{
 "id": "r8f4c29f3a52",
 "found": true,
 "parent_id": "r13f88a3ffda",
 "created_at": "2026-08-31 22:00:49.452815+00:00",
 "rows": [
  {
   "model": "Llama 4 Scout",
   "organization": "Meta",
   "release_date": "2026",
   "parameters": "109B MoE / 17B active (16 experts)",
   "context_window": "10M",
   "license": "Meta custom (700M MAU clause)",
   "source_count": 7,
   "significance": "Industry-leading context window. The license is not OSI-approved and carries a monthly-active-user ceiling.",
   "example_url": "https://fazm.ai/t/latest-open-source-llm-releases-2026"
  },
  {
   "model": "Qwen3-Coder-Next",
   "organization": "Alibaba",
   "release_date": "2026-02",
   "parameters": "80B / 3B",
   "context_window": "128K",
   "license": "Apache 2.0",
   "source_count": 2,
   "significance": "Outperforms much larger models like DeepSeek V3.2 (37B active) on coding tasks at a fraction of the inference cost",
   "example_url": "https://klymentiev.com/blog/best-open-source-llm-2026"
  },
  {
   "model": "Qwen 3.5 397B",
   "organization": "Alibaba Cloud",
   "release_date": "2026-02",
   "parameters": "397B (17B active)",
   "context_window": "256K",
   "license": "Apache 2.0",
   "source_count": 2,
   "significance": "Flagship open-weight",
   "example_url": "https://techunfoldedai.com/qwen-ai"
  },
  {
   "model": "Qwen 3.5 122B-A10B",
   "organization": "Alibaba",
   "release_date": "2026-02-01",
   "parameters": "122B / 10B",
   "context_window": "256K",
   "license": "Apache 2.0",
   "source_count": 3,
   "significance": "Medium model series variant combining high-throughput inference with efficient sparse Mixture-of-Experts architecture.",
   "example_url": "https://tech-insider.org/llama-4-vs-qwen-vs-mistral-2026"
  },
  {
   "model": "Qwen 3.5 27B",
   "organization": "Alibaba",
   "release_date": "2026-02-01",
   "parameters": "27B / 27B",
   "context_window": "256K",
   "license": "Apache 2.0",
   "source_count": 3,
   "significance": "Efficient and high-performing dense model delivering strong results in coding, reasoning, and output consistency.",
   "example_url": "https://tech-insider.org/llama-4-vs-qwen-vs-mistral-2026"
  },
  {
   "model": "MiniMax M2.5",
   "organization": "MiniMax",
   "release_date": "2026-02-12",
   "parameters": null,
   "context_window": "204K",
   "license": "Open",
   "source_count": 2,
   "significance": "Agent-native coding flagship released weeks after MiniMax's Hong Kong IPO, with top-tier SWE-bench results at a fraction of frontier cost.",
   "example_url": "https://outyet.ai/providers/minimax"
  },
  {
   "model": "GLM-5",
   "organization": "Zhipu AI",
   "release_date": "2026-02-13",
   "parameters": "744B (44B active)",
   "context_window": "1 million-token",
   "license": "MIT",
   "source_count": 10,
   "significance": "rank #1 among open-source models on Vending Bench 2; achieves best-in-class performance among all open-source models in the world on reasoning, coding, and agentic tasks",
   "example_url": "https://z.ai/blog/glm-5"
  },
  {
   "model": "Qwen 3.5",
   "organization": "Alibaba",
   "release_date": "2026-02-16",
   "parameters": "17B active / 397B total",
   "context_window": "256K",
   "license": "Apache 2.0",
   "source_count": 4,
   "significance": "Qwen3.5 (397B total / 17B active MoE); 256K native context; 256 experts",
   "example_url": "https://jaredwatkins.com/research/local-llm-finetuning/fundamentals/base-models"
  },
  {
   "model": "Mistral Large 3",
   "organization": "Mistral",
   "release_date": "2026-03-06",
   "parameters": "41B active parameters and 675B total parameters",
   "context_window": "256K tokens",
   "license": "Apache 2.0",
   "source_count": 2,
   "significance": "Currently ranks as the #2 open-source non-reasoning model on LMArena",
   "example_url": "https://insights.marvin-42.com/articles/mistral-launches-mistral-3-open-multimodal-family-under-apache-20"
  },
  {
   "model": "Nemotron 3 Super",
   "organization": "NVIDIA",
   "release_date": "2026-03-11",
   "parameters": "120.6B",
   "context_window": "Up to 1M",
   "license": "Open weights",
   "source_count": 4,
   "significance": "36.0 on the Artificial Analysis Intelligence Index",
   "example_url": "https://tech-insider.org/ca/nvidia-nemotron-3-ultra-2026"
  },
  {
   "model": "Mistral Small 4",
   "organization": "Mistral AI",
   "release_date": "2026-03-16",
   "parameters": "119B / 6B",
   "context_window": "256,000",
   "license": "Apache 2.0",
   "source_count": 11,
   "significance": "unifies reasoning, multimodal understanding, and agentic coding in one model",
   "example_url": "https://hidekazu-konishi.com/entry/open_weights_llm_release_history_and_timeline.html"
  },
  {
   "model": "Mistral Small (latest)",
   "organization": "Mistral",
   "release_date": "2026-03-16",
   "parameters": "119B",
   "context_window": "256,000",
   "license": "Open",
   "source_count": 2,
   "significance": null,
   "example_url": "https://models.opencode.ai/labs/mistral"
  },
  {
   "model": "Qwen3.5-Omni",
   "organization": "Alibaba",
   "release_date": "2026-03-30",
   "parameters": null,
   "context_window": "256K",
   "license": null,
   "source_count": 2,
   "significance": "A Thinker-Talker architecture capable of real-time speech, multi-hour audio comprehension, and video understanding, featuring Alibaba's ARIA system and an open-weight Light variant on Hugging Face.",
   "example_url": "https://ai2.work/blog/alibaba-s-three-model-blitz-fuels-the-open-weight-ai-arms-race"
  },
  {
   "model": "GLM-5.1",
   "organization": "Z.ai",
   "release_date": "2026-04",
   "parameters": "754B",
   "context_window": "200K",
   "license": "MIT",
   "source_count": 13,
   "significance": "91.7% MMLU, 85.7% GPQA, 58.4% SWE-Bench Pro",
   "example_url": "https://qtithow.com/2026/07/comparing-120-open-source-llms-best.html"
  },
  {
   "model": "Llama 4 Maverick",
   "organization": "Meta",
   "release_date": "2026-04",
   "parameters": "400B",
   "context_window": "1M",
   "license": "Llama Community",
   "source_count": 6,
   "significance": "matches GPT-4o and Gemini 2.0 Flash at an LMArena ELO score of 1,417",
   "example_url": "https://temperaturezero.com/2026/05/18/meta-open-sourced-llama-4-kept-muse-spark"
  },
  {
   "model": "Gemma 4",
   "organization": "Google",
   "release_date": "2026-04-02",
   "parameters": "edge E2B/E4B sizes plus larger dense and Mixture-of-Experts models",
   "context_window": "256K",
   "license": "Apache 2.0",
   "source_count": 22,
   "significance": "focus on agentic workflows",
   "example_url": "https://hidekazu-konishi.com/entry/open_weights_llm_release_history_and_timeline.html"
  },
  {
   "model": "Gemma 4 E4B",
   "organization": "Google",
   "release_date": "2026-04-02",
   "parameters": "8B (Effective: ~4.5B)",
   "context_window": "128K",
   "license": "Apache 2.0",
   "source_count": 7,
   "significance": "Spans from a 2-billion-parameter edge model that runs on a phone to a 31-billion-parameter dense model that ranks #3 among all open models on Arena AI.",
   "example_url": "https://decodethefuture.org/en/google-gemma-4-explained"
  },
  {
   "model": "Gemma 4 26B Mixture of Experts",
   "organization": "Google DeepMind",
   "release_date": "2026-04-02",
   "parameters": "26B (3.8B active)",
   "context_window": "256K",
   "license": "Apache 2.0",
   "source_count": 6,
   "significance": "Spans from a 2-billion-parameter edge model that runs on a phone to a 31-billion-parameter dense model that ranks #3 among all open models on Arena AI.",
   "example_url": "https://decodethefuture.org/en/google-gemma-4-explained"
  },
  {
   "model": "Gemma 4 31B Dense",
   "organization": "Google DeepMind",
   "release_date": "2026-04-02",
   "parameters": "31B",
   "context_window": "256K",
   "license": "Apache 2.0",
   "source_count": 5,
   "significance": "Spans from a 2-billion-parameter edge model that runs on a phone to a 31-billion-parameter dense model that ranks #3 among all open models on Arena AI.",
   "example_url": "https://decodethefuture.org/en/google-gemma-4-explained"
  },
  {
   "model": "Gemma 4 E2B",
   "organization": "Google",
   "release_date": "2026-04-02",
   "parameters": "5.1B (Effective: ~2.3B)",
   "context_window": "128K",
   "license": "Apache 2.0",
   "source_count": 5,
   "significance": "Spans from a 2-billion-parameter edge model that runs on a phone to a 31-billion-parameter dense model that ranks #3 among all open models on Arena AI.",
   "example_url": "https://decodethefuture.org/en/google-gemma-4-explained"
  },
  {
   "model": "Gemma 4 31B",
   "organization": "Google",
   "release_date": "2026-04-02",
   "parameters": "30.7B",
   "context_window": "256K",
   "license": "Apache 2.0",
   "source_count": 2,
   "significance": "LMArena score of 1,452 \u2014 ranked #3 among open models globally and #1 US open model. Achieves 89.2% on AIME 2026 and a Codeforces ELO of 2,150 (professional competitive programming level).",
   "example_url": "https://aiautomationglobal.com/blog/google-gemma-4-open-source-apache-agentic-ai-2026"
  },
  {
   "model": "Gemma 4 26B A4B",
   "organization": "Google",
   "release_date": "2026-04-02",
   "parameters": "25.2B",
   "context_window": "256K",
   "license": "Apache 2.0",
   "source_count": 2,
   "significance": "Ranks #6 among all open-weight models globally. Google calls it \u201cthe world\u2019s most parameter-efficient model at this scale.\u201d",
   "example_url": "https://aiautomationglobal.com/blog/google-gemma-4-open-source-apache-agentic-ai-2026"
  },
  {
   "model": "MiniMax M2.7",
   "organization": "MiniMax",
   "release_date": "2026-04-12",
   "parameters": "229B total parameters, 10B active per token",
   "context_window": "200K tokens",
   "license": "Modified MIT",
   "source_count": 6,
   "significance": "Hits 56.22% on SWE-Bench Pro and 57.0% on Terminal Bench 2, with downloads already in the tens of thousands.",
   "example_url": "https://ai-tldr.dev/releases/minimax-m2-7-open-weights"
  },
  {
   "model": "Qwen 3.6-35B-A3B",
   "organization": "Alibaba",
   "release_date": "2026-04-16",
   "parameters": "35B-A3B",
   "context_window": "256K",
   "license": "Apache 2.0",
   "source_count": 11,
   "significance": "Explicitly tuned for agentic coding, activating only 3B parameters per token.",
   "example_url": "https://tech-insider.org/llama-4-vs-qwen-vs-mistral-2026"
  },
  {
   "model": "Qwen 3.6",
   "organization": "Alibaba",
   "release_date": "2026-04-16",
   "parameters": "35B-A3B and 27B",
   "context_window": "1M",
   "license": "Apache 2.0",
   "source_count": 5,
   "significance": "building on Qwen3.5 while prioritizing stability, real-world coding utility, and direct community feedback",
   "example_url": "https://hidekazu-konishi.com/entry/open_weights_llm_release_history_and_timeline.html"
  },
  {
   "model": "Kimi K2.6",
   "organization": "Moonshot AI",
   "release_date": "2026-04-20",
   "parameters": "1T total / 32B active (MoE)",
   "context_window": "256K",
   "license": "Modified MIT",
   "source_count": 7,
   "significance": "Native multimodal agentic model. MIT-equivalent below 100M MAU / $20M monthly revenue, attribution clause above.",
   "example_url": "https://fazm.ai/t/latest-open-source-llm-releases-2026"
  },
  {
   "model": "Qwen 3.6 27B",
   "organization": "Alibaba",
   "release_date": "2026-04-22",
   "parameters": "27B",
   "context_window": "256K",
   "license": "Apache 2.0",
   "source_count": 8,
   "significance": "The local dev sweet spot",
   "example_url": "https://digitalapplied.com/blog/open-weight-models-h1-2026-retrospective-deepseek-qwen-llama"
  },
  {
   "model": "DeepSeek V4 Flash",
   "organization": "DeepSeek",
   "release_date": "2026-04-24",
   "parameters": "284B total / 13B active",
   "context_window": "1,000,000 tokens",
   "license": "MIT",
   "source_count": 24,
   "significance": "Terminal Bench 2.1 score of 82.7, a massive +25.8 point leap from its initial April preview score of 56.9",
   "example_url": "https://cline.bot/blog/best-open-weight-models-that-matter-in-2026"
  },
  {
   "model": "DeepSeek V4 Pro",
   "organization": "DeepSeek",
   "release_date": "2026-04-24",
   "parameters": "1.6 trillion total parameters",
   "context_window": "1,000,000 tokens",
   "license": "MIT",
   "source_count": 22,
   "significance": "Reported at 80.6% on SWE-bench Verified, putting it in the same range as leading closed models",
   "example_url": "https://cline.bot/blog/best-open-weight-models-that-matter-in-2026"
  },
  {
   "model": "DeepSeek V4",
   "organization": "DeepSeek",
   "release_date": "2026-04-24",
   "parameters": "V4-Pro 1.6T/49B active; V4-Flash 284B/13B active",
   "context_window": "1M",
   "license": "MIT",
   "source_count": 8,
   "significance": "Two MoE sizes, native 1M context, trained on 32T+ tokens. The clean-license frontier MoE of the spring.",
   "example_url": "https://fazm.ai/t/latest-open-source-llm-releases-2026"
  },
  {
   "model": "DeepSeek V4 Preview",
   "organization": "DeepSeek",
   "release_date": "2026-04-24",
   "parameters": "DeepSeek-V4-Pro (1.6T total / 49B active parameters) and DeepSeek-V4-Flash (284B total / 13B active)",
   "context_window": "1M-token",
   "license": "open weights",
   "source_count": 2,
   "significance": "built on token-wise compression and DeepSeek Sparse Attention, and making a 1M-token context window the standard across DeepSeek's official services",
   "example_url": "https://hidekazu-konishi.com/entry/open_weights_llm_release_history_and_timeline.html"
  },
  {
   "model": "Mistral Medium 3.5",
   "organization": "Mistral",
   "release_date": "2026-04-29",
   "parameters": "128B",
   "context_window": "256K",
   "license": "modified MIT license",
   "source_count": 10,
   "significance": "merges instruction-following, reasoning, and coding in a single set of weights",
   "example_url": "https://hidekazu-konishi.com/entry/open_weights_llm_release_history_and_timeline.html"
  },
  {
   "model": "Mistral Medium (latest)",
   "organization": "Mistral",
   "release_date": "2026-04-29",
   "parameters": null,
   "context_window": "262,144",
   "license": "Open",
   "source_count": 2,
   "significance": null,
   "example_url": "https://models.opencode.ai/labs/mistral"
  },
  {
   "model": "MiniMax M3",
   "organization": "MiniMax",
   "release_date": "2026-06-01",
   "parameters": "Undisclosed (MoE)",
   "context_window": "1M",
   "license": "MiniMax Community License",
   "source_count": 16,
   "significance": "First 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.",
   "example_url": "https://fazm.ai/t/latest-open-source-llm-releases-2026"
  },
  {
   "model": "Gemma 4 12B",
   "organization": "Google DeepMind",
   "release_date": "2026-06-03",
   "parameters": "12B",
   "context_window": "128K",
   "license": "Apache 2.0",
   "source_count": 4,
   "significance": "unified multimodal architecture; processes images and audio without separate encoders",
   "example_url": "https://aiproductivity.ai/news/google-gemma-4-12b-open-model-release-hugging-face"
  },
  {
   "model": "Nemotron 3 Ultra",
   "organization": "NVIDIA",
   "release_date": "2026-06-04",
   "parameters": "550-billion-parameter",
   "context_window": "1 million tokens",
   "license": "commercial-friendly license",
   "source_count": 5,
   "significance": "Tops US open-weight rankings with an Intelligence Index of 48",
   "example_url": "https://techfastforward.com/articles/nvidia-nemotron-3-ultra-beats-us-open-weight-rivals"
  },
  {
   "model": "Kimi K2.7 Code",
   "organization": "Moonshot AI",
   "release_date": "2026-06-12",
   "parameters": "~1T total (MoE)",
   "context_window": "Agentic coding",
   "license": "Modified MIT",
   "source_count": 4,
   "significance": "Coding-focused refresh of the K2 line shipped the same week as GLM-5.2.",
   "example_url": "https://fazm.ai/t/latest-open-source-llm-releases-2026"
  },
  {
   "model": "GLM 5.2",
   "organization": "Zhipu AI",
   "release_date": "2026-06-13",
   "parameters": "744B (40B active)",
   "context_window": "1 million tokens",
   "license": "MIT",
   "source_count": 20,
   "significance": "Successor to GLM-5.1 (vendor-reported 58.4% SWE-Bench Pro). Positioned for long-horizon agentic engineering.",
   "example_url": "https://fazm.ai/t/latest-open-source-llm-releases-2026"
  },
  {
   "model": "Qwen-AgentWorld-35B-A3B",
   "organization": "Qwen",
   "release_date": "2026-06-24",
   "parameters": "35B total / 3B active",
   "context_window": "256K",
   "license": "Apache 2.0",
   "source_count": 2,
   "significance": "Qwen-AgentWorld-35B-A3B shows +8.66 improvement over Qwen3.5-35B-A3B without LWM training.",
   "example_url": "https://github.com/QwenLM/Qwen-AgentWorld/blob/main/README.md"
  },
  {
   "model": "Leanstral 1.5",
   "organization": "Mistral AI",
   "release_date": "2026-07-02",
   "parameters": "119B",
   "context_window": "256,000-token",
   "license": "Apache-2.0",
   "source_count": 2,
   "significance": "saturates miniF2F, solves 587 of 672 PutnamBench problems, and sets new state-of-the-art scores on two algebra proof benchmarks",
   "example_url": "https://theagenttimes.com/articles/mistral-ai-releases-leanstral-1-5-with-119b-parameters-under-b10303d1"
  },
  {
   "model": "Inkling",
   "organization": "Thinking Machines Lab",
   "release_date": "2026-07-15",
   "parameters": "975B-parameter",
   "context_window": null,
   "license": "Apache 2.0",
   "source_count": 3,
   "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",
   "example_url": "https://best-ai.news/ai-model-releases-2026"
  },
  {
   "model": "Laguna S 2.1",
   "organization": "Poolside",
   "release_date": "2026-07-21",
   "parameters": "118B-parameter",
   "context_window": "1M-token",
   "license": "OpenMDW-1.1",
   "source_count": 2,
   "significance": "First-party benchmarks show 78.5% on SWE-Bench Multilingual, but independent verification is still pending.",
   "example_url": "https://teqvolt.com/tag/open-weights"
  },
  {
   "model": "Kimi K3",
   "organization": "Moonshot AI",
   "release_date": "2026-07-27",
   "parameters": "2.8-trillion-parameter",
   "context_window": "1-million-token",
   "license": "Kimi K3 License",
   "source_count": 28,
   "significance": "matched or surpassed Anthropic's Fable 5 on key benchmarks at roughly half the cost, erasing the closed-source lead",
   "example_url": "https://edgen.tech/tr/news/post/kimi-k3-open-weight-release-draws-100k-downloads-erases-us-ai-lead"
  },
  {
   "model": "DeepSeek V4 Flash 0731",
   "organization": "DeepSeek",
   "release_date": "2026-07-31",
   "parameters": null,
   "context_window": null,
   "license": "MIT License",
   "source_count": 2,
   "significance": "Terminal-Bench 82.7% \u2014 the cheapest capable agentic model with published benchmarks",
   "example_url": "https://vorplabs.com/models/open-weights-tracker"
  },
  {
   "model": "Muse Glimmer",
   "organization": "Meta",
   "release_date": "2026-08-10",
   "parameters": "30-billion-parameter",
   "context_window": "128K",
   "license": "Apache 2.0",
   "source_count": 4,
   "significance": "a 30-billion-parameter model distilled from Muse Spark",
   "example_url": "https://absolutedigitalpublishers.com/articles/llama-mistral-deepseek-qwen-gemma-and-olmo-what-open-actually-means"
  },
  {
   "model": "Qwen3.8-27B",
   "organization": "Alibaba",
   "release_date": "2026-08-14",
   "parameters": "27B",
   "context_window": "262K",
   "license": "Apache 2.0",
   "source_count": 2,
   "significance": "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.",
   "example_url": "https://codersera.com/blog/open-source-llms-landscape-2026"
  },
  {
   "model": "GLM-5.3-Flash",
   "organization": "Z.ai",
   "release_date": "2026-08-26",
   "parameters": "320-billion-parameter",
   "context_window": "above one million tokens",
   "license": "MIT license",
   "source_count": 2,
   "significance": "the first natively multimodal model in the GLM-5 series: image and video input, a context window above one million tokens, a 63.4 result on the DeepSWE evaluation shown on the model page",
   "example_url": "https://startupfortune.com/zhipus-glm-53-flash-undercuts-claude-and-gpt-on-price-not-on-hardware"
  }
 ],
 "notes": []
}