5.1 (Z.ai, BenchLM 83)",
    "GLM-5, released by Z.ai in February 2026, was reportedly trained on 8\u00d7 Huawei Ascend H20 chips."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Why Chinese Labs Now Dominate Open-Source AI",
    "GLM-5, released by Z.ai in February 2026, was reportedly trained on 8\u00d7 Huawei Ascend H20 chips."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Why Chinese Labs Now Dominate Open-Source AI",
    "GLM-5 / 5.1 (Z.ai, BenchLM 83)",
    "GLM-5, released by Z.ai in February 2026, was reportedly trained on 8\u00d7 Huawei Ascend H20 chips."
   ]
  },
  {
   "url": "https://localaimaster.com/blog/best-open-source-llms-2026",
   "title": "Best Open-Source LLMs (2026): Free Models Ranked",
   "published_at": "2026-02-04T00:00:00",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-06-17",
   "access_type": null,
   "license": null,
   "significance": "753B MoE / ~40B",
   "organization_evidence": [
    "Best Open Source LLMs 2026: Which One Should You Self-Host?",
    "## New: The June-July 2026 Giant-Model Wave",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **[GLM-5.2](/models/glm-5-2)** (Z.ai) | Jun 17 | 753B MoE / ~40B | MIT | **Big rigs, yes** \u2014 112 ready-made quants for llama.cpp/Ollama/LM Studio; think Mac Studio 512GB or EPYC + KTransformers, ~200-400GB quantized |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Best Open Source LLMs 2026: Which One Should You Self-Host?",
    "## New: The June-July 2026 Giant-Model Wave",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **[GLM-5.2](/models/glm-5-2)** (Z.ai) | Jun 17 | 753B MoE / ~40B | MIT | **Big rigs, yes** \u2014 112 ready-made quants for llama.cpp/Ollama/LM Studio; think Mac Studio 512GB or EPYC + KTransformers, ~200-400GB quantized |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Best Open Source LLMs 2026: Which One Should You Self-Host?",
    "## New: The June-July 2026 Giant-Model Wave",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **[GLM-5.2](/models/glm-5-2)** (Z.ai) | Jun 17 | 753B MoE / ~40B | MIT | **Big rigs, yes** \u2014 112 ready-made quants for llama.cpp/Ollama/LM Studio; think Mac Studio 512GB or EPYC + KTransformers, ~200-400GB quantized |"
   ]
  },
  {
   "url": "https://localaimaster.com/blog/best-open-source-llms-2026",
   "title": "Best Open-Source LLMs (2026): Free Models Ranked",
   "published_at": "2026-02-04T00:00:00",
   "organization": "Nvidia",
   "model": null,
   "release_date": "2026-06-04",
   "access_type": null,
   "license": null,
   "significance": "~550B / 55B",
   "organization_evidence": [
    "Best Open Source LLMs 2026: Which One Should You Self-Host?",
    "## New: The June-July 2026 Giant-Model Wave",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **Nemotron 3 Ultra** (Nvidia) | Jun 4 | ~550B / 55B | OpenMDW | Workstation-class multi-GPU; NVFP4 weights published |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Best Open Source LLMs 2026: Which One Should You Self-Host?",
    "## New: The June-July 2026 Giant-Model Wave",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **Nemotron 3 Ultra** (Nvidia) | Jun 4 | ~550B / 55B | OpenMDW | Workstation-class multi-GPU; NVFP4 weights published |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Best Open Source LLMs 2026: Which One Should You Self-Host?",
    "## New: The June-July 2026 Giant-Model Wave",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **Nemotron 3 Ultra** (Nvidia) | Jun 4 | ~550B / 55B | OpenMDW | Workstation-class multi-GPU; NVFP4 weights published |"
   ]
  },
  {
   "url": "https://localaimaster.com/blog/best-open-source-llms-2026",
   "title": "Best Open-Source LLMs (2026): Free Models Ranked",
   "published_at": "2026-02-04T00:00:00",
   "organization": null,
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "428B / ~23B",
   "organization_evidence": null,
   "model_evidence": null,
   "release_date_evidence": [
    "Best Open Source LLMs 2026: Which One Should You Self-Host?",
    "## New: The June-July 2026 Giant-Model Wave",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **MiniMax M3** | Jun 1 | 428B / ~23B | Open weights | Big-RAM rigs quantized (official MXFP8 + community GGUF) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Best Open Source LLMs 2026: Which One Should You Self-Host?",
    "## New: The June-July 2026 Giant-Model Wave",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **MiniMax M3** | Jun 1 | 428B / ~23B | Open weights | Big-RAM rigs quantized (official MXFP8 + community GGUF) |"
   ]
  },
  {
   "url": "https://localaimaster.com/blog/best-open-source-llms-2026",
   "title": "Best Open-Source LLMs (2026): Free Models Ranked",
   "published_at": "2026-02-04T00:00:00",
   "organization": "Tencent",
   "model": null,
   "release_date": "2026-07-06",
   "access_type": null,
   "license": null,
   "significance": "295B / 21B",
   "organization_evidence": [
    "Best Open Source LLMs 2026: Which One Should You Self-Host?",
    "## New: The June-July 2026 Giant-Model Wave",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **Hy3** (Tencent) | Jul 6 | 295B / 21B | Apache 2.0 | The most home-plausible giant \u2014 official FP8; big multi-GPU or unified-memory rigs |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Best Open Source LLMs 2026: Which One Should You Self-Host?",
    "## New: The June-July 2026 Giant-Model Wave",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **Hy3** (Tencent) | Jul 6 | 295B / 21B | Apache 2.0 | The most home-plausible giant \u2014 official FP8; big multi-GPU or unified-memory rigs |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Best Open Source LLMs 2026: Which One Should You Self-Host?",
    "## New: The June-July 2026 Giant-Model Wave",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **Hy3** (Tencent) | Jul 6 | 295B / 21B | Apache 2.0 | The most home-plausible giant \u2014 official FP8; big multi-GPU or unified-memory rigs |"
   ]
  },
  {
   "url": "https://localaimaster.com/blog/best-open-source-llms-2026",
   "title": "Best Open-Source LLMs (2026): Free Models Ranked",
   "published_at": "2026-02-04T00:00:00",
   "organization": "Thinking Machines",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "975B / 41B",
   "organization_evidence": [
    "Best Open Source LLMs 2026: Which One Should You Self-Host?",
    "## New: The June-July 2026 Giant-Model Wave",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **Inkling** (Thinking Machines) | Jul 15 | 975B / 41B | Apache 2.0 (reported) | Datacenter-class (~2TB at BF16); watch for distilled variants |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Best Open Source LLMs 2026: Which One Should You Self-Host?",
    "## New: The June-July 2026 Giant-Model Wave",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **Inkling** (Thinking Machines) | Jul 15 | 975B / 41B | Apache 2.0 (reported) | Datacenter-class (~2TB at BF16); watch for distilled variants |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Best Open Source LLMs 2026: Which One Should You Self-Host?",
    "## New: The June-July 2026 Giant-Model Wave",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **Inkling** (Thinking Machines) | Jul 15 | 975B / 41B | Apache 2.0 (reported) | Datacenter-class (~2TB at BF16); watch for distilled variants |"
   ]
  },
  {
   "url": "https://localaimaster.com/blog/best-open-source-llms-2026",
   "title": "Best Open-Source LLMs (2026): Free Models Ranked",
   "published_at": "2026-02-04T00:00:00",
   "organization": "Moonshot",
   "model": null,
   "release_date": "2026-07-16",
   "access_type": null,
   "license": null,
   "significance": "2.8T MoE / ~1.8% active",
   "organization_evidence": [
    "Best Open Source LLMs 2026: Which One Should You Self-Host?",
    "## New: The June-July 2026 Giant-Model Wave",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **[Kimi K3](/models/kimi-k3)** (Moonshot) | Jul 16 | 2.8T MoE / ~1.8% active | TBD | **No** \u2014 ~1.4TB even at native 4-bit; weights land ~Jul 27, datacenter-only |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Best Open Source LLMs 2026: Which One Should You Self-Host?",
    "## New: The June-July 2026 Giant-Model Wave",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **[Kimi K3](/models/kimi-k3)** (Moonshot) | Jul 16 | 2.8T MoE / ~1.8% active | TBD | **No** \u2014 ~1.4TB even at native 4-bit; weights land ~Jul 27, datacenter-only |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Best Open Source LLMs 2026: Which One Should You Self-Host?",
    "## New: The June-July 2026 Giant-Model Wave",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **[Kimi K3](/models/kimi-k3)** (Moonshot) | Jul 16 | 2.8T MoE / ~1.8% active | TBD | **No** \u2014 ~1.4TB even at native 4-bit; weights land ~Jul 27, datacenter-only |"
   ]
  },
  {
   "url": "https://localaimaster.com/blog/best-open-source-llms-2026",
   "title": "Best Open-Source LLMs (2026): Free Models Ranked",
   "published_at": "2026-02-04T00:00:00",
   "organization": "Google",
   "model": null,
   "release_date": "2026-06-03",
   "access_type": null,
   "license": null,
   "significance": "12B",
   "organization_evidence": [
    "Best Open Source LLMs 2026: Which One Should You Self-Host?",
    "## New: The June-July 2026 Giant-Model Wave",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **[Gemma 4 12B Unified](/models/gemma-4)** (Google) | Jun 3 | 12B | Apache 2.0 | **Yes, easily** \u2014 the one genuinely consumer-class release of the wave; text+audio+image input, 256K context |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Best Open Source LLMs 2026: Which One Should You Self-Host?",
    "## New: The June-July 2026 Giant-Model Wave",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **[Gemma 4 12B Unified](/models/gemma-4)** (Google) | Jun 3 | 12B | Apache 2.0 | **Yes, easily** \u2014 the one genuinely consumer-class release of the wave; text+audio+image input, 256K context |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Best Open Source LLMs 2026: Which One Should You Self-Host?",
    "## New: The June-July 2026 Giant-Model Wave",
    "| Model | Released | Size (total/active) | License | Can you actually run it? |",
    "| **[Gemma 4 12B Unified](/models/gemma-4)** (Google) | Jun 3 | 12B | Apache 2.0 | **Yes, easily** \u2014 the one genuinely consumer-class release of the wave; text+audio+image input, 256K context |"
   ]
  },
  {
   "url": "https://en.wedoany.com/shortnews/420681.html",
   "title": "US-based Poolside Releases Open-Weight Coding Model with 118 Billion Parameters",
   "published_at": null,
   "organization": "Poolside",
   "model": null,
   "release_date": "2026-07-22",
   "access_type": null,
   "license": null,
   "significance": "an open-weight coding model with 118 billion parameters, employing a mixture-of-experts architecture that activates 8 billion parameters per token",
   "organization_evidence": [
    "US-based Poolside Releases Open-Weight Coding Model with 118 Billion Parameters",
    "*en.Wedoany.com* Reported - US startup Poolside has released Laguna S 2.1, an open-weight coding model with 118 billion parameters, employing a mixture-of-experts architecture that activates 8 billion parameters per token."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "US-based Poolside Releases Open-Weight Coding Model with 118 Billion Parameters",
    "2026-07-22 10:27"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "US-based Poolside Releases Open-Weight Coding Model with 118 Billion Parameters",
    "*en.Wedoany.com* Reported - US startup Poolside has released Laguna S 2.1, an open-weight coding model with 118 billion parameters, employing a mixture-of-experts architecture that activates 8 billion parameters per token."
   ]
  },
  {
   "url": "https://heybeagle.com/blog/kimi-k3-is-open-weight-running-it-yourself-is-another-matter",
   "title": "Kimi K3 Is Open-Weight. Running It Yourself Is Another Matter.",
   "published_at": "2026-07-31T12:01:50",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-26",
   "access_type": null,
   "license": null,
   "significance": "flagship 2.8-trillion-parameter multimodal reasoning model",
   "organization_evidence": [
    "Kimi K3 Is Open-Weight. Running It Yourself Is Another Matter.",
    "Kimi K3 is Moonshot AI's flagship 2.8-trillion-parameter multimodal reasoning model.",
    "Moonshot AI released the weights on July 26, 2026, roughly a day ahead of the July 27 target the company had previously communicated."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Kimi K3 Is Open-Weight. Running It Yourself Is Another Matter.",
    "Moonshot AI dropped Kimi K3's 2.8-trillion-parameter weights on July 26.",
    "Moonshot AI released the weights on July 26, 2026, roughly a day ahead of the July 27 target the company had previously communicated."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Kimi K3 Is Open-Weight. Running It Yourself Is Another Matter.",
    "Kimi K3 is Moonshot AI's flagship 2.8-trillion-parameter multimodal reasoning model."
   ]
  },
  {
   "url": "https://whatllm.org/blog/llm-releases-march-2026",
   "title": "New LLMs March 2026: GPT-5.4 Tied for #1. Nobody Talked About It.",
   "published_at": "2026-03-24T00:00:00",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-03-18",
   "access_type": null,
   "license": null,
   "significance": "Intelligence Index 49.62 with low hallucination rate at a competitive price",
   "organization_evidence": [
    "New LLMs March 2026: GPT-5.4 Tied for #1. Nobody Talked About It.",
    "MiniMax-M2.7",
    "MiniMax \u00b7 March 18 \u00b7 Open"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "New LLMs March 2026: GPT-5.4 Tied for #1. Nobody Talked About It.",
    "**MiniMax-M2.7** landed March 18 with Intelligence Index 49.62 at just $0.53 per million tokens.",
    "MiniMax \u00b7 March 18 \u00b7 Open"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "New LLMs March 2026: GPT-5.4 Tied for #1. Nobody Talked About It.",
    "**MiniMax-M2.7** landed March 18 with Intelligence Index 49.62 at just $0.53 per million tokens.",
    "Low\n\nHallucination rate",
    "The M2.7 is the best price-to-quality ratio in its tier."
   ]
  },
  {
   "url": "https://whatllm.org/blog/llm-releases-march-2026",
   "title": "New LLMs March 2026: GPT-5.4 Tied for #1. Nobody Talked About It.",
   "published_at": "2026-03-24T00:00:00",
   "organization": "Xiaomi",
   "model": null,
   "release_date": "2026-03-18",
   "access_type": null,
   "license": null,
   "significance": "Intelligence Index 49 and Elo 1426 on GDPval-AA for agentic tasks",
   "organization_evidence": [
    "New LLMs March 2026: GPT-5.4 Tied for #1. Nobody Talked About It.",
    "**MiMo-V2-Pro**, also March 18, from Xiaomi."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "New LLMs March 2026: GPT-5.4 Tied for #1. Nobody Talked About It.",
    "**MiMo-V2-Pro**, also March 18, from Xiaomi."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "New LLMs March 2026: GPT-5.4 Tied for #1. Nobody Talked About It.",
    "**MiMo-V2-Pro**, also March 18, from Xiaomi.",
    "Intelligence Index 49. Elo 1426 on GDPval-AA for agentic tasks."
   ]
  },
  {
   "url": "https://whatllm.org/blog/llm-releases-march-2026",
   "title": "New LLMs March 2026: GPT-5.4 Tied for #1. Nobody Talked About It.",
   "published_at": "2026-03-24T00:00:00",
   "organization": "Mistral AI",
   "model": null,
   "release_date": "2026-03-01",
   "access_type": null,
   "license": null,
   "significance": "119 billion total parameters with only 6.5 billion active parameters, supporting image and text inputs",
   "organization_evidence": [
    "New LLMs March 2026: GPT-5.4 Tied for #1. Nobody Talked About It.",
    "**Mistral Small 4**, Mistral (Apache 2.0)"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "New LLMs March 2026: GPT-5.4 Tied for #1. Nobody Talked About It.",
    "Every text-focused model that shipped in March 2026, ordered chronologically."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "New LLMs March 2026: GPT-5.4 Tied for #1. Nobody Talked About It.",
    "Mistral Small 4 is the one worth lingering on.",
    "119 billion total parameters with only 6.5 billion active."
   ]
  },
  {
   "url": "https://whatllm.org/blog/llm-releases-march-2026",
   "title": "New LLMs March 2026: GPT-5.4 Tied for #1. Nobody Talked About It.",
   "published_at": "2026-03-24T00:00:00",
   "organization": "NVIDIA",
   "model": null,
   "release_date": "2026-03-01",
   "access_type": null,
   "license": null,
   "significance": "120B total parameters, 12B active, open weights, Intelligence Index of 36",
   "organization_evidence": [
    "New LLMs March 2026: GPT-5.4 Tied for #1. Nobody Talked About It.",
    "**Nemotron 3 Super**, NVIDIA"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "New LLMs March 2026: GPT-5.4 Tied for #1. Nobody Talked About It.",
    "Every text-focused model that shipped in March 2026, ordered chronologically."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "New LLMs March 2026: GPT-5.4 Tied for #1. Nobody Talked About It.",
    "NVIDIA's Nemotron 3 Super tells a similar story: 120B total, 12B active, open weights, Intelligence Index of 36."
   ]
  },
  {
   "url": "https://whatllm.org/blog/llm-releases-march-2026",
   "title": "New LLMs March 2026: GPT-5.4 Tied for #1. Nobody Talked About It.",
   "published_at": "2026-03-24T00:00:00",
   "organization": "NVIDIA",
   "model": null,
   "release_date": "2026-03-01",
   "access_type": null,
   "license": null,
   "significance": "Open-weight voice chat model",
   "organization_evidence": [
    "New LLMs March 2026: GPT-5.4 Tied for #1. Nobody Talked About It.",
    "**Nemotron 3 VoiceChat**, NVIDIA"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "New LLMs March 2026: GPT-5.4 Tied for #1. Nobody Talked About It.",
    "Every text-focused model that shipped in March 2026, ordered chronologically."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "New LLMs March 2026: GPT-5.4 Tied for #1. Nobody Talked About It.",
    "**Nemotron 3 VoiceChat**, NVIDIA"
   ]
  },
  {
   "url": "https://whatllm.org/blog/llm-releases-march-2026",
   "title": "New LLMs March 2026: GPT-5.4 Tied for #1. Nobody Talked About It.",
   "published_at": "2026-03-24T00:00:00",
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-03-01",
   "access_type": null,
   "license": null,
   "significance": "Product line featuring reasoning variants and MoE variants ranging from 0.8B to 397B parameters",
   "organization_evidence": [
    "New LLMs March 2026: GPT-5.4 Tied for #1. Nobody Talked About It.",
    "**Qwen3.5** (8 variants), Alibaba"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "New LLMs March 2026: GPT-5.4 Tied for #1. Nobody Talked About It.",
    "Every text-focused model that shipped in March 2026, ordered chronologically."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "New LLMs March 2026: GPT-5.4 Tied for #1. Nobody Talked About It.",
    "Alibaba went wide with Qwen3.5",
    "Qwen3.5 isn't a model. It's a product line. Alibaba shipped reasoning variants at 0.8B, 2B, 4B, and 9B (dense), plus MoE variants at 27B, 35B (3B active), 122B (10B active), and 397B."
   ]
  },
  {
   "url": "https://droiddynasty.com/kimi-k3-open-weights-ship",
   "title": "Kimi K3 Open Weights Ships Early With Custom License",
   "published_at": "2026-07-28T17:03:30",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-26",
   "access_type": null,
   "license": null,
   "significance": "A 2.8 trillion-parameter mixture of experts model with a 1-million-token context window, activating 104 billion parameters per token.",
   "organization_evidence": [
    "# Kimi K3 Open Weights Ships Early With Custom License",
    "The kimi k3 open weights dropped on July 26, a day ahead of schedule, making Moonshot AI the provider of the largest publicly available open-weight model."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Kimi K3 Open Weights Ships Early With Custom License",
    "The kimi k3 open weights dropped on July 26, a day ahead of schedule, making Moonshot AI the provider of the largest publicly available open-weight model."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Kimi K3 Open Weights Ships Early With Custom License",
    "Developers can now download, modify, and self-host a 2.8 trillion-parameter mixture of experts model with a 1-million-token context window."
   ]
  },
  {
   "url": "https://medium.com/@ffguci8/meituan-open-sources-longcat-2-0-a-1-6t-parameter-model-cad8df656ba2",
   "title": "Meituan Open-Sources LongCat-2.0, a 1.6T-Parameter Model",
   "published_at": "2026-07-09T18:29:54",
   "organization": "Meituan",
   "model": null,
   "release_date": "2026-06-30",
   "access_type": null,
   "license": null,
   "significance": "a 1.6-trillion-parameter model with a native 1-million-token context window, trained entirely on Chinese-made chips",
   "organization_evidence": [
    "# Meituan Open-Sources LongCat-2.0, a 1.6T-Parameter Model",
    "On June 30, 2026, Meituan became exactly that source: [LongCat-2.0](https://github.com/meituan-longcat/LongCat-2.0), a 1.6-trillion-parameter model, free for commercial use under the MIT license, trained and served entirely on Chinese-made chips."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Meituan Open-Sources LongCat-2.0, a 1.6T-Parameter Model",
    "On June 30, 2026, Meituan became exactly that source: [LongCat-2.0](https://github.com/meituan-longcat/LongCat-2.0), a 1.6-trillion-parameter model, free for commercial use under the MIT license, trained and served entirely on Chinese-made chips."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Meituan Open-Sources LongCat-2.0, a 1.6T-Parameter Model",
    "On June 30, 2026, Meituan became exactly that source: [LongCat-2.0](https://github.com/meituan-longcat/LongCat-2.0), a 1.6-trillion-parameter model, free for commercial use under the MIT license, trained and served entirely on Chinese-made chips."
   ]
  },
  {
   "url": "https://ai-tldr.dev/models/inkling",
   "title": "Inkling \u2014 Thinking Machines' Open-Weight Multimodal MoE | AI/TLDR",
   "published_at": "2026-07-15T00:00:00",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "sparse Mixture-of-Experts transformer with 975B total parameters and 41B active per token, trained on roughly 45 trillion tokens of text, images, audio, and video",
   "organization_evidence": [
    "# Inkling",
    "Inkling is Thinking Machines Lab's first foundation model, released on July 15, 2026."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Inkling",
    "Inkling is Thinking Machines Lab's first foundation model, released on July 15, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Inkling",
    "It is a sparse Mixture-of-Experts transformer with 975B total parameters and 41B active per token, trained on roughly 45 trillion tokens of text, images, audio, and video."
   ]
  },
  {
   "url": "http://ai-tldr.dev/releases/poolside-laguna-s-2-1",
   "title": "Poolside Laguna S 2.1 \u2014 118B open-weight coding MoE with 8B active",
   "published_at": "2026-07-21T12:00:00",
   "organization": "Poolside",
   "model": "Laguna S 2.1",
   "release_date": "2026-07-21",
   "access_type": "open-weight",
   "license": "OpenMDW-1.1",
   "significance": "118B/8B-active open-weight coding MoE with a 1M-token context",
   "organization_evidence": [
    "Poolside \u00b7 2026-07-21 \u00b7 major",
    "# Poolside Laguna S 2.1 \u2014 118B open-weight coding MoE with 8B active",
    "Poolside ships Laguna S 2.1, a 118B/8B-active open-weight coding MoE with a 1M-token context."
   ],
   "model_evidence": [
    "Poolside \u00b7 2026-07-21 \u00b7 major",
    "# Poolside Laguna S 2.1 \u2014 118B open-weight coding MoE with 8B active",
    "Poolside ships Laguna S 2.1, a 118B/8B-active open-weight coding MoE with a 1M-token context."
   ],
   "release_date_evidence": [
    "Poolside \u00b7 2026-07-21 \u00b7 major",
    "# Poolside Laguna S 2.1 \u2014 118B open-weight coding MoE with 8B active",
    "Laguna S 2.1 is Poolside's new open-weight coding foundation model, released July 21, 2026 with weights on Hugging Face under the Linux Foundation's OpenMDW-1.1 license."
   ],
   "access_type_evidence": [
    "Poolside \u00b7 2026-07-21 \u00b7 major",
    "# Poolside Laguna S 2.1 \u2014 118B open-weight coding MoE with 8B active",
    "Poolside ships Laguna S 2.1, a 118B/8B-active open-weight coding MoE with a 1M-token context."
   ],
   "license_evidence": [
    "Poolside \u00b7 2026-07-21 \u00b7 major",
    "# Poolside Laguna S 2.1 \u2014 118B open-weight coding MoE with 8B active"
   ],
   "significance_evidence": [
    "Poolside \u00b7 2026-07-21 \u00b7 major",
    "# Poolside Laguna S 2.1 \u2014 118B open-weight coding MoE with 8B active",
    "Poolside ships Laguna S 2.1, a 118B/8B-active open-weight coding MoE with a 1M-token context."
   ]
  },
  {
   "url": "https://businessoutreach.in/kimi-k3-ai-model-launch-openai-anthropic-rival",
   "title": "China\u2019s AI model Kimi K3 set to rival Anthropic, OpenAI; Public launch on July 27",
   "published_at": "2026-07-18T06:05:20",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-17",
   "access_type": null,
   "license": null,
   "significance": "a 2.8-trillion-parameter open-weight model it calls the world\u2019s largest",
   "organization_evidence": [
    "Beijing-based Moonshot AI on July 17, 2026 launched Kimi K3, a 2.8-trillion-parameter open-weight model it calls the world\u2019s largest."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Beijing-based Moonshot AI on July 17, 2026 launched Kimi K3, a 2.8-trillion-parameter open-weight model it calls the world\u2019s largest."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Beijing-based Moonshot AI on July 17, 2026 launched Kimi K3, a 2.8-trillion-parameter open-weight model it calls the world\u2019s largest."
   ]
  },
  {
   "url": "https://pivotnews.ai/five/story/thinking-machines-releases-first-open-model",
   "title": "Thinking Machines Releases First Open Model",
   "published_at": "2026-07-16T14:13:14",
   "organization": "Thinking Machines",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "A multimodal Mixture-of-Experts foundation model with 975B total parameters (41B active) and controllable reasoning effort.",
   "organization_evidence": [
    "# Thinking Machines Releases First Open Model",
    "Inkling: Our Open-Weights Model - Thinking Machines Lab Inkling: Our Open-Weights Model Jul 15, 2026 Try on Tinker Model card Hugging Face Our mission is to build AI that extends human will and judgment."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Inkling: Our Open-Weights Model - Thinking Machines Lab Inkling: Our Open-Weights Model Jul 15, 2026 Try on Tinker Model card Hugging Face Our mission is to build AI that extends human will and judgment."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Thinking Machines Releases First Open Model",
    "Our first open-weights model: multimodal, Mixture-of-Experts, with controllable reasoning effort. Available to fine-tune on Tinker.",
    "Our model, called Inkling, is a Mixture-of-Experts transformer with 975B total parameters, 41B active."
   ]
  },
  {
   "url": "https://github.com/phlx0/awesome-open-weight-models",
   "title": "GitHub - phlx0/awesome-open-weight-models: The practitioner's guide to open-weight LLMs \u2014 benchmarks, licensing, hardware, deployment, fine-tuning",
   "published_at": null,
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-04-16",
   "access_type": null,
   "license": null,
   "significance": "Multimodal: text, image, and video inputs.",
   "organization_evidence": [
    "### Qwen \u2014 Alibaba",
    "#### Qwen3.6 \u2014 released April 16, 2026",
    "| Model | Params | Context | License |",
    "| Qwen3.6-35B-A3B | 35B total / 3B active | 256K (1M w/YaRN) | Apache 2.0 |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "### Qwen \u2014 Alibaba",
    "#### Qwen3.6 \u2014 released April 16, 2026",
    "| Model | Params | Context | License |",
    "| Qwen3.6-35B-A3B | 35B total / 3B active | 256K (1M w/YaRN) | Apache 2.0 |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "### Qwen \u2014 Alibaba",
    "#### Qwen3.6 \u2014 released April 16, 2026",
    "Multimodal: text, image, and video inputs. Qwen3.6-Plus is API-only with no public weights."
   ]
  },
  {
   "url": "https://github.com/phlx0/awesome-open-weight-models",
   "title": "GitHub - phlx0/awesome-open-weight-models: The practitioner's guide to open-weight LLMs \u2014 benchmarks, licensing, hardware, deployment, fine-tuning",
   "published_at": null,
   "organization": "Google DeepMind",
   "model": null,
   "release_date": "2026-04-02",
   "access_type": null,
   "license": null,
   "significance": "Near-30B quality at 8B speed and multimodal flagship",
   "organization_evidence": [
    "### Gemma \u2014 Google DeepMind",
    "#### Gemma 4 \u2014 released April 2, 2026 \u2014 `Apache 2.0`",
    "**Models**",
    "| Model | Params | Context | Notes |",
    "| gemma-4-E2B | 2.3B effective (5.1B total) | 128K | Multimodal: text, image, audio |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "### Gemma \u2014 Google DeepMind",
    "#### Gemma 4 \u2014 released April 2, 2026 \u2014 `Apache 2.0`",
    "**Models**",
    "| Model | Params | Context | Notes |",
    "| gemma-4-E2B | 2.3B effective (5.1B total) | 128K | Multimodal: text, image, audio |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "### Gemma \u2014 Google DeepMind",
    "#### Gemma 4 \u2014 released April 2, 2026 \u2014 `Apache 2.0`",
    "**Models**",
    "| Model | Params | Context | Notes |",
    "| gemma-4-26B-A4B | 26B total / 4B active (MoE) | 256K | Near-30B quality at 8B speed |",
    "| gemma-4-31B | 31B dense | 256K | Multimodal flagship |"
   ]
  },
  {
   "url": "https://github.com/phlx0/awesome-open-weight-models",
   "title": "GitHub - phlx0/awesome-open-weight-models: The practitioner's guide to open-weight LLMs \u2014 benchmarks, licensing, hardware, deployment, fine-tuning",
   "published_at": null,
   "organization": "Microsoft",
   "model": null,
   "release_date": "2026-03",
   "access_type": null,
   "license": null,
   "significance": "Vision and reasoning",
   "organization_evidence": [
    "### Phi \u2014 Microsoft",
    "**Models**",
    "| Model | Params | Context | Notes |",
    "| Phi-4-reasoning-vision-15B | 15B | 16K | Vision and reasoning (Mar 2026) |"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "### Phi \u2014 Microsoft",
    "**Models**",
    "| Model | Params | Context | Notes |",
    "| Phi-4-reasoning-vision-15B | 15B | 16K | Vision and reasoning (Mar 2026) |"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "### Phi \u2014 Microsoft",
    "**Models**",
    "| Model | Params | Context | Notes |",
    "| Phi-4-reasoning-vision-15B | 15B | 16K | Vision and reasoning (Mar 2026) |"
   ]
  },
  {
   "url": "https://admix.software/blog/thinking-machines-inkling-open-weights",
   "title": "Mira Murati's Thinking Machines Just Shipped Inkling, a 975B Open-Weights Model Under Apache 2.0",
   "published_at": "2026-07-15T00:00:00",
   "organization": "Thinking Machines",
   "model": "Inkling",
   "release_date": "2026-07-14",
   "access_type": "open-weights",
   "license": "Apache 2.0",
   "significance": "975B-parameter mixture-of-experts model handling text, images, and audio with a 1M-token context window",
   "organization_evidence": [
    "Mira Murati's Thinking Machines Just Shipped Inkling, a 975B Open-Weights Model Under Apache 2.0",
    "Thinking Machines Lab, the startup Mira Murati started after leaving OpenAI, just released its first open-weights model."
   ],
   "model_evidence": [
    "Mira Murati's Thinking Machines Just Shipped Inkling, a 975B Open-Weights Model Under Apache 2.0",
    "Inkling is a 975B-parameter mixture-of-experts model with 41B active per token, it takes text, images, and audio, and it ships under Apache 2.0 with a 1M-token context."
   ],
   "release_date_evidence": [
    "Mira Murati's Thinking Machines Just Shipped Inkling, a 975B Open-Weights Model Under Apache 2.0",
    "6 min readJuly 14, 2026"
   ],
   "access_type_evidence": [
    "Mira Murati's Thinking Machines Just Shipped Inkling, a 975B Open-Weights Model Under Apache 2.0",
    "Thinking Machines Lab, the startup Mira Murati started after leaving OpenAI, just released its first open-weights model."
   ],
   "license_evidence": [
    "Mira Murati's Thinking Machines Just Shipped Inkling, a 975B Open-Weights Model Under Apache 2.0",
    "Inkling is a 975B-parameter mixture-of-experts model with 41B active per token, it takes text, images, and audio, and it ships under Apache 2.0 with a 1M-token context."
   ],
   "significance_evidence": [
    "Mira Murati's Thinking Machines Just Shipped Inkling, a 975B Open-Weights Model Under Apache 2.0",
    "Inkling is a 975B-parameter mixture-of-experts model with 41B active per token, it takes text, images, and audio, and it ships under Apache 2.0 with a 1M-token context."
   ]
  },
  {
   "url": "https://k3-kimi.com/blog/kimi-k3-complete-guide",
   "title": "Kimi K3: Complete Guide to Moonshot's 2.8T Model (2026)",
   "published_at": "2026-07-19T00:00:00",
   "organization": "Moonshot AI",
   "model": "Kimi K3",
   "release_date": "2026-07-16",
   "access_type": "open-weight",
   "license": "Modified MIT",
   "significance": "A 2.8-trillion-parameter mixture-of-experts system with a 1-million-token context window, native image and video understanding, and benchmark scores that put it within striking distance of GPT-5.6 Sol and Claude Fable 5 at roughly half the cost per task",
   "organization_evidence": [
    "Kimi K3: Complete Guide to Moonshot's 2.8T Model (2026)",
    "**Kimi K3** is the newest flagship AI model from **Moonshot AI**, officially launched on **July 16, 2026**."
   ],
   "model_evidence": [
    "Kimi K3: Complete Guide to Moonshot's 2.8T Model (2026)",
    "**Kimi K3** is the newest flagship AI model from **Moonshot AI**, officially launched on **July 16, 2026**."
   ],
   "release_date_evidence": [
    "Kimi K3: Complete Guide to Moonshot's 2.8T Model (2026)",
    "**Kimi K3** is the newest flagship AI model from **Moonshot AI**, officially launched on **July 16, 2026**."
   ],
   "access_type_evidence": [
    "Kimi K3: Complete Guide to Moonshot's 2.8T Model (2026)",
    "**Kimi K3** is the newest flagship AI model from **Moonshot AI**, officially launched on **July 16, 2026**.",
    "It is the largest open-weight model ever released \u2014 a 2.8-trillion-parameter mixture-of-experts system with a 1-million-token context window, native image and video understanding, and benchmark scores that put it within striking distance of GPT-5.6 Sol and Claude Fable 5 at roughly half the cost per task."
   ],
   "license_evidence": [
    "Kimi K3: Complete Guide to Moonshot's 2.8T Model (2026)",
    "Key specifications at a glance"
   ],
   "significance_evidence": [
    "Kimi K3: Complete Guide to Moonshot's 2.8T Model (2026)",
    "**Kimi K3** is the newest flagship AI model from **Moonshot AI**, officially launched on **July 16, 2026**.",
    "It is the largest open-weight model ever released \u2014 a 2.8-trillion-parameter mixture-of-experts system with a 1-million-token context window, native image and video understanding, and benchmark scores that put it within striking distance of GPT-5.6 Sol and Claude Fable 5 at roughly half the cost per task."
   ]
  },
  {
   "url": "https://aichina.news/blog/telechat-1b-a-tiny-openly-licensed-chinese-language-model-for-huawei-d3zltl",
   "title": "TeleChat-1B: A Tiny, Openly Licensed Chinese Language Model for Huawei Ascend NPUs \u2014 AICHINA.news Blog",
   "published_at": "2026-07-24T10:37:39",
   "organization": "TeleAI (China Telecom)",
   "model": null,
   "release_date": "2026-07-24",
   "access_type": null,
   "license": null,
   "significance": "A tiny 1-billion-parameter Chinese foundation model purpose-built for Huawei Ascend NPUs and edge devices.",
   "organization_evidence": [
    "TeleChat-1B: A Tiny, Openly Licensed Chinese Language Model for Huawei Ascend NPUs",
    "If you\u2019ve been hunting for a small, permissively licensed Chinese-language foundation model that can actually run on Huawei Ascend hardware, the new **TeleChat-1B** from TeleAI (China Telecom) might just be the find of the week."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "TeleChat-1B: A Tiny, Openly Licensed Chinese Language Model for Huawei Ascend NPUs",
    "models/TeleAI July 24, 2026 2 min read Auto-generated from Modelers"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "TeleChat-1B: A Tiny, Openly Licensed Chinese Language Model for Huawei Ascend NPUs",
    "Released under Apache-2.0 and hosted exclusively on the Modelers.cn platform, this 1-billion-parameter model is purpose-built for developers who need to deploy lightweight language understanding on edge devices, in resource-constrained environments, or directly on Ascend NPUs without wrestling with proprietary runtimes."
   ]
  },
  {
   "url": "https://thursdai.news/topics/reasoning",
   "title": "Reasoning & Math: AI Releases & Expert Coverage \u2014 ThursdAI",
   "published_at": "2026-08-04T00:00:00",
   "organization": "Liquid AI",
   "model": null,
   "release_date": "2026-07",
   "access_type": null,
   "license": null,
   "significance": "removes the reasoning doom-loop and improves eval scores across the board",
   "organization_evidence": [
    "Liquid AI open-sources Antidoom, removing the reasoning doom-loop",
    "Reasoning models, chain-of-thought, math performance, and test-time compute. \u2014 93 releases covered on the show.",
    "Antidoom"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## [July 2026](/releases/2026-07)",
    "Reasoning models, chain-of-thought, math performance, and test-time compute. \u2014 93 releases covered on the show.",
    "Antidoom"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Liquid AI open-sources Antidoom, removing the reasoning doom-loop",
    "Reasoning models, chain-of-thought, math performance, and test-time compute. \u2014 93 releases covered on the show.",
    "Antidoom"
   ]
  },
  {
   "url": "https://thursdai.news/topics/reasoning",
   "title": "Reasoning & Math: AI Releases & Expert Coverage \u2014 ThursdAI",
   "published_at": "2026-08-04T00:00:00",
   "organization": "NVIDIA",
   "model": null,
   "release_date": "2026-06",
   "access_type": null,
   "license": null,
   "significance": "a 550B-parameter sparse MoE with 55B active parameters built for long-running agentic harnesses",
   "organization_evidence": [
    "NVIDIA releases Nemotron 3 Ultra, a 550B open-weight MoE for agents",
    "Reasoning models, chain-of-thought, math performance, and test-time compute. \u2014 93 releases covered on the show.",
    "Nemotron 3 Ultra"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## [June 2026](/releases/2026-06)",
    "Reasoning models, chain-of-thought, math performance, and test-time compute. \u2014 93 releases covered on the show.",
    "Nemotron 3 Ultra"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "NVIDIA dropped Nemotron 3 Ultra the day of the show, a 550B-parameter sparse MoE with 55B active parameters built for long-running agentic harnesses like OpenCode, Hermes, and OpenClaw.",
    "Reasoning models, chain-of-thought, math performance, and test-time compute. \u2014 93 releases covered on the show.",
    "Nemotron 3 Ultra"
   ]
  },
  {
   "url": "https://thursdai.news/topics/reasoning",
   "title": "Reasoning & Math: AI Releases & Expert Coverage \u2014 ThursdAI",
   "published_at": "2026-08-04T00:00:00",
   "organization": "Mistral",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "128B dense flagship model with 256K context and configurable reasoning",
   "organization_evidence": [
    "Mistral Medium 3.5: 128B dense flagship with 256K context",
    "Reasoning models, chain-of-thought, math performance, and test-time compute. \u2014 93 releases covered on the show.",
    "Mistral Medium 3.5"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## [April 2026](/releases/2026-04)",
    "Reasoning models, chain-of-thought, math performance, and test-time compute. \u2014 93 releases covered on the show.",
    "Mistral Medium 3.5"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Mistral launched Medium 3.5, a 128B dense flagship model with 256K context and configurable reasoning, released with weights on Hugging Face.",
    "Reasoning models, chain-of-thought, math performance, and test-time compute. \u2014 93 releases covered on the show.",
    "Mistral Medium 3.5"
   ]
  },
  {
   "url": "https://thursdai.news/topics/reasoning",
   "title": "Reasoning & Math: AI Releases & Expert Coverage \u2014 ThursdAI",
   "published_at": "2026-08-04T00:00:00",
   "organization": "NVIDIA",
   "model": null,
   "release_date": "2026-03",
   "access_type": null,
   "license": null,
   "significance": "a 120B Hybrid Mamba-Transformer MoE model with 12B active parameters and a 1M-token context window",
   "organization_evidence": [
    "NVIDIA releases Nemotron 3 Super 120B with $26B open-source bet",
    "Reasoning models, chain-of-thought, math performance, and test-time compute. \u2014 93 releases covered on the show.",
    "Nemotron 3 Super 120B"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## [March 2026](/releases/2026-03)",
    "Reasoning models, chain-of-thought, math performance, and test-time compute. \u2014 93 releases covered on the show.",
    "Nemotron 3 Super 120B"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "NVIDIA launched Nemotron 3 Super, a 120B Hybrid Mamba-Transformer MoE model with 12B active parameters, a 1M-token context window, and 450 tok/s throughput.",
    "Reasoning models, chain-of-thought, math performance, and test-time compute. \u2014 93 releases covered on the show.",
    "Nemotron 3 Super 120B"
   ]
  },
  {
   "url": "https://thursdai.news/topics/reasoning",
   "title": "Reasoning & Math: AI Releases & Expert Coverage \u2014 ThursdAI",
   "published_at": "2026-08-04T00:00:00",
   "organization": "InternLM",
   "model": null,
   "release_date": "2026-02",
   "access_type": null,
   "license": null,
   "significance": "1 trillion parameter open-source MoE model targeting SOTA scientific reasoning across chemistry, biology, materials, and earth sciences",
   "organization_evidence": [
    "Intern-S1-Pro: 1 trillion parameter open MoE for scientific reasoning",
    "Reasoning models, chain-of-thought, math performance, and test-time compute. \u2014 93 releases covered on the show.",
    "Intern-S1-Pro"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## [February 2026](/releases/2026-02)",
    "Reasoning models, chain-of-thought, math performance, and test-time compute. \u2014 93 releases covered on the show.",
    "Intern-S1-Pro"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "InternLM released Intern-S1-Pro, a 1 trillion parameter open-source MoE model targeting SOTA scientific reasoning across chemistry, biology, materials, and earth sciences.",
    "Reasoning models, chain-of-thought, math performance, and test-time compute. \u2014 93 releases covered on the show.",
    "Intern-S1-Pro"
   ]
  },
  {
   "url": "https://thursdai.news/topics/reasoning",
   "title": "Reasoning & Math: AI Releases & Expert Coverage \u2014 ThursdAI",
   "published_at": "2026-08-04T00:00:00",
   "organization": "StepFun",
   "model": null,
   "release_date": "2026-02",
   "access_type": null,
   "license": null,
   "significance": "196B sparse MoE model with only 11B active parameters, claiming frontier-level reasoning while generating at 100-350 tokens per second",
   "organization_evidence": [
    "StepFun Step 3.5 Flash: frontier reasoning claims at 11B active params",
    "Reasoning models, chain-of-thought, math performance, and test-time compute. \u2014 93 releases covered on the show.",
    "Step 3.5 Flash"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## [February 2026](/releases/2026-02)",
    "Reasoning models, chain-of-thought, math performance, and test-time compute. \u2014 93 releases covered on the show.",
    "Step 3.5 Flash"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "StepFun released Step 3.5 Flash, a 196B sparse MoE model with only 11B active parameters, claiming frontier-level reasoning while generating at 100-350 tokens per second.",
    "Reasoning models, chain-of-thought, math performance, and test-time compute. \u2014 93 releases covered on the show.",
    "Step 3.5 Flash"
   ]
  },
  {
   "url": "https://thursdai.news/topics/reasoning",
   "title": "Reasoning & Math: AI Releases & Expert Coverage \u2014 ThursdAI",
   "published_at": "2026-08-04T00:00:00",
   "organization": "Liquid AI",
   "model": null,
   "release_date": "2026-01",
   "access_type": null,
   "license": null,
   "significance": "a 1.2B parameter reasoning model that runs entirely on-device with under 900MB of memory",
   "organization_evidence": [
    "Liquid AI's LFM2.5-1.2B-Thinking: on-device reasoning under 900MB",
    "Reasoning models, chain-of-thought, math performance, and test-time compute. \u2014 93 releases covered on the show.",
    "LFM2.5-1.2B-Thinking"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## [January 2026](/releases/2026-01)",
    "Reasoning models, chain-of-thought, math performance, and test-time compute. \u2014 93 releases covered on the show.",
    "LFM2.5-1.2B-Thinking"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Liquid AI released LFM2.5-1.2B-Thinking, a 1.2B parameter reasoning model that runs entirely on-device with under 900MB of memory.",
    "Reasoning models, chain-of-thought, math performance, and test-time compute. \u2014 93 releases covered on the show.",
    "LFM2.5-1.2B-Thinking"
   ]
  },
  {
   "url": "https://thursdai.news/topics/reasoning",
   "title": "Reasoning & Math: AI Releases & Expert Coverage \u2014 ThursdAI",
   "published_at": "2026-08-04T00:00:00",
   "organization": "Meituan",
   "model": null,
   "release_date": "2026-01",
   "access_type": null,
   "license": null,
   "significance": "open-source reasoning MoE with 560B total parameters and only 27B active",
   "organization_evidence": [
    "Meituan's LongCat Flash Thinking: 560B MoE with 27B active, MIT licensed",
    "Reasoning models, chain-of-thought, math performance, and test-time compute. \u2014 93 releases covered on the show.",
    "LongCat Flash Thinking"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## [January 2026](/releases/2026-01)",
    "Reasoning models, chain-of-thought, math performance, and test-time compute. \u2014 93 releases covered on the show.",
    "LongCat Flash Thinking"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Meituan released LongCat Flash Thinking, an open-source reasoning MoE with 560B total parameters and only 27B active, under an MIT license.",
    "Reasoning models, chain-of-thought, math performance, and test-time compute. \u2014 93 releases covered on the show.",
    "LongCat Flash Thinking"
   ]
  },
  {
   "url": "https://thursdai.news/topics/reasoning",
   "title": "Reasoning & Math: AI Releases & Expert Coverage \u2014 ThursdAI",
   "published_at": "2026-08-04T00:00:00",
   "organization": "NVIDIA",
   "model": null,
   "release_date": "2026-01",
   "access_type": null,
   "license": null,
   "significance": "a family of open source reasoning-based self-driving AI models",
   "organization_evidence": [
    "NVIDIA Alpha Mayo: open source reasoning self-driving models",
    "Reasoning models, chain-of-thought, math performance, and test-time compute. \u2014 93 releases covered on the show.",
    "Alpha Mayo"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## [January 2026](/releases/2026-01)",
    "Reasoning models, chain-of-thought, math performance, and test-time compute. \u2014 93 releases covered on the show.",
    "Alpha Mayo"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "NVIDIA announced Alpha Mayo at CES, a family of open source reasoning-based self-driving AI models.",
    "Reasoning models, chain-of-thought, math performance, and test-time compute. \u2014 93 releases covered on the show.",
    "Alpha Mayo"
   ]
  },
  {
   "url": "https://llmcheck.net/blog/state-of-open-source-local-llms-july-2026",
   "title": "State of Open-Source Local LLMs \u2014 July 2026: GLM 5.2 Breaks the Frontier, Qwen 4.1, Llama 5 405B",
   "published_at": "2026-07-11T00:00:00",
   "organization": "Zhipu AI",
   "model": null,
   "release_date": "2026-07-08",
   "access_type": null,
   "license": null,
   "significance": "the FIRST open model to beat GPT-5 and Claude",
   "organization_evidence": [
    "## The 30-Day Recap (TL;DR)",
    "Every major open-weights release between June 11 and July 2026, with a one-line takeaway.",
    "On July 8, Zhipu AI released GLM 5.2: a 744-billion-parameter mixture-of-experts model with 40 billion active parameters per token, shipped under the MIT license."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## The 30-Day Recap (TL;DR)",
    "Every major open-weights release between June 11 and July 2026, with a one-line takeaway.",
    "- **GLM 5.2** Jul 8 \u2014 MIT, 744B-A40B MoE, 68.5% SWE-Bench Pro \u2014 the FIRST open model to beat GPT-5 and Claude. The headline.",
    "On July 8, Zhipu AI released GLM 5.2: a 744-billion-parameter mixture-of-experts model with 40 billion active parameters per token, shipped under the MIT license."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## The 30-Day Recap (TL;DR)",
    "Every major open-weights release between June 11 and July 2026, with a one-line takeaway.",
    "- **GLM 5.2** Jul 8 \u2014 MIT, 744B-A40B MoE, 68.5% SWE-Bench Pro \u2014 the FIRST open model to beat GPT-5 and Claude. The headline."
   ]
  },
  {
   "url": "https://llmcheck.net/blog/state-of-open-source-local-llms-july-2026",
   "title": "State of Open-Source Local LLMs \u2014 July 2026: GLM 5.2 Breaks the Frontier, Qwen 4.1, Llama 5 405B",
   "published_at": "2026-07-11T00:00:00",
   "organization": "Zhipu",
   "model": null,
   "release_date": "2026-07-08",
   "access_type": null,
   "license": null,
   "significance": "frontier-adjacent on consumer silicon",
   "organization_evidence": [
    "## The 30-Day Recap (TL;DR)",
    "Every major open-weights release between June 11 and July 2026, with a one-line takeaway.",
    "**GLM 5.2 Air brings it home.** Alongside the flagship, Zhipu shipped **GLM 5.2 Air** \u2014 a 106B-A12B MoE distilled from the full model, also MIT-licensed."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## The 30-Day Recap (TL;DR)",
    "Every major open-weights release between June 11 and July 2026, with a one-line takeaway.",
    "- **GLM 5.2 Air 106B-A12B** Jul 8 \u2014 MIT, runs on a 64 GB Mac at ~30 tok/s, 58% SWE-Bench Pro \u2014 frontier-adjacent on consumer silicon."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## The 30-Day Recap (TL;DR)",
    "Every major open-weights release between June 11 and July 2026, with a one-line takeaway.",
    "- **GLM 5.2 Air 106B-A12B** Jul 8 \u2014 MIT, runs on a 64 GB Mac at ~30 tok/s, 58% SWE-Bench Pro \u2014 frontier-adjacent on consumer silicon."
   ]
  },
  {
   "url": "https://llmcheck.net/blog/state-of-open-source-local-llms-july-2026",
   "title": "State of Open-Source Local LLMs \u2014 July 2026: GLM 5.2 Breaks the Frontier, Qwen 4.1, Llama 5 405B",
   "published_at": "2026-07-11T00:00:00",
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-07-02",
   "access_type": null,
   "license": null,
   "significance": "the new Mac-runnable #1, score 76",
   "organization_evidence": [
    "## The 30-Day Recap (TL;DR)",
    "Every major open-weights release between June 11 and July 2026, with a one-line takeaway.",
    "Six weeks after the full Qwen 4 release, Alibaba shipped Qwen 4.1 32B-A3B on July 2 \u2014 a point-release refinement rather than a new generation."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## The 30-Day Recap (TL;DR)",
    "Every major open-weights release between June 11 and July 2026, with a one-line takeaway.",
    "- **Qwen 4.1 32B-A3B** Jul 2 \u2014 Apache 2.0, 80% SWE-Verified, ~62 tok/s on M4 Pro 24 GB \u2014 the new Mac-runnable #1, score 76.",
    "Six weeks after the full Qwen 4 release, Alibaba shipped Qwen 4.1 32B-A3B on July 2 \u2014 a point-release refinement rather than a new generation."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## The 30-Day Recap (TL;DR)",
    "Every major open-weights release between June 11 and July 2026, with a one-line takeaway.",
    "- **Qwen 4.1 32B-A3B** Jul 2 \u2014 Apache 2.0, 80% SWE-Verified, ~62 tok/s on M4 Pro 24 GB \u2014 the new Mac-runnable #1, score 76."
   ]
  },
  {
   "url": "https://llmcheck.net/blog/state-of-open-source-local-llms-july-2026",
   "title": "State of Open-Source Local LLMs \u2014 July 2026: GLM 5.2 Breaks the Frontier, Qwen 4.1, Llama 5 405B",
   "published_at": "2026-07-11T00:00:00",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-07-05",
   "access_type": null,
   "license": null,
   "significance": "frontier reasoning, server-class",
   "organization_evidence": [
    "## The 30-Day Recap (TL;DR)",
    "Every major open-weights release between June 11 and July 2026, with a one-line takeaway.",
    "- **DeepSeek R3** Jul 5 \u2014 MIT, 685B-A37B, AIME 95% \u2014 frontier reasoning, server-class."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## The 30-Day Recap (TL;DR)",
    "Every major open-weights release between June 11 and July 2026, with a one-line takeaway.",
    "- **DeepSeek R3** Jul 5 \u2014 MIT, 685B-A37B, AIME 95% \u2014 frontier reasoning, server-class."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## The 30-Day Recap (TL;DR)",
    "Every major open-weights release between June 11 and July 2026, with a one-line takeaway.",
    "- **DeepSeek R3** Jul 5 \u2014 MIT, 685B-A37B, AIME 95% \u2014 frontier reasoning, server-class."
   ]
  },
  {
   "url": "https://llmcheck.net/blog/state-of-open-source-local-llms-july-2026",
   "title": "State of Open-Source Local LLMs \u2014 July 2026: GLM 5.2 Breaks the Frontier, Qwen 4.1, Llama 5 405B",
   "published_at": "2026-07-11T00:00:00",
   "organization": "Meta",
   "model": null,
   "release_date": "2026-07-06",
   "access_type": null,
   "license": null,
   "significance": "Meta's dense frontier, MMLU 91%",
   "organization_evidence": [
    "## The 30-Day Recap (TL;DR)",
    "Every major open-weights release between June 11 and July 2026, with a one-line takeaway.",
    "- **Llama 5 405B** Jul 6 \u2014 Meta's dense frontier, MMLU 91%, server / 192 GB Q2 only (~5 tok/s on M4 Ultra)."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## The 30-Day Recap (TL;DR)",
    "Every major open-weights release between June 11 and July 2026, with a one-line takeaway.",
    "- **Llama 5 405B** Jul 6 \u2014 Meta's dense frontier, MMLU 91%, server / 192 GB Q2 only (~5 tok/s on M4 Ultra)."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## The 30-Day Recap (TL;DR)",
    "Every major open-weights release between June 11 and July 2026, with a one-line takeaway.",
    "- **Llama 5 405B** Jul 6 \u2014 Meta's dense frontier, MMLU 91%, server / 192 GB Q2 only (~5 tok/s on M4 Ultra)."
   ]
  },
  {
   "url": "https://llmcheck.net/blog/state-of-open-source-local-llms-july-2026",
   "title": "State of Open-Source Local LLMs \u2014 July 2026: GLM 5.2 Breaks the Frontier, Qwen 4.1, Llama 5 405B",
   "published_at": "2026-07-11T00:00:00",
   "organization": "Google",
   "model": null,
   "release_date": "2026-07-03",
   "access_type": null,
   "license": null,
   "significance": "1M context, MMLU 86%",
   "organization_evidence": [
    "## The 30-Day Recap (TL;DR)",
    "Every major open-weights release between June 11 and July 2026, with a one-line takeaway.",
    "Google scaled its Gemma 4.5 refresh up to a 27B dense variant on July 3."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## The 30-Day Recap (TL;DR)",
    "Every major open-weights release between June 11 and July 2026, with a one-line takeaway.",
    "- **Gemma 4.5 27B** Jul 3 \u2014 Gemma, 1M context, MMLU 86%, ~42 tok/s on M5 Max, 32 GB Mac.",
    "Google scaled its Gemma 4.5 refresh up to a 27B dense variant on July 3."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## The 30-Day Recap (TL;DR)",
    "Every major open-weights release between June 11 and July 2026, with a one-line takeaway.",
    "- **Gemma 4.5 27B** Jul 3 \u2014 Gemma, 1M context, MMLU 86%, ~42 tok/s on M5 Max, 32 GB Mac."
   ]
  },
  {
   "url": "https://llmcheck.net/blog/state-of-open-source-local-llms-july-2026",
   "title": "State of Open-Source Local LLMs \u2014 July 2026: GLM 5.2 Breaks the Frontier, Qwen 4.1, Llama 5 405B",
   "published_at": "2026-07-11T00:00:00",
   "organization": "Mistral",
   "model": null,
   "release_date": "2026-07-04",
   "access_type": null,
   "license": null,
   "significance": "Apache 2.0 MoE, SWE-Verified 70%",
   "organization_evidence": [
    "## The 30-Day Recap (TL;DR)",
    "Every major open-weights release between June 11 and July 2026, with a one-line takeaway.",
    "- **Mistral Medium 4 41B-A13B** Jul 4 \u2014 Apache 2.0 MoE, SWE-Verified 70%, ~48 tok/s, 32 GB Mac."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## The 30-Day Recap (TL;DR)",
    "Every major open-weights release between June 11 and July 2026, with a one-line takeaway.",
    "- **Mistral Medium 4 41B-A13B** Jul 4 \u2014 Apache 2.0 MoE, SWE-Verified 70%, ~48 tok/s, 32 GB Mac.",
    "Mistral's Medium 4 is a 41B-A13B MoE under Apache 2.0, shipped July 4."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## The 30-Day Recap (TL;DR)",
    "Every major open-weights release between June 11 and July 2026, with a one-line takeaway.",
    "- **Mistral Medium 4 41B-A13B** Jul 4 \u2014 Apache 2.0 MoE, SWE-Verified 70%, ~48 tok/s, 32 GB Mac."
   ]
  },
  {
   "url": "https://llmcheck.net/blog/state-of-open-source-local-llms-july-2026",
   "title": "State of Open-Source Local LLMs \u2014 July 2026: GLM 5.2 Breaks the Frontier, Qwen 4.1, Llama 5 405B",
   "published_at": "2026-07-11T00:00:00",
   "organization": "Microsoft",
   "model": null,
   "release_date": "2026-06-28",
   "access_type": null,
   "license": null,
   "significance": "MMLU 88%, AIME 80%",
   "organization_evidence": [
    "## The 30-Day Recap (TL;DR)",
    "Every major open-weights release between June 11 and July 2026, with a one-line takeaway.",
    "Microsoft completed the Phi-5 family on June 28 with Phi-5 Large 28B, MIT-licensed and dense."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## The 30-Day Recap (TL;DR)",
    "Every major open-weights release between June 11 and July 2026, with a one-line takeaway.",
    "- **Phi-5 Large 28B** Jun 28 \u2014 MIT, MMLU 88%, AIME 80%, ~38 tok/s on M5 Max, 24\u201332 GB Mac.",
    "Microsoft completed the Phi-5 family on June 28 with Phi-5 Large 28B, MIT-licensed and dense."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## The 30-Day Recap (TL;DR)",
    "Every major open-weights release between June 11 and July 2026, with a one-line takeaway.",
    "- **Phi-5 Large 28B** Jun 28 \u2014 MIT, MMLU 88%, AIME 80%, ~38 tok/s on M5 Max, 24\u201332 GB Mac."
   ]
  },
  {
   "url": "https://llmcheck.net/blog/state-of-open-source-local-llms-july-2026",
   "title": "State of Open-Source Local LLMs \u2014 July 2026: GLM 5.2 Breaks the Frontier, Qwen 4.1, Llama 5 405B",
   "published_at": "2026-07-11T00:00:00",
   "organization": "Moonshot",
   "model": null,
   "release_date": "2026-07-07",
   "access_type": null,
   "license": null,
   "significance": "agentic-coding leader, server-class",
   "organization_evidence": [
    "## The 30-Day Recap (TL;DR)",
    "Every major open-weights release between June 11 and July 2026, with a one-line takeaway.",
    "Moonshot's ~1-trillion-parameter, 32B-active MoE (MIT) is the long-horizon agentic-coding leader, scoring 66% on SWE-Bench Pro and pairing it with a 2M-token context window \u2014 the largest of any model this month."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## The 30-Day Recap (TL;DR)",
    "Every major open-weights release between June 11 and July 2026, with a one-line takeaway.",
    "- **Kimi K3 ~1T-A32B** Jul 7 \u2014 MIT, agentic-coding leader, server-class."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## The 30-Day Recap (TL;DR)",
    "Every major open-weights release between June 11 and July 2026, with a one-line takeaway.",
    "- **Kimi K3 ~1T-A32B** Jul 7 \u2014 MIT, agentic-coding leader, server-class."
   ]
  },
  {
   "url": "https://llmcheck.net/blog/state-of-open-source-local-llms-july-2026",
   "title": "State of Open-Source Local LLMs \u2014 July 2026: GLM 5.2 Breaks the Frontier, Qwen 4.1, Llama 5 405B",
   "published_at": "2026-07-11T00:00:00",
   "organization": "Cohere",
   "model": null,
   "release_date": "2026-07-01",
   "access_type": null,
   "license": null,
   "significance": "enterprise RAG, ~28 tok/s on a 64 GB Mac",
   "organization_evidence": [
    "## The 30-Day Recap (TL;DR)",
    "Every major open-weights release between June 11 and July 2026, with a one-line takeaway.",
    "- **Command R+ 2 104B** Jul 1 \u2014 CC-BY-NC, enterprise RAG, ~28 tok/s on a 64 GB Mac."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## The 30-Day Recap (TL;DR)",
    "Every major open-weights release between June 11 and July 2026, with a one-line takeaway.",
    "- **Command R+ 2 104B** Jul 1 \u2014 CC-BY-NC, enterprise RAG, ~28 tok/s on a 64 GB Mac."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## The 30-Day Recap (TL;DR)",
    "Every major open-weights release between June 11 and July 2026, with a one-line takeaway.",
    "- **Command R+ 2 104B** Jul 1 \u2014 CC-BY-NC, enterprise RAG, ~28 tok/s on a 64 GB Mac."
   ]
  },
  {
   "url": "https://felloai.com/es/longcat-2-0",
   "title": "LongCat-2.0: Meituan\u2019s 1.6 Trillion-Parameter Open-Source Coding Model",
   "published_at": "2026-06-30T18:28:44",
   "organization": "Meituan",
   "model": null,
   "release_date": "2026-06-30",
   "access_type": null,
   "license": null,
   "significance": "a 1.6T-parameter Mixture-of-Experts model with a native 1-million-token context window",
   "organization_evidence": [
    "**Meituan** has open-sourced **LongCat-2.0**, a **1.6 trillion-parameter** Mixture-of-Experts model with a native **1-million-token context window**, released on **June 30, 2026** under a permissive **MIT license**.",
    "**LongCat-2.0** is a large language model from **Meituan**, the Chinese delivery and local-services giant that has been quietly building a serious AI lab."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "**Meituan** has open-sourced **LongCat-2.0**, a **1.6 trillion-parameter** Mixture-of-Experts model with a native **1-million-token context window**, released on **June 30, 2026** under a permissive **MIT license**."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "**Meituan** has open-sourced **LongCat-2.0**, a **1.6 trillion-parameter** Mixture-of-Experts model with a native **1-million-token context window**, released on **June 30, 2026** under a permissive **MIT license**.",
    "LongCat-2.0 is Meituan\u2019s open-source Mixture-of-Experts model with 1.6 trillion total parameters, roughly 48 billion active per token, and a native 1-million-token context window."
   ]
  },
  {
   "url": "https://divyam.ai/blog/open-weights-moment",
   "title": "Open Source LLMs Just Caught Up: Why Your LLM Router Needs to Switch in a Day",
   "published_at": "2026-04-14T04:24:09",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-04-07",
   "access_type": null,
   "license": null,
   "significance": "Sits at the top of SWE-Bench Pro at 58.4, ahead of GPT-5.4 and Claude Opus 4.6",
   "organization_evidence": [
    "# Open Source LLMs Just Caught Up: Why Your LLM Router Needs to Switch in a Day",
    "Z.ai, \"GLM-5.1: The Next Level of Open Source\" (April 7, 2026)"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Open Source LLMs Just Caught Up: Why Your LLM Router Needs to Switch in a Day",
    "Z.ai's GLM-5.1, released April 7, sits at the top of SWE-Bench Pro at 58.4, ahead of GPT-5.4 (57.7) and Claude Opus 4.6 (57.3).1",
    "Z.ai, \"GLM-5.1: The Next Level of Open Source\" (April 7, 2026)"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Open Source LLMs Just Caught Up: Why Your LLM Router Needs to Switch in a Day",
    "Z.ai's GLM-5.1, released April 7, sits at the top of SWE-Bench Pro at 58.4, ahead of GPT-5.4 (57.7) and Claude Opus 4.6 (57.3).1"
   ]
  },
  {
   "url": "https://divyam.ai/blog/open-weights-moment",
   "title": "Open Source LLMs Just Caught Up: Why Your LLM Router Needs to Switch in a Day",
   "published_at": "2026-04-14T04:24:09",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-01-27",
   "access_type": null,
   "license": null,
   "significance": "Scores 50.2% on Humanity's Last Exam with tools, ahead of GPT-5.2 and Claude 4.5 Opus",
   "organization_evidence": [
    "# Open Source LLMs Just Caught Up: Why Your LLM Router Needs to Switch in a Day",
    "Moonshot AI, Kimi K2.5 release (January 27, 2026)"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Open Source LLMs Just Caught Up: Why Your LLM Router Needs to Switch in a Day",
    "Moonshot AI, Kimi K2.5 release (January 27, 2026)"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Open Source LLMs Just Caught Up: Why Your LLM Router Needs to Switch in a Day",
    "Moonshot's Kimi K2.5 scores 50.2% on Humanity's Last Exam with tools, ahead of GPT-5.2 (45.5%) and Claude 4.5 Opus (43.2%).2"
   ]
  },
  {
   "url": "https://divyam.ai/blog/open-weights-moment",
   "title": "Open Source LLMs Just Caught Up: Why Your LLM Router Needs to Switch in a Day",
   "published_at": "2026-04-14T04:24:09",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026",
   "access_type": null,
   "license": null,
   "significance": "Matches GPT-4o on MMLU at 94.2% and offers low chat-mode pricing",
   "organization_evidence": [
    "# Open Source LLMs Just Caught Up: Why Your LLM Router Needs to Switch in a Day",
    "DeepSeek API pricing (V3.2, 2026)"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Open Source LLMs Just Caught Up: Why Your LLM Router Needs to Switch in a Day",
    "DeepSeek API pricing (V3.2, 2026)"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Open Source LLMs Just Caught Up: Why Your LLM Router Needs to Switch in a Day",
    "DeepSeek V3.2 matches GPT-4o on MMLU at 94.2% and offers chat-mode pricing of $0.14 per million input tokens, $0.28 per million output.3"
   ]
  },
  {
   "url": "https://divyam.ai/blog/open-weights-moment",
   "title": "Open Source LLMs Just Caught Up: Why Your LLM Router Needs to Switch in a Day",
   "published_at": "2026-04-14T04:24:09",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026",
   "access_type": null,
   "license": null,
   "significance": "Holds the highest open-weight score on SWE-Bench Verified at 80.2%",
   "organization_evidence": [
    "# Open Source LLMs Just Caught Up: Why Your LLM Router Needs to Switch in a Day",
    "BenchLM, \"Best Open Source LLM 2026\""
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Open Source LLMs Just Caught Up: Why Your LLM Router Needs to Switch in a Day",
    "BenchLM, \"Best Open Source LLM 2026\""
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Open Source LLMs Just Caught Up: Why Your LLM Router Needs to Switch in a Day",
    "MiniMax M2.5 holds the highest open-weight score on SWE-Bench Verified at 80.2%.4"
   ]
  },
  {
   "url": "https://opensourceforu.com/2026/07/poolside-ai-launches-open-weight-laguna-coding-models",
   "title": "Poolside AI Launches Open-Weight 'Laguna' Coding Models - Open Source For You",
   "published_at": "2026-07-08T06:33:40",
   "organization": "Poolside AI",
   "model": null,
   "release_date": "2026-07-02",
   "access_type": null,
   "license": null,
   "significance": "lightweight model with 33-billion parameter Mixture-of-Experts architecture tailored for agentic coding",
   "organization_evidence": [
    "# Poolside AI Launches Open-Weight \u2018Laguna\u2019 Coding Models",
    "Poolside AI, a heavily funded US-based foundation model lab, released its first public AI models, Laguna XS 2.1, the lightweight model, and Laguna M.1, the heavyweight model, tailored specifically for agentic coding, on 2 July, 2026."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Poolside AI Launches Open-Weight \u2018Laguna\u2019 Coding Models",
    "Poolside AI, a heavily funded US-based foundation model lab, released its first public AI models, Laguna XS 2.1, the lightweight model, and Laguna M.1, the heavyweight model, tailored specifically for agentic coding, on 2 July, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Poolside AI Launches Open-Weight \u2018Laguna\u2019 Coding Models",
    "Poolside AI, a heavily funded US-based foundation model lab, released its first public AI models, Laguna XS 2.1, the lightweight model, and Laguna M.1, the heavyweight model, tailored specifically for agentic coding, on 2 July, 2026.",
    "Laguna XS 2.1 is their fast, fully open-weight coding model."
   ]
  },
  {
   "url": "https://opensourceforu.com/2026/07/poolside-ai-launches-open-weight-laguna-coding-models",
   "title": "Poolside AI Launches Open-Weight 'Laguna' Coding Models - Open Source For You",
   "published_at": "2026-07-08T06:33:40",
   "organization": "Poolside AI",
   "model": null,
   "release_date": "2026-07-02",
   "access_type": null,
   "license": null,
   "significance": "flagship and highly capable model optimized for long-horizon software engineering in enterprise and government settings with 225 billion total parameters",
   "organization_evidence": [
    "# Poolside AI Launches Open-Weight \u2018Laguna\u2019 Coding Models",
    "Poolside AI, a heavily funded US-based foundation model lab, released its first public AI models, Laguna XS 2.1, the lightweight model, and Laguna M.1, the heavyweight model, tailored specifically for agentic coding, on 2 July, 2026."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Poolside AI Launches Open-Weight \u2018Laguna\u2019 Coding Models",
    "Poolside AI, a heavily funded US-based foundation model lab, released its first public AI models, Laguna XS 2.1, the lightweight model, and Laguna M.1, the heavyweight model, tailored specifically for agentic coding, on 2 July, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Poolside AI Launches Open-Weight \u2018Laguna\u2019 Coding Models",
    "Poolside AI, a heavily funded US-based foundation model lab, released its first public AI models, Laguna XS 2.1, the lightweight model, and Laguna M.1, the heavyweight model, tailored specifically for agentic coding, on 2 July, 2026.",
    "Laguna M.1 is their flagship, and highly capable model optimized for long-horizon software engineering in enterprise and government settings."
   ]
  },
  {
   "url": "https://techjacksolutions.com/ai-news/technology?tag=local-ai-models",
   "title": "LLM News Today - Model Releases, Benchmarks & AI Research Updates",
   "published_at": "2026-02-28T23:10:41",
   "organization": "Google DeepMind",
   "model": null,
   "release_date": "2026-06-03",
   "access_type": null,
   "license": null,
   "significance": "Encoder-free multimodal model capable of natively ingesting audio and video, running locally on consumer hardware.",
   "organization_evidence": [
    "Tech Intelligence",
    "## Daily Briefings",
    "Google DeepMind released Gemma 4 12B Unified on June 3, 2026."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Tech Intelligence",
    "## Daily Briefings",
    "Google DeepMind released Gemma 4 12B Unified on June 3, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Tech Intelligence",
    "## Daily Briefings",
    "Google\u2019s developer documentation, confirmed via cross-reference, describes Gemma 4 12B as [\u201cthe first medium-sized, encoder-free multimodal model capable of natively ingesting audio and video.\u201d](https://developers.googleblog.com/gemma-4-12b-the-developer-guide/)"
   ]
  },
  {
   "url": "https://binaryverseai.com/kimi-k3-benchmarks-pricing-fable-5-comparison",
   "title": "Kimi K3: Benchmarks, Pricing, And How It Really Compares To Fable 5",
   "published_at": "2026-07-17T09:07:23",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-07-27",
   "access_type": null,
   "license": null,
   "significance": "A 2.8 trillion parameter open model taking on frontier models with native vision understanding and a 1M context window",
   "organization_evidence": [
    "# Kimi K3: The 2.8 Trillion Parameter Open Model Taking on Fable 5",
    "Moonshot AI just did something no Chinese lab had pulled off before."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "# Kimi K3: The 2.8 Trillion Parameter Open Model Taking on Fable 5",
    "Moonshot says those land by July 27, 2026, alongside a full technical report.",
    "*Answer:* Yes, Moonshot AI has confirmed that Kimi K3 will be an open-weight model. While the API and web applications are live now, the full 2.8 trillion parameter model weights will be officially released on July 27, 2026."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "# Kimi K3: The 2.8 Trillion Parameter Open Model Taking on Fable 5",
    "Moonshot AI just did something no Chinese lab had pulled off before.",
    "Kimi K3 [landed on July 16, 2026](https://x.com/Kimi_Moonshot/status/2077830229968683203), a 2.8 trillion parameter model that Moonshot calls the world\u2019s first open 3T class system, and the numbers behind that claim are hard to wave away."
   ]
  },
  {
   "url": "https://thewaffleaffair.com/international-relations/china-s-ai-revolution-four-frontier-class-open-models-unveiled-in-only-eight-wee",
   "title": "China's AI Revolution: Four Frontier-Class Open Models Unveiled In Only Eight Weeks - The Waffle Affair",
   "published_at": "2026-07-14T14:27:24",
   "organization": "DeepSeek",
   "model": null,
   "release_date": "2026-04-24",
   "access_type": null,
   "license": null,
   "significance": "1.6-trillion-parameter mixture-of-experts model activating 49 billion parameters per pass, with a one-million-token context window",
   "organization_evidence": [
    "**Four Chinese AI laboratories** released advanced open-weight models in roughly **eight weeks between April 24 and mid-June 2026**, according to a July 13 review by Thorsten Meyer AI.",
    "The releases from **DeepSeek, MiniMax, Moonshot AI and Z.ai** matter because their downloadable weights, long context windows and lower hosted prices could make high-capability AI more accessible outside closed Western platforms.",
    "Four Frontier-Class Open Models in Eight Weeks"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The sequence began with **DeepSeek V4 Pro and Flash on April 24**, followed by **MiniMax M3 on June 1**.",
    "Four Frontier-Class Open Models in Eight Weeks"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "DeepSeek V4 is described as a **1.6-trillion-parameter mixture-of-experts model** activating 49 billion parameters per pass, with a **one-million-token context window**.",
    "Four Frontier-Class Open Models in Eight Weeks"
   ]
  },
  {
   "url": "https://thewaffleaffair.com/international-relations/china-s-ai-revolution-four-frontier-class-open-models-unveiled-in-only-eight-wee",
   "title": "China's AI Revolution: Four Frontier-Class Open Models Unveiled In Only Eight Weeks - The Waffle Affair",
   "published_at": "2026-07-14T14:27:24",
   "organization": "MiniMax",
   "model": null,
   "release_date": "2026-06-01",
   "access_type": null,
   "license": null,
   "significance": "combines a similarly long context window with native multimodal functions",
   "organization_evidence": [
    "**Four Chinese AI laboratories** released advanced open-weight models in roughly **eight weeks between April 24 and mid-June 2026**, according to a July 13 review by Thorsten Meyer AI.",
    "The releases from **DeepSeek, MiniMax, Moonshot AI and Z.ai** matter because their downloadable weights, long context windows and lower hosted prices could make high-capability AI more accessible outside closed Western platforms.",
    "Four Frontier-Class Open Models in Eight Weeks"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "The sequence began with **DeepSeek V4 Pro and Flash on April 24**, followed by **MiniMax M3 on June 1**.",
    "Four Frontier-Class Open Models in Eight Weeks"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "MiniMax M3 combines a similarly long context window with native multimodal functions.",
    "Four Frontier-Class Open Models in Eight Weeks"
   ]
  },
  {
   "url": "https://thewaffleaffair.com/international-relations/china-s-ai-revolution-four-frontier-class-open-models-unveiled-in-only-eight-wee",
   "title": "China's AI Revolution: Four Frontier-Class Open Models Unveiled In Only Eight Weeks - The Waffle Affair",
   "published_at": "2026-07-14T14:27:24",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "agent-based coding work and uses about 30% fewer reasoning tokens than K2.6",
   "organization_evidence": [
    "**Four Chinese AI laboratories** released advanced open-weight models in roughly **eight weeks between April 24 and mid-June 2026**, according to a July 13 review by Thorsten Meyer AI.",
    "The releases from **DeepSeek, MiniMax, Moonshot AI and Z.ai** matter because their downloadable weights, long context windows and lower hosted prices could make high-capability AI more accessible outside closed Western platforms.",
    "Four Frontier-Class Open Models in Eight Weeks"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Moonshot AI released **Kimi K2.7-Code around June 13**, while Z.ai introduced GLM-5.2 during the same mid-June period.",
    "Four Frontier-Class Open Models in Eight Weeks"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Moonshot positions Kimi K2.7-Code for agent-based coding work and says it uses **about 30% fewer reasoning tokens** than K2.6.",
    "Four Frontier-Class Open Models in Eight Weeks"
   ]
  },
  {
   "url": "https://thewaffleaffair.com/international-relations/china-s-ai-revolution-four-frontier-class-open-models-unveiled-in-only-eight-wee",
   "title": "China's AI Revolution: Four Frontier-Class Open Models Unveiled In Only Eight Weeks - The Waffle Affair",
   "published_at": "2026-07-14T14:27:24",
   "organization": "Z.ai",
   "model": null,
   "release_date": "2026-06-13",
   "access_type": null,
   "license": null,
   "significance": "753-billion-parameter mixture-of-experts model",
   "organization_evidence": [
    "**Four Chinese AI laboratories** released advanced open-weight models in roughly **eight weeks between April 24 and mid-June 2026**, according to a July 13 review by Thorsten Meyer AI.",
    "The releases from **DeepSeek, MiniMax, Moonshot AI and Z.ai** matter because their downloadable weights, long context windows and lower hosted prices could make high-capability AI more accessible outside closed Western platforms.",
    "Four Frontier-Class Open Models in Eight Weeks"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Moonshot AI released **Kimi K2.7-Code around June 13**, while Z.ai introduced GLM-5.2 during the same mid-June period.",
    "Four Frontier-Class Open Models in Eight Weeks"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Z.ai\u2019s GLM-5.2 is reported as a **753-billion-parameter mixture-of-experts model**.",
    "Four Frontier-Class Open Models in Eight Weeks"
   ]
  },
  {
   "url": "https://aidigitalcrew.com/models",
   "title": "The AI Model Timeline",
   "published_at": "2026-07-24T00:00:00",
   "organization": "Ant Group",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Efficiency-focused mixture-of-experts",
   "organization_evidence": [
    "## 2026",
    "\u00b7Ant GroupNew",
    "### [Ling-3.0-flash](https://github.com/inclusionAI)"
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## 2026",
    "\u00b7Ant GroupNew",
    "### [Ling-3.0-flash](https://github.com/inclusionAI)",
    "Efficiency-focused MoE, closing out the busiest release week on record."
   ]
  },
  {
   "url": "https://aidigitalcrew.com/models",
   "title": "The AI Model Timeline",
   "published_at": "2026-07-24T00:00:00",
   "organization": "poolside",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Open-weight coding model",
   "organization_evidence": [
    "## 2026",
    "\u00b7poolsideNew",
    "### [Laguna S 2.1](https://poolside.ai/blog)"
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## 2026",
    "\u00b7poolsideNew",
    "### [Laguna S 2.1](https://poolside.ai/blog)",
    "Open-weight coding model from a lab that had been entirely closed until now."
   ]
  },
  {
   "url": "https://aidigitalcrew.com/models",
   "title": "The AI Model Timeline",
   "published_at": "2026-07-24T00:00:00",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Near-trillion-scale weights released under Apache 2.0",
   "organization_evidence": [
    "## 2026",
    "\u00b7Thinking Machines LabLandmarkNew",
    "### [Inkling](https://thinkingmachines.ai/)"
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## 2026",
    "\u00b7Thinking Machines LabLandmarkNew",
    "### [Inkling](https://thinkingmachines.ai/)",
    "Thinking Machines' debut \u2014 near-trillion-scale weights released under Apache 2.0."
   ]
  },
  {
   "url": "https://aidigitalcrew.com/models",
   "title": "The AI Model Timeline",
   "published_at": "2026-07-24T00:00:00",
   "organization": "Tencent",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Model family moved to a fully permissive licence",
   "organization_evidence": [
    "## 2026",
    "\u00b7TencentNew",
    "### [Hy3](https://hunyuan.tencent.com/)"
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## 2026",
    "\u00b7TencentNew",
    "### [Hy3](https://hunyuan.tencent.com/)",
    "Tencent's Hunyuan line moved to a fully permissive licence."
   ]
  },
  {
   "url": "https://aidigitalcrew.com/models",
   "title": "The AI Model Timeline",
   "published_at": "2026-07-24T00:00:00",
   "organization": "Meituan",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "A food-delivery company shipping one of the largest models of the year",
   "organization_evidence": [
    "## 2026",
    "\u00b7MeituanNew",
    "### [LongCat-2.0](https://github.com/meituan-longcat)"
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## 2026",
    "\u00b7MeituanNew",
    "### [LongCat-2.0](https://github.com/meituan-longcat)",
    "A food-delivery company shipping one of the largest models of the year."
   ]
  },
  {
   "url": "https://aidigitalcrew.com/models",
   "title": "The AI Model Timeline",
   "published_at": "2026-07-24T00:00:00",
   "organization": "Z.ai",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Open-weight cadence through 2026",
   "organization_evidence": [
    "## 2026",
    "\u00b7Z.ai",
    "### [GLM-5.2](https://z.ai)"
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## 2026",
    "\u00b7Z.ai",
    "### [GLM-5.2](https://z.ai)",
    "Z.ai held to a roughly two-month open-weight cadence through 2026."
   ]
  },
  {
   "url": "https://aidigitalcrew.com/models",
   "title": "The AI Model Timeline",
   "published_at": "2026-07-24T00:00:00",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "A code-specialised trillion-parameter model with open weights",
   "organization_evidence": [
    "## 2026",
    "\u00b7Moonshot AI",
    "### [Kimi K2.7 Code](https://moonshotai.github.io/)"
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## 2026",
    "\u00b7Moonshot AI",
    "### [Kimi K2.7 Code](https://moonshotai.github.io/)",
    "A code-specialised trillion-parameter model with open weights."
   ]
  },
  {
   "url": "https://aidigitalcrew.com/models",
   "title": "The AI Model Timeline",
   "published_at": "2026-07-24T00:00:00",
   "organization": "NVIDIA",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "NVIDIA's largest open release, tuned for its own inference stack",
   "organization_evidence": [
    "## 2026",
    "\u00b7NVIDIA",
    "### [Nemotron 3 Ultra](https://blogs.nvidia.com/blog/category/generative-ai/)"
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## 2026",
    "\u00b7NVIDIA",
    "### [Nemotron 3 Ultra](https://blogs.nvidia.com/blog/category/generative-ai/)",
    "NVIDIA's largest open release, tuned for its own inference stack."
   ]
  },
  {
   "url": "https://aidigitalcrew.com/models",
   "title": "The AI Model Timeline",
   "published_at": "2026-07-24T00:00:00",
   "organization": "MiniMax",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Agentic MoE aimed squarely at long-horizon tool use",
   "organization_evidence": [
    "## 2026",
    "\u00b7MiniMax",
    "### [MiniMax-M3](https://www.minimax.io/news)"
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## 2026",
    "\u00b7MiniMax",
    "### [MiniMax-M3](https://www.minimax.io/news)",
    "Agentic MoE aimed squarely at long-horizon tool use."
   ]
  },
  {
   "url": "https://aidigitalcrew.com/models",
   "title": "The AI Model Timeline",
   "published_at": "2026-07-24T00:00:00",
   "organization": "StepFun",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Multimodal refresh of the sparse Flash line",
   "organization_evidence": [
    "## 2026",
    "\u00b7StepFun",
    "### [Step 3.7 Flash](https://www.stepfun.com/)"
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## 2026",
    "\u00b7StepFun",
    "### [Step 3.7 Flash](https://www.stepfun.com/)",
    "Multimodal refresh of the sparse Flash line."
   ]
  },
  {
   "url": "https://aidigitalcrew.com/models",
   "title": "The AI Model Timeline",
   "published_at": "2026-07-24T00:00:00",
   "organization": "Mistral AI",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Mistral's dense flagship, still open",
   "organization_evidence": [
    "## 2026",
    "\u00b7Mistral AI",
    "### [Mistral Medium 3.5](https://mistral.ai/news)"
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## 2026",
    "\u00b7Mistral AI",
    "### [Mistral Medium 3.5](https://mistral.ai/news)",
    "Mistral's dense flagship, still open at a time when most flagships were not."
   ]
  },
  {
   "url": "https://aidigitalcrew.com/models",
   "title": "The AI Model Timeline",
   "published_at": "2026-07-24T00:00:00",
   "organization": "DeepSeek",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Split into a cheap Flash tier and a 1.6T Pro tier",
   "organization_evidence": [
    "## 2026",
    "\u00b7DeepSeekLandmark",
    "### [DeepSeek-V4 (Flash + Pro)](https://api-docs.deepseek.com/news/)"
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## 2026",
    "\u00b7DeepSeekLandmark",
    "### [DeepSeek-V4 (Flash + Pro)](https://api-docs.deepseek.com/news/)",
    "Split into a cheap Flash tier and a 1.6T Pro tier, both under MIT."
   ]
  },
  {
   "url": "https://aidigitalcrew.com/models",
   "title": "The AI Model Timeline",
   "published_at": "2026-07-24T00:00:00",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Kept trillion-parameter open weights on a roughly quarterly cadence",
   "organization_evidence": [
    "## 2026",
    "\u00b7Moonshot AI",
    "### [Kimi K2.6](https://moonshotai.github.io/)"
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## 2026",
    "\u00b7Moonshot AI",
    "### [Kimi K2.6](https://moonshotai.github.io/)",
    "Kept trillion-parameter open weights on a roughly quarterly cadence."
   ]
  },
  {
   "url": "https://aidigitalcrew.com/models",
   "title": "The AI Model Timeline",
   "published_at": "2026-07-24T00:00:00",
   "organization": "Alibaba",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "3B active parameters at near-frontier quality",
   "organization_evidence": [
    "## 2026",
    "\u00b7Alibaba",
    "### [Qwen3.6](https://qwen.ai/blog)"
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## 2026",
    "\u00b7Alibaba",
    "### [Qwen3.6](https://qwen.ai/blog)",
    "3B active parameters at near-frontier quality \u2014 the efficiency frontier of its moment."
   ]
  },
  {
   "url": "https://aidigitalcrew.com/models",
   "title": "The AI Model Timeline",
   "published_at": "2026-07-24T00:00:00",
   "organization": "Z.ai",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Refinement of GLM-5 focused on agentic tool use",
   "organization_evidence": [
    "## 2026",
    "\u00b7Z.ai",
    "### [GLM-5.1](https://z.ai)"
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## 2026",
    "\u00b7Z.ai",
    "### [GLM-5.1](https://z.ai)",
    "Refinement of GLM-5 focused on agentic tool use."
   ]
  },
  {
   "url": "https://aidigitalcrew.com/models",
   "title": "The AI Model Timeline",
   "published_at": "2026-07-24T00:00:00",
   "organization": "Google DeepMind",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Gemma moved to a true Apache licence",
   "organization_evidence": [
    "## 2026",
    "\u00b7Google DeepMind",
    "### [Gemma 4](https://deepmind.google/models/)"
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## 2026",
    "\u00b7Google DeepMind",
    "### [Gemma 4](https://deepmind.google/models/)",
    "Gemma moved to a true Apache licence, dropping the custom terms."
   ]
  },
  {
   "url": "https://aidigitalcrew.com/models",
   "title": "The AI Model Timeline",
   "published_at": "2026-07-24T00:00:00",
   "organization": "Mistral AI",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "6B active parameters",
   "organization_evidence": [
    "## 2026",
    "\u00b7Mistral AI",
    "### [Mistral Small 4](https://mistral.ai/news)"
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## 2026",
    "\u00b7Mistral AI",
    "### [Mistral Small 4](https://mistral.ai/news)",
    "6B active parameters \u2014 \"small\" now means active cost, not total size."
   ]
  },
  {
   "url": "https://aidigitalcrew.com/models",
   "title": "The AI Model Timeline",
   "published_at": "2026-07-24T00:00:00",
   "organization": "Sarvam AI",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "India's first independently trained large model, built for Indic languages",
   "organization_evidence": [
    "## 2026",
    "\u00b7Sarvam AI",
    "### [Sarvam-105B](https://www.sarvam.ai/blogs)"
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## 2026",
    "\u00b7Sarvam AI",
    "### [Sarvam-105B](https://www.sarvam.ai/blogs)",
    "India's first independently trained large model, built for Indic languages."
   ]
  },
  {
   "url": "https://aidigitalcrew.com/models",
   "title": "The AI Model Timeline",
   "published_at": "2026-07-24T00:00:00",
   "organization": "Alibaba",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Apache-licensed frontier-class weights, still free to use commercially",
   "organization_evidence": [
    "## 2026",
    "\u00b7Alibaba",
    "### [Qwen3.5](https://qwen.ai/blog)"
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## 2026",
    "\u00b7Alibaba",
    "### [Qwen3.5](https://qwen.ai/blog)",
    "Apache-licensed frontier-class weights, still free to use commercially."
   ]
  },
  {
   "url": "https://aidigitalcrew.com/models",
   "title": "The AI Model Timeline",
   "published_at": "2026-07-24T00:00:00",
   "organization": "Z.ai",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "MIT-licensed weights at a scale that had been strictly proprietary a year earlier",
   "organization_evidence": [
    "## 2026",
    "\u00b7Z.ai",
    "### [GLM-5](https://z.ai)"
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## 2026",
    "\u00b7Z.ai",
    "### [GLM-5](https://z.ai)",
    "MIT-licensed weights at a scale that had been strictly proprietary a year earlier."
   ]
  },
  {
   "url": "https://aidigitalcrew.com/models",
   "title": "The AI Model Timeline",
   "published_at": "2026-07-24T00:00:00",
   "organization": "StepFun",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "Extreme sparsity aimed at cheap multimodal serving",
   "organization_evidence": [
    "## 2026",
    "\u00b7StepFun",
    "### [Step 3.5 Flash](https://www.stepfun.com/)"
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## 2026",
    "\u00b7StepFun",
    "### [Step 3.5 Flash](https://www.stepfun.com/)",
    "Extreme sparsity \u2014 11B active out of 196B \u2014 aimed at cheap multimodal serving."
   ]
  },
  {
   "url": "https://aidigitalcrew.com/models",
   "title": "The AI Model Timeline",
   "published_at": "2026-07-24T00:00:00",
   "organization": "Moonshot AI",
   "model": null,
   "release_date": null,
   "access_type": null,
   "license": null,
   "significance": "A trillion-parameter open MoE",
   "organization_evidence": [
    "## 2026",
    "\u00b7Moonshot AILandmark",
    "### [Kimi K2.5](https://moonshotai.github.io/)"
   ],
   "model_evidence": null,
   "release_date_evidence": null,
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## 2026",
    "\u00b7Moonshot AILandmark",
    "### [Kimi K2.5](https://moonshotai.github.io/)",
    "A trillion-parameter open MoE \u2014 the largest open weights anyone had published."
   ]
  },
  {
   "url": "https://valarian.com/article/introducing-the-open-weight-league",
   "title": "Introducing the Open-Weight League",
   "published_at": null,
   "organization": "Moonshot",
   "model": null,
   "release_date": "2026-07-27",
   "access_type": null,
   "license": null,
   "significance": "League-best intelligence and coding",
   "organization_evidence": [
    "Introducing the Open-Weight League",
    "Why does Kimi K3 have a \u2020 next to its name?",
    "Moonshot announced it on 16 July with weights publicly committed for 27 July; Artificial Analysis has already benchmarked it at League-best intelligence and coding."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Introducing the Open-Weight League",
    "Why does Kimi K3 have a \u2020 next to its name?",
    "Moonshot announced it on 16 July with weights publicly committed for 27 July; Artificial Analysis has already benchmarked it at League-best intelligence and coding."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Introducing the Open-Weight League",
    "Why does Kimi K3 have a \u2020 next to its name?",
    "Moonshot announced it on 16 July with weights publicly committed for 27 July; Artificial Analysis has already benchmarked it at League-best intelligence and coding."
   ]
  },
  {
   "url": "https://radarai.top/en/topics/china-ai-model-release-tracker",
   "title": "China AI model release tracker (how to keep the watchlist current)",
   "published_at": "2026-05-13T00:00:00",
   "organization": "Alibaba",
   "model": null,
   "release_date": "2026-04",
   "access_type": null,
   "license": null,
   "significance": "Established the current capability ceiling for open-weight Chinese models with the 30B-A3B MoE matching GPT-4o on MMLU",
   "organization_evidence": [
    "## Decision in 20 seconds",
    "**The best China AI model release tracker in 2026 maintains a stable watchlist for 6 core labs \u2014 Qwen (Alibaba), DeepSeek, Kimi (Moonshot), GLM (Zhipu), MiniMax, and Step (StepFun) \u2014 and uses release cadence, benchmark changes, and license shifts as the three actionable signals. Qwen3 (April 2026) established the current capability ceiling for open-weight Chinese models with the 30B-A3B MoE matching GPT-4o on MMLU under Apache 2.0. DeepSeek R2 is expected H2 2026 based on lab hiring signals. A tracker that covers only headline launches misses the smaller iterative updates (quantized variants, API pricing changes) that often have more immediate builder impact.**"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "## Decision in 20 seconds",
    "**The best China AI model release tracker in 2026 maintains a stable watchlist for 6 core labs \u2014 Qwen (Alibaba), DeepSeek, Kimi (Moonshot), GLM (Zhipu), MiniMax, and Step (StepFun) \u2014 and uses release cadence, benchmark changes, and license shifts as the three actionable signals. Qwen3 (April 2026) established the current capability ceiling for open-weight Chinese models with the 30B-A3B MoE matching GPT-4o on MMLU under Apache 2.0. DeepSeek R2 is expected H2 2026 based on lab hiring signals. A tracker that covers only headline launches misses the smaller iterative updates (quantized variants, API pricing changes) that often have more immediate builder impact.**"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "## Decision in 20 seconds",
    "**The best China AI model release tracker in 2026 maintains a stable watchlist for 6 core labs \u2014 Qwen (Alibaba), DeepSeek, Kimi (Moonshot), GLM (Zhipu), MiniMax, and Step (StepFun) \u2014 and uses release cadence, benchmark changes, and license shifts as the three actionable signals. Qwen3 (April 2026) established the current capability ceiling for open-weight Chinese models with the 30B-A3B MoE matching GPT-4o on MMLU under Apache 2.0. DeepSeek R2 is expected H2 2026 based on lab hiring signals. A tracker that covers only headline launches misses the smaller iterative updates (quantized variants, API pricing changes) that often have more immediate builder impact.**"
   ]
  },
  {
   "url": "https://precisionaiacademy.com/news/thinking-machines-inkling-model",
   "title": "Thinking Machines Ships Inkling: A 975B Open-Weights Model Under Apache 2.0",
   "published_at": "2026-07-15T00:00:00",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "975B total parameters with 41B active per token, a 66-layer mixture-of-experts transformer, pretraining on 45 trillion tokens, and a context window of up to 1M tokens.",
   "organization_evidence": [
    "Thinking Machines ships Inkling, a 975B open-weights model",
    "Thinking Machines Lab released Inkling on July 15, 2026 \u2014 its first in-house foundation model, and its first open-weights release."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Thinking Machines ships Inkling, a 975B open-weights model",
    "Thinking Machines Lab released Inkling on July 15, 2026 \u2014 its first in-house foundation model, and its first open-weights release."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Thinking Machines ships Inkling, a 975B open-weights model",
    "The model card lists 975B total parameters with 41B active per token, a 66-layer mixture-of-experts transformer, pretraining on 45 trillion tokens, and a context window of up to 1M tokens."
   ]
  },
  {
   "url": "https://precisionaiacademy.com/news/thinking-machines-inkling-model",
   "title": "Thinking Machines Ships Inkling: A 975B Open-Weights Model Under Apache 2.0",
   "published_at": "2026-07-15T00:00:00",
   "organization": "Thinking Machines Lab",
   "model": null,
   "release_date": "2026-07-30",
   "access_type": null,
   "license": null,
   "significance": "276B total parameters and 12B active across a 42-layer mixture-of-experts transformer, the same 1M-token context.",
   "organization_evidence": [
    "Update: Inkling-Small, two weeks later"
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Update: Inkling-Small, two weeks later"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Update: Inkling-Small, two weeks later",
    "It lists 276B total parameters and 12B active across a 42-layer mixture-of-experts transformer, the same 1M-token context, the same text/image/audio input, and the same Apache 2.0 licence."
   ]
  },
  {
   "url": "https://futureagi.com/blog/best-llms-may-2026",
   "title": "GPT-5.5, Claude, Gemini, DeepSeek V4",
   "published_at": "2026-05-06T00:00:00",
   "organization": "DeepSeek",
   "model": "DeepSeek V4-Pro",
   "release_date": "2026-04-23",
   "access_type": "open-weight",
   "license": "MIT license",
   "significance": "1.6 trillion total parameters with 49 billion active per token, MoE, hybrid attention architecture, and 80.6% SWE-bench Verified",
   "organization_evidence": [
    "# Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "DeepSeek V4-Pro and V4-Flash. Best cost-performance on the frontier",
    "DeepSeek. Released April 23\u201324, 2026. MIT license, downloadable weights."
   ],
   "model_evidence": [
    "# Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "DeepSeek V4-Pro and V4-Flash. Best cost-performance on the frontier",
    "DeepSeek. Released April 23\u201324, 2026. MIT license, downloadable weights."
   ],
   "release_date_evidence": [
    "# Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "DeepSeek V4-Pro and V4-Flash. Best cost-performance on the frontier",
    "DeepSeek. Released April 23\u201324, 2026. MIT license, downloadable weights."
   ],
   "access_type_evidence": [
    "# Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "DeepSeek V4-Pro and V4-Flash. Best cost-performance on the frontier",
    "DeepSeek. Released April 23\u201324, 2026. MIT license, downloadable weights."
   ],
   "license_evidence": [
    "# Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "DeepSeek V4-Pro and V4-Flash. Best cost-performance on the frontier",
    "DeepSeek. Released April 23\u201324, 2026. MIT license, downloadable weights."
   ],
   "significance_evidence": [
    "# Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "DeepSeek V4-Pro and V4-Flash. Best cost-performance on the frontier",
    "**V4-Pro architecture:** 1.6 trillion total parameters with 49 billion active per token, pre-trained on more than 32 trillion tokens (per DeepSeek model card)."
   ]
  },
  {
   "url": "https://futureagi.com/blog/best-llms-may-2026",
   "title": "GPT-5.5, Claude, Gemini, DeepSeek V4",
   "published_at": "2026-05-06T00:00:00",
   "organization": "DeepSeek",
   "model": "DeepSeek V4-Flash",
   "release_date": "2026-04-24",
   "access_type": "open-weight",
   "license": "MIT license",
   "significance": "284 billion total / 13 billion active, same hybrid attention, 1M-token context, latency-optimized",
   "organization_evidence": [
    "# Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "DeepSeek V4-Pro and V4-Flash. Best cost-performance on the frontier",
    "DeepSeek. Released April 23\u201324, 2026. MIT license, downloadable weights."
   ],
   "model_evidence": [
    "# Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "DeepSeek V4-Pro and V4-Flash. Best cost-performance on the frontier",
    "DeepSeek. Released April 23\u201324, 2026. MIT license, downloadable weights."
   ],
   "release_date_evidence": [
    "# Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "DeepSeek V4-Pro and V4-Flash. Best cost-performance on the frontier",
    "DeepSeek. Released April 23\u201324, 2026. MIT license, downloadable weights."
   ],
   "access_type_evidence": [
    "# Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "DeepSeek V4-Pro and V4-Flash. Best cost-performance on the frontier",
    "DeepSeek. Released April 23\u201324, 2026. MIT license, downloadable weights."
   ],
   "license_evidence": [
    "# Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "DeepSeek V4-Pro and V4-Flash. Best cost-performance on the frontier",
    "DeepSeek. Released April 23\u201324, 2026. MIT license, downloadable weights."
   ],
   "significance_evidence": [
    "# Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "DeepSeek V4-Pro and V4-Flash. Best cost-performance on the frontier",
    "**V4-Flash architecture:** 284 billion total / 13 billion active. Same hybrid attention. 1M-token context. Latency-optimized."
   ]
  },
  {
   "url": "https://futureagi.com/blog/best-llms-may-2026",
   "title": "GPT-5.5, Claude, Gemini, DeepSeek V4",
   "published_at": "2026-05-06T00:00:00",
   "organization": "Alibaba",
   "model": "Qwen 3.6-72B-dense",
   "release_date": "2026-04-11",
   "access_type": "open-weight",
   "license": "Apache 2.0",
   "significance": "flagship dense, 256K context, 94.8% HumanEval, 68.2% SWE-bench Verified, 71.4% LiveCodeBench",
   "organization_evidence": [
    "# Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "Top open-weight and Chinese-frontier LLMs in May 2026",
    "Qwen 3.6 open-weight family. Best Apache 2.0 multilingual base",
    "Alibaba. Released April 11\u201321, 2026. Apache 2.0, three sizes (72B / 35B-A3B / 27B)."
   ],
   "model_evidence": [
    "# Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "Top open-weight and Chinese-frontier LLMs in May 2026",
    "Qwen 3.6 open-weight family. Best Apache 2.0 multilingual base",
    "- **Qwen 3.6-72B-dense** (April 11). flagship dense, 256K context, 94.8% HumanEval, 68.2% SWE-bench Verified, 71.4% LiveCodeBench."
   ],
   "release_date_evidence": [
    "# Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "Top open-weight and Chinese-frontier LLMs in May 2026",
    "Qwen 3.6 open-weight family. Best Apache 2.0 multilingual base",
    "- **Qwen 3.6-72B-dense** (April 11). flagship dense, 256K context, 94.8% HumanEval, 68.2% SWE-bench Verified, 71.4% LiveCodeBench."
   ],
   "access_type_evidence": [
    "# Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "Top open-weight and Chinese-frontier LLMs in May 2026",
    "Qwen 3.6 open-weight family. Best Apache 2.0 multilingual base",
    "Alibaba. Released April 11\u201321, 2026. Apache 2.0, three sizes (72B / 35B-A3B / 27B)."
   ],
   "license_evidence": [
    "# Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "Top open-weight and Chinese-frontier LLMs in May 2026",
    "Qwen 3.6 open-weight family. Best Apache 2.0 multilingual base",
    "Alibaba. Released April 11\u201321, 2026. Apache 2.0, three sizes (72B / 35B-A3B / 27B)."
   ],
   "significance_evidence": [
    "# Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "Top open-weight and Chinese-frontier LLMs in May 2026",
    "Qwen 3.6 open-weight family. Best Apache 2.0 multilingual base",
    "- **Qwen 3.6-72B-dense** (April 11). flagship dense, 256K context, 94.8% HumanEval, 68.2% SWE-bench Verified, 71.4% LiveCodeBench."
   ]
  },
  {
   "url": "https://futureagi.com/blog/best-llms-may-2026",
   "title": "GPT-5.5, Claude, Gemini, DeepSeek V4",
   "published_at": "2026-05-06T00:00:00",
   "organization": "Alibaba",
   "model": "Qwen 3.6-35B-A3B",
   "release_date": "2026-04-16",
   "access_type": "open-weight",
   "license": "Apache 2.0",
   "significance": "MoE with 35B total / 3B active, optimized for inference cost",
   "organization_evidence": [
    "# Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "Top open-weight and Chinese-frontier LLMs in May 2026",
    "Qwen 3.6 open-weight family. Best Apache 2.0 multilingual base",
    "Alibaba. Released April 11\u201321, 2026. Apache 2.0, three sizes (72B / 35B-A3B / 27B)."
   ],
   "model_evidence": [
    "# Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "Top open-weight and Chinese-frontier LLMs in May 2026",
    "Qwen 3.6 open-weight family. Best Apache 2.0 multilingual base",
    "- **Qwen 3.6-35B-A3B** (April 16). MoE with 35B total / 3B active, optimized for inference cost"
   ],
   "release_date_evidence": [
    "# Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "Top open-weight and Chinese-frontier LLMs in May 2026",
    "Qwen 3.6 open-weight family. Best Apache 2.0 multilingual base",
    "- **Qwen 3.6-35B-A3B** (April 16). MoE with 35B total / 3B active, optimized for inference cost"
   ],
   "access_type_evidence": [
    "# Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "Top open-weight and Chinese-frontier LLMs in May 2026",
    "Qwen 3.6 open-weight family. Best Apache 2.0 multilingual base",
    "Alibaba. Released April 11\u201321, 2026. Apache 2.0, three sizes (72B / 35B-A3B / 27B)."
   ],
   "license_evidence": [
    "# Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "Top open-weight and Chinese-frontier LLMs in May 2026",
    "Qwen 3.6 open-weight family. Best Apache 2.0 multilingual base",
    "Alibaba. Released April 11\u201321, 2026. Apache 2.0, three sizes (72B / 35B-A3B / 27B)."
   ],
   "significance_evidence": [
    "# Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "Top open-weight and Chinese-frontier LLMs in May 2026",
    "Qwen 3.6 open-weight family. Best Apache 2.0 multilingual base",
    "- **Qwen 3.6-35B-A3B** (April 16). MoE with 35B total / 3B active, optimized for inference cost"
   ]
  },
  {
   "url": "https://futureagi.com/blog/best-llms-may-2026",
   "title": "GPT-5.5, Claude, Gemini, DeepSeek V4",
   "published_at": "2026-05-06T00:00:00",
   "organization": "Alibaba",
   "model": "Qwen 3.6-27B",
   "release_date": "2026-04-21",
   "access_type": "open-weight",
   "license": "Apache 2.0",
   "significance": "dense, image+text, 27B at fine-tunable size",
   "organization_evidence": [
    "# Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "Top open-weight and Chinese-frontier LLMs in May 2026",
    "Qwen 3.6 open-weight family. Best Apache 2.0 multilingual base",
    "Alibaba. Released April 11\u201321, 2026. Apache 2.0, three sizes (72B / 35B-A3B / 27B)."
   ],
   "model_evidence": [
    "# Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "Top open-weight and Chinese-frontier LLMs in May 2026",
    "Qwen 3.6 open-weight family. Best Apache 2.0 multilingual base",
    "- **Qwen 3.6-27B** (April 21). dense, image+text, 27B at fine-tunable size"
   ],
   "release_date_evidence": [
    "# Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "Top open-weight and Chinese-frontier LLMs in May 2026",
    "Qwen 3.6 open-weight family. Best Apache 2.0 multilingual base",
    "- **Qwen 3.6-27B** (April 21). dense, image+text, 27B at fine-tunable size"
   ],
   "access_type_evidence": [
    "# Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "Top open-weight and Chinese-frontier LLMs in May 2026",
    "Qwen 3.6 open-weight family. Best Apache 2.0 multilingual base",
    "Alibaba. Released April 11\u201321, 2026. Apache 2.0, three sizes (72B / 35B-A3B / 27B)."
   ],
   "license_evidence": [
    "# Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "Top open-weight and Chinese-frontier LLMs in May 2026",
    "Qwen 3.6 open-weight family. Best Apache 2.0 multilingual base",
    "Alibaba. Released April 11\u201321, 2026. Apache 2.0, three sizes (72B / 35B-A3B / 27B)."
   ],
   "significance_evidence": [
    "# Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "Top open-weight and Chinese-frontier LLMs in May 2026",
    "Qwen 3.6 open-weight family. Best Apache 2.0 multilingual base",
    "- **Qwen 3.6-27B** (April 21). dense, image+text, 27B at fine-tunable size"
   ]
  },
  {
   "url": "https://futureagi.com/blog/best-llms-may-2026",
   "title": "GPT-5.5, Claude, Gemini, DeepSeek V4",
   "published_at": "2026-05-06T00:00:00",
   "organization": "Mistral AI",
   "model": "Mistral Small 4",
   "release_date": "2026-03-16",
   "access_type": "open-weight",
   "license": "Apache 2.0",
   "significance": "119B total parameters with 6.5B active per forward pass, hybrid reasoning, image and text inputs",
   "organization_evidence": [
    "# Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "Top open-weight and Chinese-frontier LLMs in May 2026",
    "Mistral Small 4. Smallest frontier-class MoE for self-hosting",
    "Mistral AI. Released March 16, 2026. Apache 2.0, 119B total / 6.5B active."
   ],
   "model_evidence": [
    "# Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "Top open-weight and Chinese-frontier LLMs in May 2026",
    "Mistral Small 4. Smallest frontier-class MoE for self-hosting",
    "Mistral AI. Released March 16, 2026. Apache 2.0, 119B total / 6.5B active."
   ],
   "release_date_evidence": [
    "# Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "Top open-weight and Chinese-frontier LLMs in May 2026",
    "Mistral Small 4. Smallest frontier-class MoE for self-hosting",
    "Mistral AI. Released March 16, 2026. Apache 2.0, 119B total / 6.5B active."
   ],
   "access_type_evidence": [
    "# Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "Top open-weight and Chinese-frontier LLMs in May 2026",
    "Mistral Small 4. Smallest frontier-class MoE for self-hosting",
    "Mistral AI. Released March 16, 2026. Apache 2.0, 119B total / 6.5B active."
   ],
   "license_evidence": [
    "# Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "Top open-weight and Chinese-frontier LLMs in May 2026",
    "Mistral Small 4. Smallest frontier-class MoE for self-hosting",
    "Mistral AI. Released March 16, 2026. Apache 2.0, 119B total / 6.5B active."
   ],
   "significance_evidence": [
    "# Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "Top open-weight and Chinese-frontier LLMs in May 2026",
    "Mistral Small 4. Smallest frontier-class MoE for self-hosting",
    "The smallest-footprint frontier pick. 119B total parameters with 6.5B active per forward pass."
   ]
  },
  {
   "url": "https://futureagi.com/blog/best-llms-may-2026",
   "title": "GPT-5.5, Claude, Gemini, DeepSeek V4",
   "published_at": "2026-05-06T00:00:00",
   "organization": "Google",
   "model": "Gemma 4",
   "release_date": "2026-04-02",
   "access_type": "open-weight",
   "license": "Apache 2.0",
   "significance": "Four sizes ranging up to 31B parameters, open-weight family well-supported in the Google ecosystem",
   "organization_evidence": [
    "# Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "Top open-weight and Chinese-frontier LLMs in May 2026",
    "Gemma 4. Best small fine-tuning base in the Google ecosystem",
    "Google. Released April 2, 2026. Apache 2.0, sizes up to 31B."
   ],
   "model_evidence": [
    "# Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "Top open-weight and Chinese-frontier LLMs in May 2026",
    "Gemma 4. Best small fine-tuning base in the Google ecosystem",
    "Google. Released April 2, 2026. Apache 2.0, sizes up to 31B."
   ],
   "release_date_evidence": [
    "# Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "Top open-weight and Chinese-frontier LLMs in May 2026",
    "Gemma 4. Best small fine-tuning base in the Google ecosystem",
    "Google. Released April 2, 2026. Apache 2.0, sizes up to 31B."
   ],
   "access_type_evidence": [
    "# Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "Top open-weight and Chinese-frontier LLMs in May 2026",
    "Gemma 4. Best small fine-tuning base in the Google ecosystem",
    "Google. Released April 2, 2026. Apache 2.0, sizes up to 31B."
   ],
   "license_evidence": [
    "# Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "Top open-weight and Chinese-frontier LLMs in May 2026",
    "Gemma 4. Best small fine-tuning base in the Google ecosystem",
    "Google. Released April 2, 2026. Apache 2.0, sizes up to 31B."
   ],
   "significance_evidence": [
    "# Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "Top open-weight and Chinese-frontier LLMs in May 2026",
    "Gemma 4. Best small fine-tuning base in the Google ecosystem",
    "Google\u2019s open-weight family. Four sizes ranging up to 31B parameters."
   ]
  },
  {
   "url": "https://futureagi.com/blog/best-llms-may-2026",
   "title": "GPT-5.5, Claude, Gemini, DeepSeek V4",
   "published_at": "2026-05-06T00:00:00",
   "organization": "Z.AI / Zhipu",
   "model": "GLM-5.1",
   "release_date": "2026-04-07",
   "access_type": "open-weight",
   "license": "MIT license",
   "significance": "754B parameter MoE, largest practically self-hostable MIT-licensed flagship",
   "organization_evidence": [
    "# Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "Top open-weight and Chinese-frontier LLMs in May 2026",
    "GLM-5.1. Largest open-weight MIT-licensed flagship",
    "Z.AI / Zhipu. Released April 7, 2026. MIT license, 754B parameter MoE."
   ],
   "model_evidence": [
    "# Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "Top open-weight and Chinese-frontier LLMs in May 2026",
    "GLM-5.1. Largest open-weight MIT-licensed flagship",
    "Z.AI / Zhipu. Released April 7, 2026. MIT license, 754B parameter MoE."
   ],
   "release_date_evidence": [
    "# Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "Top open-weight and Chinese-frontier LLMs in May 2026",
    "GLM-5.1. Largest open-weight MIT-licensed flagship",
    "Z.AI / Zhipu. Released April 7, 2026. MIT license, 754B parameter MoE."
   ],
   "access_type_evidence": [
    "# Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "Top open-weight and Chinese-frontier LLMs in May 2026",
    "GLM-5.1. Largest open-weight MIT-licensed flagship",
    "Z.AI / Zhipu. Released April 7, 2026. MIT license, 754B parameter MoE."
   ],
   "license_evidence": [
    "# Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "Top open-weight and Chinese-frontier LLMs in May 2026",
    "GLM-5.1. Largest open-weight MIT-licensed flagship",
    "Z.AI / Zhipu. Released April 7, 2026. MIT license, 754B parameter MoE."
   ],
   "significance_evidence": [
    "# Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "Top open-weight and Chinese-frontier LLMs in May 2026",
    "GLM-5.1. Largest open-weight MIT-licensed flagship",
    "**754B parameter MoE under MIT license.** GLM-5.1 is the largest practically self-hostable MIT-licensed flagship in May 2026"
   ]
  },
  {
   "url": "https://futureagi.com/blog/best-llms-may-2026",
   "title": "GPT-5.5, Claude, Gemini, DeepSeek V4",
   "published_at": "2026-05-06T00:00:00",
   "organization": "Moonshot AI",
   "model": "Kimi K2.6",
   "release_date": "2026-04-20",
   "access_type": "open-weight",
   "license": null,
   "significance": "#1 open-weight model on the Artificial Analysis Intelligence Index with 1.1 trillion parameters MoE",
   "organization_evidence": [
    "# Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "Top open-weight and Chinese-frontier LLMs in May 2026",
    "Kimi K2.6. Top open-weight on AA Intelligence Index",
    "Moonshot AI. Released April 20, 2026. Open-weight, 1.1T MoE."
   ],
   "model_evidence": [
    "# Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "Top open-weight and Chinese-frontier LLMs in May 2026",
    "Kimi K2.6. Top open-weight on AA Intelligence Index",
    "Moonshot AI. Released April 20, 2026. Open-weight, 1.1T MoE."
   ],
   "release_date_evidence": [
    "# Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "Top open-weight and Chinese-frontier LLMs in May 2026",
    "Kimi K2.6. Top open-weight on AA Intelligence Index",
    "Moonshot AI. Released April 20, 2026. Open-weight, 1.1T MoE."
   ],
   "access_type_evidence": [
    "# Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "Top open-weight and Chinese-frontier LLMs in May 2026",
    "Kimi K2.6. Top open-weight on AA Intelligence Index",
    "Moonshot AI. Released April 20, 2026. Open-weight, 1.1T MoE."
   ],
   "license_evidence": null,
   "significance_evidence": [
    "# Best LLMs of May 2026: Top Closed-Source, Open-Weight, Multimodal, and Coding Picks",
    "Top open-weight and Chinese-frontier LLMs in May 2026",
    "Kimi K2.6. Top open-weight on AA Intelligence Index",
    "**The current #1 open-weight model on the Artificial Analysis Intelligence Index.** Moonshot\u2019s latest, an upgrade to K2.5 with the primary improvement being **agentic stability over extended sessions**."
   ]
  },
  {
   "url": "http://ai-tldr.dev/releases/upstage-solar-open2",
   "title": "Solar Open 2 \u2014 Upstage's 250B/15B open-weight MoE built for agentic use",
   "published_at": "2026-07-22T12:00:00",
   "organization": "Upstage",
   "model": null,
   "release_date": "2026-07-22",
   "access_type": null,
   "license": null,
   "significance": "a 250B-parameter Mixture-of-Experts model that activates 15B per token, offers a 1M-token context, and targets agentic tool-calling",
   "organization_evidence": [
    "Upstage \u00b7 2026-07-22 \u00b7 major",
    "Upstage releases Solar Open 2, a 250B-parameter Mixture-of-Experts model that activates 15B per token, offers a 1M-token context, and targets agentic tool-calling."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Upstage \u00b7 2026-07-22 \u00b7 major",
    "Weights, a technical report, and a hosted playground all shipped together on July 22, 2026, alongside a permissive-with-attribution Upstage Solar License that allows commercial use."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Upstage releases Solar Open 2, a 250B-parameter Mixture-of-Experts model that activates 15B per token, offers a 1M-token context, and targets agentic tool-calling."
   ]
  },
  {
   "url": "https://nationpress.com/sciencetech/meituan-trains-16t-param-ai-on-china-chips",
   "title": "Meituan's LongCat-2.0: China's largest AI model trained on domestic chips | Nation Press",
   "published_at": "2026-06-17T06:30:32",
   "organization": "Meituan",
   "model": null,
   "release_date": "2026-06-30",
   "access_type": null,
   "license": null,
   "significance": "1.6-trillion-parameter large language model with a 1 million-token context window trained entirely on domestic semiconductor hardware",
   "organization_evidence": [
    "# Meituan's LongCat-2.0: China's largest AI model trained on domestic chips",
    "Meituan has open-sourced LongCat-2.0, a 1.6-trillion-parameter LLM it claims is China's first frontier-scale model trained \u2014 not just run \u2014 on domestic AI chips, challenging the assumption that home-grown silicon can only handle inference workloads."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Meituan open-sourced LongCat-2.0 on 30 June 2026 , a large language model with 1.6 trillion parameters and a 1 million-token context window.",
    "# Meituan's LongCat-2.0: China's largest AI model trained on domestic chips"
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Meituan open-sourced LongCat-2.0 on 30 June 2026 , a large language model with 1.6 trillion parameters and a 1 million-token context window.",
    "# Meituan's LongCat-2.0: China's largest AI model trained on domestic chips"
   ]
  },
  {
   "url": "https://knowledgenuts.com/2026/07/17/what-is-inkling-thinking-machines-labs-open-weight-ai-model-explained",
   "title": "What Is Inkling? Thinking Machines Lab\u2019s Open-Weight AI Model Explained",
   "published_at": "2026-07-17T11:39:08",
   "organization": "Thinking Machines",
   "model": null,
   "release_date": "2026-07-15",
   "access_type": null,
   "license": null,
   "significance": "A 975B total parameter (41B active) mixture-of-experts transformer supporting a one-million-token context window and multimodal inputs.",
   "organization_evidence": [
    "Thinking Machines Lab released Inkling on July 15, 2026, giving developers its first general-purpose AI model and making the full weights available for download."
   ],
   "model_evidence": null,
   "release_date_evidence": [
    "Thinking Machines Lab released Inkling on July 15, 2026, giving developers its first general-purpose AI model and making the full weights available for download."
   ],
   "access_type_evidence": null,
   "license_evidence": null,
   "significance_evidence": [
    "Thinking Machines Lab released Inkling on July 15, 2026, giving developers its first general-purpose AI model and making the full weights available for download.",
    "Inkling is a 66-layer mixture-of-experts transformer with 975 billion total parameters and 41 billion active parameters.",
    "The model supports a context window of up to one million tokens and was pretrained on 45 trillion tokens drawn from text, images, audio and video."
   ]
  }
 ],
 "notes": [
  "Merged 8 queries (1566 total) \u2192 1477 unique results",
  "Search result set: r5247d91c07d (1477 rows), reference it as FROM r5247d91c07d in follow-up queries"
 ]
}