Keenable SELECT showcaseAsk your own question →GitHub

Major Open & Open-Weight LLM Releases — 2026 Timeline

Trajectory · 14 messages, 9 tool calls · report · all reports

user

assemble a release time line of major open source LLMs being released in 2026

assistant

tool call · select

SELECT
  url,
  title,
  published_at,
  UNNEST(
    SEM_EXTRACT_ALL(
      content,
      'a major openly released large language model or major new model-family generation released during 2026, excluding proprietary API-only models and minor patches',
      organization := 'developer or releasing organization',
      model := 'official model or model-family release name',
      release_date := 'date publicly released or made available, YYYY-MM-DD',
      access_type := 'whether source calls it open source, open weight, or openly available',
      license := 'license name or stated usage restrictions',
      significance := 'one short clause describing capabilities, scale, or why the launch was major',
      evidence := 'per_field'
    ),
    recursive := true
  )
FROM WEB_SEARCH(
  '2026 open source LLM releases timeline',
  '2026 open weight language model launch news',
  'site:huggingface.co/blog 2026 new open model LLM',
  'site:techcrunch.com 2026 open source AI model released',
  'site:venturebeat.com 2026 open source LLM launch',
  'site:artificialanalysis.ai 2026 open weights model release',
  '2026 Chinese open source large language model launch',
  '2026 reasoning model open weights announcement'
)
WHERE SEM_MATCH(
  content,
  'reports a public release in calendar year 2026 of a major large language model with downloadable weights or source, and gives the model name and release timing'
)

{
 "show_preview": false
}

tool call · select

SELECT
  url,
  title,
  published_at,
  UNNEST(
    SEM_EXTRACT_ALL(
      content,
      'a major downloadable/open-weight large language model released in 2026 by one of the named developers',
      organization := 'developer or releasing organization',
      model := 'official model or model-family release name',
      release_date := 'date publicly released or made available, YYYY-MM-DD',
      parameters := 'parameter count or architecture size as stated',
      context_window := 'context window as stated',
      license := 'license name or stated usage restrictions',
      significance := 'major new capabilities or benchmark positioning stated on the page',
      evidence := 'per_field'
    ),
    recursive := true
  )
FROM WEB_SEARCH(
  'DeepSeek 2026 model open weights release',
  'Qwen 2026 LLM open source release official',
  'Meta Llama 2026 open model release official',
  'Mistral 2026 open model release official',
  'Google Gemma 2026 open model release official',
  'Moonshot Kimi 2026 open source model release',
  'Zhipu GLM 2026 open source model release',
  'MiniMax 2026 open source LLM release',
  'NVIDIA 2026 open language model release',
  'Allen AI 2026 open language model release'
)
WHERE SEM_MATCH(
  content,
  'states that a specific large language model was released during 2026 with weights publicly downloadable or under an open license'
)

{
 "show_preview": false
}

tool result (4873 chars)

result_set_id: r2f99bf28246
total_rows: 1488
columns[15]: access_type,access_type_evidence,license,license_evidence,model,model_evidence,organization,organization_evidence,published_at,release_date,release_date_evidence,significance,significance_evidence,title,url
preview:
  sample_rows[3]:
    - 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[1]: "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[2]: "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[3]: "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://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[2]: "## The Great Unshackling: Major Releases That Changed the Game","In February, [Hugging Face’s 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[2]: "## The Great Unshackling: Major Releases That Changed the Game","In February, [Hugging Face’s 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[2]: "## The Great Unshackling: Major Releases That Changed the Game","In February, [Hugging Face’s 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://knowledgenuts.com/2026/07/17/what-is-inkling-thinking-machines-labs-open-weight-ai-model-explained"
      title: What Is Inkling? Thinking Machines Lab’s 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[1]: "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[1]: "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[3]: "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... [truncated in preview]"
notes[3,]{text}:
  "The full table is too large to return in one response. This is a preview, not the data — do not draw conclusions from it alone. Query the result set by id: SEM_EXTRACT fields, filter with exact WHERE, aggregate, or page with LIMIT/OFFSET. Even if the preview looks irrelevant, probe the full result set before searching again."
  Merged 8 queries (1566 total) → 1477 unique results
  "Search result set: r5247d91c07d (1477 rows), reference it as FROM r5247d91c07d in follow-up queries"

result set r2f99bf28246 · 1488 rows

url	title	published_at	organization	model	release_date	access_type	license	significance	organization_evidence	model_evidence	release_date_evidence	access_type_evidence	license_evidence	significance_evidence
https://en.wikipedia.org/wiki/GLM_(AI)	GLM (AI)	2026-08-30T18:59:23	Z.ai	null	2026-02	null	null	major model family generation starting the GLM-5 series utilizing the DeepSeek Sparse Attention mechanism	["GLM, short for General Language Model, is a series of open weight large language models developed by Chinese software company Z.ai."]	null	["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 202…	null	null	["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, …
https://en.wikipedia.org/wiki/GLM_(AI)	GLM (AI)	2026-08-30T18:59:23	Z.ai	null	2026-04	null	null	iteration of the GLM-5 series	["GLM, short for General Language Model, is a series of open weight large language models developed by Chinese software company Z.ai."]	null	["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…	null	null	["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…
https://en.wikipedia.org/wiki/GLM_(AI)	GLM (AI)	2026-08-30T18:59:23	Z.ai	null	2026-06	null	null	5.2 iteration of the model released in June 2026 was nearly as performant as Claude Opus 4.7 to 4.8	["GLM, short for General Language Model, is a series of open weight large language models developed by Chinese software company Z.ai."]	null	["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 benchmar…	null	null	["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 benchmar…
https://en.wikipedia.org/wiki/GLM_(AI)	GLM (AI)	2026-08-30T18:59:23	Z.ai	null	2026-08-14	null	null	major model version whose weights were made available two weeks after its August 14 release	["GLM, short for General Language Model, is a series of open weight large language models developed by Chinese software company Z.ai."]	null	["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 …	null	null	["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 …
https://en.wikipedia.org/wiki/GLM_(AI)	GLM (AI)	2026-08-30T18:59:23	Z.ai	null	2026-08-26	null	null	model incorporating Kimi Delta Attention developed by Moonshot AI and served using AI chips manufactured in China	["GLM, short for General Language Model, is a series of open weight large language models developed by Chinese software company Z.ai."]	null	["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…	null	null	["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…

Download all 1488 rows · 2394 KB

tool result (6923 chars)

result_set_id: r13f88a3ffda
total_rows: 1893
columns[17]: context_window,context_window_evidence,license,license_evidence,model,model_evidence,organization,organization_evidence,parameters,parameters_evidence,published_at,release_date,release_date_evidence,significance,significance_evidence,title,url
preview:
  sample_rows[3]:
    - url: "https://best-ai.news/ai-model-releases-2026"
      title: "Latest AI Models 2026: 99 Released — Complete Catalog"
      published_at: "2026-08-24T09:16:53"
      organization: Thinking Machines Lab
      model: Inkling
      release_date: 2026-07
      parameters: 975B-parameter
      context_window: null
      license: Apache 2.0
      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[3]: Latest AI Models 2026 — 99 Released,## July 202615 models,"Inkling — 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... [truncated in preview]"
      model_evidence[3]: Latest AI Models 2026 — 99 Released,## July 202615 models,"Inkling — 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... [truncated in preview]"
      release_date_evidence[3]: Latest AI Models 2026 — 99 Released,## July 202615 models,"Inkling — 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... [truncated in preview]"
      parameters_evidence[3]: Latest AI Models 2026 — 99 Released,## July 202615 models,"Inkling — 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... [truncated in preview]"
      context_window_evidence: null
      license_evidence[3]: Latest AI Models 2026 — 99 Released,## July 202615 models,"Inkling — 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... [truncated in preview]"
      significance_evidence[3]: Latest AI Models 2026 — 99 Released,## July 202615 models,"Inkling — 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... [truncated in preview]"
    - url: "https://agenticbrew.ai/news/bf8b9700-d02e-4add-b00b-7b20e0a9ab18/moonshot-ai-s-kimi-k3-open-weight-model-release-and-rollout"
      title: Moonshot AI's Kimi K3 open-weight model release and rollout
      published_at: "2026-08-02T00:36:29"
      organization: Moonshot AI
      model: null
      release_date: 2026-07-16
      parameters: 2.8-trillion-parameter mixture-of-experts model with 104 billion activated parameters
      context_window: null
      license: null
      significance: "Moonshot AI unveiled Kimi K3 on July 16, 2026 as a 2.8-trillion-parameter mixture-of-experts model with 104 billion activated parameters, a 1-million-token context window, and native... [truncated in preview]"
      organization_evidence[1]: "Moonshot AI unveiled Kimi K3 on July 16, 2026 as a 2.8-trillion-parameter mixture-of-experts model with 104 billion activated parameters, a 1-million-token context window, and native... [truncated in preview]"
      model_evidence: null
      release_date_evidence[1]: "Moonshot AI unveiled Kimi K3 on July 16, 2026 as a 2.8-trillion-parameter mixture-of-experts model with 104 billion activated parameters, a 1-million-token context window, and native... [truncated in preview]"
      parameters_evidence[1]: "Moonshot AI unveiled Kimi K3 on July 16, 2026 as a 2.8-trillion-parameter mixture-of-experts model with 104 billion activated parameters, a 1-million-token context window, and native... [truncated in preview]"
      context_window_evidence: null
      license_evidence: null
      significance_evidence[1]: "Moonshot AI unveiled Kimi K3 on July 16, 2026 as a 2.8-trillion-parameter mixture-of-experts model with 104 billion activated parameters, a 1-million-token context window, and native... [truncated in preview]"
    - url: "https://aiproductivity.ai/news/nvidia-diffusiongemma-26b-nvfp4-local-model"
      title: "NVIDIA DiffusionGemma 26B: Run Locally with NVFP4"
      published_at: "2026-06-11T03:28:47"
      organization: NVIDIA
      model: null
      release_date: 2026-06-11
      parameters: 26B
      context_window: null
      license: null
      significance: A diffusion language model compressed with NVFP4 quantization for local inference on NVIDIA hardware.
      organization_evidence[3]: Open Source,NVIDIA Releases Quantized DiffusionGemma 26B on Hugging Face,"NVIDIA published [nvidia/diffusiongemma-26B-A4B-it-NVFP4](https://huggingface.co/nvidia/diffusiongemma-26B-A4B-it-NVFP4) on Hugging Face on June 11, an instruction-tuned language model compressed for local inference on NVIDIA hardware."
      model_evidence: null
      release_date_evidence[4]: Open Source,NVIDIA Releases Quantized DiffusionGemma 26B on Hugging Face,"June 11, 2026 2 min read",... +1 more items
      parameters_evidence[3]: Open Source,NVIDIA Releases Quantized DiffusionGemma 26B on Hugging Face,"The \"26B\" refers to total parameters, but this is a mixture-of-experts model, meaning only a subset of the network activates per query - specifically 4... [truncated in preview]"
      context_window_evidence: null
      license_evidence: null
      significance_evidence[3]: Open Source,NVIDIA Releases Quantized DiffusionGemma 26B on Hugging Face,"NVIDIA published [nvidia/diffusiongemma-26B-A4B-it-NVFP4](https://huggingface.co/nvidia/diffusiongemma-26B-A4B-it-NVFP4) on Hugging Face on June 11, an instruction-tuned language model compressed for local inference on NVIDIA hardware."
notes[4,]{text}:
  "The full table is too large to return in one response. This is a preview, not the data — do not draw conclusions from it alone. Query the result set by id: SEM_EXTRACT fields, filter with exact WHERE, aggregate, or page with LIMIT/OFFSET. Even if the preview looks irrelevant, probe the full result set before searching again."
  Merged 10 queries (2000 total) → 1913 unique results
  "Search result set: r701f9a0e854 (1913 rows), reference it as FROM r701f9a0e854 in follow-up queries"
  "WARNING: semantic extraction failed on 1 of 1913 rows — extracted values are null on those rows. First error: timed_out after 90.0s"

result set r13f88a3ffda · 1893 rows

url	title	published_at	organization	model	release_date	parameters	context_window	license	significance	organization_evidence	model_evidence	release_date_evidence	parameters_evidence	context_window_evidence	license_evidence	significance_evidence
https://best-ai.news/ai-model-releases-2026	Latest AI Models 2026: 99 Released — Complete Catalog	2026-08-24T09:16:53	Thinking Machines Lab	Inkling	2026-07	975B-parameter	null	Apache 2.0	Thinking Machines Lab's first model trained from scratch: a 975B-parameter open-weights multimodal Mixture-of-Experts with 41B active parameters and controllab…	["Latest AI Models 2026 — 99 Released", "## July 202615 models", "Inkling — Thinking Machines Lab's first model trained from scratch: a 975B-parameter open-wei…	["Latest AI Models 2026 — 99 Released", "## July 202615 models", "Inkling — Thinking Machines Lab's first model trained from scratch: a 975B-parameter open-wei…	["Latest AI Models 2026 — 99 Released", "## July 202615 models", "Inkling — Thinking Machines Lab's first model trained from scratch: a 975B-parameter open-wei…	["Latest AI Models 2026 — 99 Released", "## July 202615 models", "Inkling — Thinking Machines Lab's first model trained from scratch: a 975B-parameter open-wei…	null	["Latest AI Models 2026 — 99 Released", "## July 202615 models", "Inkling — Thinking Machines Lab's first model trained from scratch: a 975B-parameter open-wei…	["Latest AI Models 2026 — 99 Released", "## July 202615 models", "Inkling — Thinking Machines Lab's first model trained from scratch: a 975B-parameter open-wei…
https://arxiv.org/html/2604.07190v1	The ATOM Report:Measuring the Open Language Model Ecosystem	null	Alibaba	null	2026-02	3.5B	null	null	Placed among the hottest launches in the reference set, reaching high relative adoption scores across 7, 14, and 30 day milestones.	["# The ATOM Report:\nMeasuring the Open Language Model Ecosystem", "## 1 Introduction", "The growth of the Chinese model ecosystem can largely be attributed t…	null	["# The ATOM Report:\nMeasuring the Open Language Model Ecosystem", "## 6 The Relative Adoption Metric (RAM)", "### 6.2 Using RAM to Measure New Models", "For …	["# The ATOM Report:\nMeasuring the Open Language Model Ecosystem", "## 6 The Relative Adoption Metric (RAM)", "### 6.2 Using RAM to Measure New Models", "In t…	null	null	["# The ATOM Report:\nMeasuring the Open Language Model Ecosystem", "## 6 The Relative Adoption Metric (RAM)", "### 6.2 Using RAM to Measure New Models", "The …
https://hidekazu-konishi.com/entry/open_weights_llm_release_history_and_timeline.html	Open-Weights LLM Release History and Timeline - Llama, Mistral, Qwen, DeepSeek, and gpt-oss Model Families, Capability Evolution, and Licenses	2026-07-26T00:00:00	Qwen	null	2026-02-16	397B total / 17B active	null	null	integrating advances in multimodal learning, architectural efficiency, and large-scale reinforcement learning	["## Open-Weights LLM Release History and Timeline - Llama, Mistral, Qwen, DeepSeek, and gpt-oss Model Families, Capability Evolution, and Licenses", "Here is …	null	["Here is the chronological timeline of major open-weights model releases across the families above, in ascending order.", "\| Date \| Family \| Release \|", "\| 20…	["Here is the chronological timeline of major open-weights model releases across the families above, in ascending order.", "\| Date \| Family \| Release \|", "\| 20…	null	null	["Here is the chronological timeline of major open-weights model releases across the families above, in ascending order.", "\| Date \| Family \| Release \|", "\| 20…
https://hidekazu-konishi.com/entry/open_weights_llm_release_history_and_timeline.html	Open-Weights LLM Release History and Timeline - Llama, Mistral, Qwen, DeepSeek, and gpt-oss Model Families, Capability Evolution, and Licenses	2026-07-26T00:00:00	Mistral AI	Mistral Small 4	2026-03-16	null	256K	Apache 2.0	unifies reasoning, multimodal understanding, and agentic coding in one model	["## Open-Weights LLM Release History and Timeline - Llama, Mistral, Qwen, DeepSeek, and gpt-oss Model Families, Capability Evolution, and Licenses", "Here is …	["Here is the chronological timeline of major open-weights model releases across the families above, in ascending order.", "\| Date \| Family \| Release \|", "\| 20…	["Here is the chronological timeline of major open-weights model releases across the families above, in ascending order.", "\| Date \| Family \| Release \|", "\| 20…	null	["Here is the chronological timeline of major open-weights model releases across the families above, in ascending order.", "\| Date \| Family \| Release \|", "\| 20…	["Here is the chronological timeline of major open-weights model releases across the families above, in ascending order.", "\| Date \| Family \| Release \|", "\| 20…	["Here is the chronological timeline of major open-weights model releases across the families above, in ascending order.", "\| Date \| Family \| Release \|", "\| 20…
https://hidekazu-konishi.com/entry/open_weights_llm_release_history_and_timeline.html	Open-Weights LLM Release History and Timeline - Llama, Mistral, Qwen, DeepSeek, and gpt-oss Model Families, Capability Evolution, and Licenses	2026-07-26T00:00:00	Google	Gemma 4	2026-04-02	edge E2B/E4B sizes plus larger dense and Mixture-of-Experts models	null	Apache 2.0	focus on agentic workflows	["## Open-Weights LLM Release History and Timeline - Llama, Mistral, Qwen, DeepSeek, and gpt-oss Model Families, Capability Evolution, and Licenses", "Here is …	["Here is the chronological timeline of major open-weights model releases across the families above, in ascending order.", "\| Date \| Family \| Release \|", "\| 20…	["Here is the chronological timeline of major open-weights model releases across the families above, in ascending order.", "\| Date \| Family \| Release \|", "\| 20…	["Here is the chronological timeline of major open-weights model releases across the families above, in ascending order.", "\| Date \| Family \| Release \|", "\| 20…	null	["Here is the chronological timeline of major open-weights model releases across the families above, in ascending order.", "\| Date \| Family \| Release \|", "\| 20…	["Here is the chronological timeline of major open-weights model releases across the families above, in ascending order.", "\| Date \| Family \| Release \|", "\| 20…

Download all 1893 rows · 3535 KB

assistant

tool call · select

SELECT
  SEM_NORM(model, 'merge spelling and size variants only when they are one release event; keep distinct generations and materially distinct releases apart') AS model,
  MODE(organization) AS organization,
  MODE(release_date) AS release_date,
  MODE(parameters) AS parameters,
  MODE(context_window) AS context_window,
  MODE(license) AS license,
  COUNT(DISTINCT url) AS source_count,
  ANY_VALUE(significance) AS significance,
  ANY_VALUE(url) AS example_url
FROM (
  SELECT
    organization,
    model,
    release_date,
    parameters,
    context_window,
    license,
    significance,
    url
  FROM r13f88a3ffda
  WHERE model IS NOT NULL
    AND release_date LIKE '2026%'
    AND organization IS NOT NULL
)
GROUP BY SEM_NORM(model, 'merge spelling and size variants only when they are one release event; keep distinct generations and materially distinct releases apart')
HAVING COUNT(DISTINCT url) >= 2
ORDER BY release_date,
  source_count DESC
LIMIT 100

{
 "show_preview": false
}

tool result (12158 chars)

result_set_id: r8f4c29f3a52
total_rows: 47
columns[9]: context_window,example_url,license,model,organization,parameters,release_date,significance,source_count
rows[47,]{model,organization,release_date,parameters,context_window,license,source_count,significance,example_url}:
  Llama 4 Scout,Meta,"2026",109B MoE / 17B active (16 experts),10M,Meta custom (700M MAU clause),7,Industry-leading context window. The license is not OSI-approved and carries a monthly-active-user ceiling.,"https://fazm.ai/t/latest-open-source-llm-releases-2026"
  Qwen3-Coder-Next,Alibaba,2026-02,80B / 3B,128K,Apache 2.0,2,Outperforms much larger models like DeepSeek V3.2 (37B active) on coding tasks at a fraction of the inference cost,"https://klymentiev.com/blog/best-open-source-llm-2026"
  Qwen 3.5 397B,Alibaba Cloud,2026-02,397B (17B active),256K,Apache 2.0,2,Flagship open-weight,"https://techunfoldedai.com/qwen-ai"
  Qwen 3.5 122B-A10B,Alibaba,2026-02-01,122B / 10B,256K,Apache 2.0,3,Medium model series variant combining high-throughput inference with efficient sparse Mixture-of-Experts architecture.,"https://tech-insider.org/llama-4-vs-qwen-vs-mistral-2026"
  Qwen 3.5 27B,Alibaba,2026-02-01,27B / 27B,256K,Apache 2.0,3,"Efficient and high-performing dense model delivering strong results in coding, reasoning, and output consistency.","https://tech-insider.org/llama-4-vs-qwen-vs-mistral-2026"
  MiniMax M2.5,MiniMax,2026-02-12,null,204K,Open,2,"Agent-native coding flagship released weeks after MiniMax's Hong Kong IPO, with top-tier SWE-bench results at a fraction of frontier cost.","https://outyet.ai/providers/minimax"
  GLM-5,Zhipu AI,2026-02-13,744B (44B active),1 million-token,MIT,10,"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","https://z.ai/blog/glm-5"
  Qwen 3.5,Alibaba,2026-02-16,17B active / 397B total,256K,Apache 2.0,4,Qwen3.5 (397B total / 17B active MoE); 256K native context; 256 experts,"https://jaredwatkins.com/research/local-llm-finetuning/fundamentals/base-models"
  Mistral Large 3,Mistral,2026-03-06,41B active parameters and 675B total parameters,256K tokens,Apache 2.0,2,Currently ranks as the #2 open-source non-reasoning model on LMArena,"https://insights.marvin-42.com/articles/mistral-launches-mistral-3-open-multimodal-family-under-apache-20"
  Nemotron 3 Super,NVIDIA,2026-03-11,120.6B,Up to 1M,Open weights,4,36.0 on the Artificial Analysis Intelligence Index,"https://tech-insider.org/ca/nvidia-nemotron-3-ultra-2026"
  Mistral Small 4,Mistral AI,2026-03-16,119B / 6B,"256,000",Apache 2.0,11,"unifies reasoning, multimodal understanding, and agentic coding in one model","https://hidekazu-konishi.com/entry/open_weights_llm_release_history_and_timeline.html"
  Mistral Small (latest),Mistral,2026-03-16,119B,"256,000",Open,2,null,"https://models.opencode.ai/labs/mistral"
  Qwen3.5-Omni,Alibaba,2026-03-30,null,256K,null,2,"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.","https://ai2.work/blog/alibaba-s-three-model-blitz-fuels-the-open-weight-ai-arms-race"
  GLM-5.1,Z.ai,2026-04,754B,200K,MIT,13,"91.7% MMLU, 85.7% GPQA, 58.4% SWE-Bench Pro","https://qtithow.com/2026/07/comparing-120-open-source-llms-best.html"
  Llama 4 Maverick,Meta,2026-04,400B,1M,Llama Community,6,"matches GPT-4o and Gemini 2.0 Flash at an LMArena ELO score of 1,417","https://temperaturezero.com/2026/05/18/meta-open-sourced-llama-4-kept-muse-spark"
  Gemma 4,Google,2026-04-02,edge E2B/E4B sizes plus larger dense and Mixture-of-Experts models,256K,Apache 2.0,22,focus on agentic workflows,"https://hidekazu-konishi.com/entry/open_weights_llm_release_history_and_timeline.html"
  Gemma 4 E4B,Google,2026-04-02,"8B (Effective: ~4.5B)",128K,Apache 2.0,7,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.,"https://decodethefuture.org/en/google-gemma-4-explained"
  Gemma 4 26B Mixture of Experts,Google DeepMind,2026-04-02,26B (3.8B active),256K,Apache 2.0,6,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.,"https://decodethefuture.org/en/google-gemma-4-explained"
  Gemma 4 31B Dense,Google DeepMind,2026-04-02,31B,256K,Apache 2.0,5,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.,"https://decodethefuture.org/en/google-gemma-4-explained"
  Gemma 4 E2B,Google,2026-04-02,"5.1B (Effective: ~2.3B)",128K,Apache 2.0,5,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.,"https://decodethefuture.org/en/google-gemma-4-explained"
  Gemma 4 31B,Google,2026-04-02,30.7B,256K,Apache 2.0,2,"LMArena score of 1,452 — 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).","https://aiautomationglobal.com/blog/google-gemma-4-open-source-apache-agentic-ai-2026"
  Gemma 4 26B A4B,Google,2026-04-02,25.2B,256K,Apache 2.0,2,Ranks #6 among all open-weight models globally. Google calls it “the world’s most parameter-efficient model at this scale.”,"https://aiautomationglobal.com/blog/google-gemma-4-open-source-apache-agentic-ai-2026"
  MiniMax M2.7,MiniMax,2026-04-12,"229B total parameters, 10B active per token",200K tokens,Modified MIT,6,"Hits 56.22% on SWE-Bench Pro and 57.0% on Terminal Bench 2, with downloads already in the tens of thousands.","https://ai-tldr.dev/releases/minimax-m2-7-open-weights"
  Qwen 3.6-35B-A3B,Alibaba,2026-04-16,35B-A3B,256K,Apache 2.0,11,"Explicitly tuned for agentic coding, activating only 3B parameters per token.","https://tech-insider.org/llama-4-vs-qwen-vs-mistral-2026"
  Qwen 3.6,Alibaba,2026-04-16,35B-A3B and 27B,1M,Apache 2.0,5,"building on Qwen3.5 while prioritizing stability, real-world coding utility, and direct community feedback","https://hidekazu-konishi.com/entry/open_weights_llm_release_history_and_timeline.html"
  Kimi K2.6,Moonshot AI,2026-04-20,1T total / 32B active (MoE),256K,Modified MIT,7,"Native multimodal agentic model. MIT-equivalent below 100M MAU / $20M monthly revenue, attribution clause above.","https://fazm.ai/t/latest-open-source-llm-releases-2026"
  Qwen 3.6 27B,Alibaba,2026-04-22,27B,256K,Apache 2.0,8,The local dev sweet spot,"https://digitalapplied.com/blog/open-weight-models-h1-2026-retrospective-deepseek-qwen-llama"
  DeepSeek V4 Flash,DeepSeek,2026-04-24,284B total / 13B active,"1,000,000 tokens",MIT,24,"Terminal Bench 2.1 score of 82.7, a massive +25.8 point leap from its initial April preview score of 56.9","https://cline.bot/blog/best-open-weight-models-that-matter-in-2026"
  DeepSeek V4 Pro,DeepSeek,2026-04-24,1.6 trillion total parameters,"1,000,000 tokens",MIT,22,"Reported at 80.6% on SWE-bench Verified, putting it in the same range as leading closed models","https://cline.bot/blog/best-open-weight-models-that-matter-in-2026"
  DeepSeek V4,DeepSeek,2026-04-24,V4-Pro 1.6T/49B active; V4-Flash 284B/13B active,1M,MIT,8,"Two MoE sizes, native 1M context, trained on 32T+ tokens. The clean-license frontier MoE of the spring.","https://fazm.ai/t/latest-open-source-llm-releases-2026"
  DeepSeek V4 Preview,DeepSeek,2026-04-24,DeepSeek-V4-Pro (1.6T total / 49B active parameters) and DeepSeek-V4-Flash (284B total / 13B active),1M-token,open weights,2,"built on token-wise compression and DeepSeek Sparse Attention, and making a 1M-token context window the standard across DeepSeek's official services","https://hidekazu-konishi.com/entry/open_weights_llm_release_history_and_timeline.html"
  Mistral Medium 3.5,Mistral,2026-04-29,128B,256K,modified MIT license,10,"merges instruction-following, reasoning, and coding in a single set of weights","https://hidekazu-konishi.com/entry/open_weights_llm_release_history_and_timeline.html"
  Mistral Medium (latest),Mistral,2026-04-29,null,"262,144",Open,2,null,"https://models.opencode.ai/labs/mistral"
  MiniMax M3,MiniMax,2026-06-01,Undisclosed (MoE),1M,MiniMax Community License,16,"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.","https://fazm.ai/t/latest-open-source-llm-releases-2026"
  Gemma 4 12B,Google DeepMind,2026-06-03,12B,128K,Apache 2.0,4,unified multimodal architecture; processes images and audio without separate encoders,"https://aiproductivity.ai/news/google-gemma-4-12b-open-model-release-hugging-face"
  Nemotron 3 Ultra,NVIDIA,2026-06-04,550-billion-parameter,1 million tokens,commercial-friendly license,5,Tops US open-weight rankings with an Intelligence Index of 48,"https://techfastforward.com/articles/nvidia-nemotron-3-ultra-beats-us-open-weight-rivals"
  Kimi K2.7 Code,Moonshot AI,2026-06-12,~1T total (MoE),Agentic coding,Modified MIT,4,Coding-focused refresh of the K2 line shipped the same week as GLM-5.2.,"https://fazm.ai/t/latest-open-source-llm-releases-2026"
  GLM 5.2,Zhipu AI,2026-06-13,744B (40B active),1 million tokens,MIT,20,Successor to GLM-5.1 (vendor-reported 58.4% SWE-Bench Pro). Positioned for long-horizon agentic engineering.,"https://fazm.ai/t/latest-open-source-llm-releases-2026"
  Qwen-AgentWorld-35B-A3B,Qwen,2026-06-24,35B total / 3B active,256K,Apache 2.0,2,Qwen-AgentWorld-35B-A3B shows +8.66 improvement over Qwen3.5-35B-A3B without LWM training.,"https://github.com/QwenLM/Qwen-AgentWorld/blob/main/README.md"
  Leanstral 1.5,Mistral AI,2026-07-02,119B,"256,000-token",Apache-2.0,2,"saturates miniF2F, solves 587 of 672 PutnamBench problems, and sets new state-of-the-art scores on two algebra proof benchmarks","https://theagenttimes.com/articles/mistral-ai-releases-leanstral-1-5-with-119b-parameters-under-b10303d1"
  Inkling,Thinking Machines Lab,2026-07-15,975B-parameter,null,Apache 2.0,3,"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","https://best-ai.news/ai-model-releases-2026"
  Laguna S 2.1,Poolside,2026-07-21,118B-parameter,1M-token,OpenMDW-1.1,2,"First-party benchmarks show 78.5% on SWE-Bench Multilingual, but independent verification is still pending.","https://teqvolt.com/tag/open-weights"
  Kimi K3,Moonshot AI,2026-07-27,2.8-trillion-parameter,1-million-token,Kimi K3 License,28,"matched or surpassed Anthropic's Fable 5 on key benchmarks at roughly half the cost, erasing the closed-source lead","https://edgen.tech/tr/news/post/kimi-k3-open-weight-release-draws-100k-downloads-erases-us-ai-lead"
  DeepSeek V4 Flash 0731,DeepSeek,2026-07-31,null,null,MIT License,2,Terminal-Bench 82.7% — the cheapest capable agentic model with published benchmarks,"https://vorplabs.com/models/open-weights-tracker"
  Muse Glimmer,Meta,2026-08-10,30-billion-parameter,128K,Apache 2.0,4,a 30-billion-parameter model distilled from Muse Spark,"https://absolutedigitalpublishers.com/articles/llama-mistral-deepseek-qwen-gemma-and-olmo-what-open-actually-means"
  Qwen3.8-27B,Alibaba,2026-08-14,27B,262K,Apache 2.0,2,"Artificial Analysis independently scores it 52 on its Intelligence Index — 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–150B range.","https://codersera.com/blog/open-source-llms-landscape-2026"
  GLM-5.3-Flash,Z.ai,2026-08-26,320-billion-parameter,above one million tokens,MIT license,2,"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","https://startupfortune.com/zhipus-glm-53-flash-undercuts-claude-and-gpt-on-price-not-on-hardware"
notes[0]:

result set r8f4c29f3a52 · 47 rows

model	organization	release_date	parameters	context_window	license	source_count	significance	example_url
Llama 4 Scout	Meta	2026	109B MoE / 17B active (16 experts)	10M	Meta custom (700M MAU clause)	7	Industry-leading context window. The license is not OSI-approved and carries a monthly-active-user ceiling.	https://fazm.ai/t/latest-open-source-llm-releases-2026
Qwen3-Coder-Next	Alibaba	2026-02	80B / 3B	128K	Apache 2.0	2	Outperforms much larger models like DeepSeek V3.2 (37B active) on coding tasks at a fraction of the inference cost	https://klymentiev.com/blog/best-open-source-llm-2026
Qwen 3.5 397B	Alibaba Cloud	2026-02	397B (17B active)	256K	Apache 2.0	2	Flagship open-weight	https://techunfoldedai.com/qwen-ai
Qwen 3.5 122B-A10B	Alibaba	2026-02-01	122B / 10B	256K	Apache 2.0	3	Medium model series variant combining high-throughput inference with efficient sparse Mixture-of-Experts architecture.	https://tech-insider.org/llama-4-vs-qwen-vs-mistral-2026
Qwen 3.5 27B	Alibaba	2026-02-01	27B / 27B	256K	Apache 2.0	3	Efficient and high-performing dense model delivering strong results in coding, reasoning, and output consistency.	https://tech-insider.org/llama-4-vs-qwen-vs-mistral-2026

Download all 47 rows · 20 KB

assistant

tool call · select

SELECT
  url,
  title,
  published_at,
  SEM_EXTRACT(
    content,
    'the official release announcement for the named open-weight model',
    organization := 'releasing organization',
    model := 'official released model-family name',
    release_date := 'date weights or model were publicly released, YYYY-MM-DD',
    parameters := 'total and active parameter count as stated',
    context_window := 'maximum context length as stated',
    license := 'license as stated',
    summary := 'main capabilities or intended use, one concise clause',
    evidence := 'per_field'
  ) AS release
FROM WEB_SEARCH(
  'site:qwenlm.github.io/blog Qwen3.5 2026',
  'site:github.com/QwenLM Qwen3.6 release',
  'site:qwenlm.github.io/blog Qwen3.8 27B',
  'site:huggingface.co/Qwen Qwen3.5 model card',
  'site:alibabacloud.com Qwen3.5 release 2026',
  'site:github.com/QwenLM 2026 releases Qwen',
  'Qwen3.5 official release February 2026',
  'Qwen3.6 official release April 2026'
)
WHERE SEM_MATCH(content, 'is an official developer page or first-party model card that confirms a 2026 public model release')

{
 "show_preview": false
}

tool call · select

SELECT
  url,
  title,
  published_at,
  SEM_EXTRACT(
    content,
    'the official release announcement for the named open-weight model',
    organization := 'releasing organization',
    model := 'official released model-family name',
    release_date := 'date weights or model were publicly released, YYYY-MM-DD',
    parameters := 'total and active parameter count as stated',
    context_window := 'maximum context length as stated',
    license := 'license as stated',
    summary := 'main capabilities or intended use, one concise clause',
    evidence := 'per_field'
  ) AS release
FROM WEB_SEARCH(
  'site:deepseek.com DeepSeek V4 release',
  'site:github.com/deepseek-ai DeepSeek-V4',
  'site:huggingface.co/deepseek-ai DeepSeek V4 model card',
  'site:z.ai/blog GLM-5 release',
  'site:github.com/zai-org GLM-5.2',
  'site:huggingface.co/zai-org GLM-5 model card',
  'site:docs.z.ai GLM-5.3 Flash',
  'DeepSeek V4 official release April 2026'
)
WHERE SEM_MATCH(content, 'is an official developer page or first-party model card that confirms a 2026 public model release')

{
 "show_preview": false
}

tool call · select

SELECT
  url,
  title,
  published_at,
  SEM_EXTRACT(
    content,
    'the official release announcement for the named open-weight model',
    organization := 'releasing organization',
    model := 'official released model-family name',
    release_date := 'date weights or model were publicly released, YYYY-MM-DD',
    parameters := 'total and active parameter count as stated',
    context_window := 'maximum context length as stated',
    license := 'license as stated',
    summary := 'main capabilities or intended use, one concise clause',
    evidence := 'per_field'
  ) AS release
FROM WEB_SEARCH(
  'site:mistral.ai/news Mistral Small 4',
  'site:mistral.ai/news Mistral Medium 3.5',
  'site:huggingface.co/mistralai Mistral Small 4 model card',
  'site:minimax.io/news MiniMax M2.5 release',
  'site:github.com/MiniMax-AI MiniMax M3',
  'site:huggingface.co/MiniMaxAI MiniMax M3 model card',
  'site:moonshotai.github.io Kimi K2.6 K3',
  'site:huggingface.co/moonshotai Kimi K3 model card'
)
WHERE SEM_MATCH(content, 'is an official developer page or first-party model card that confirms a 2026 public model release')

{
 "show_preview": false
}

tool call · select

SELECT
  url,
  title,
  published_at,
  SEM_EXTRACT(
    content,
    'the official release announcement for the named open-weight model',
    organization := 'releasing organization',
    model := 'official released model-family name',
    release_date := 'date weights or model were publicly released, YYYY-MM-DD',
    parameters := 'total and active parameter count as stated',
    context_window := 'maximum context length as stated',
    license := 'license as stated',
    summary := 'main capabilities or intended use, one concise clause',
    evidence := 'per_field'
  ) AS release
FROM WEB_SEARCH(
  'site:blog.google Gemma 4 release',
  'site:developers.googleblog.com Gemma 4',
  'site:huggingface.co/google Gemma 4 model card',
  'site:nvidia.com Nemotron 3 Ultra release',
  'site:huggingface.co/nvidia Nemotron 3 Ultra model card',
  'site:thinkingmachines.ai Inkling open weights',
  'site:huggingface.co Thinking Machines Inkling model card',
  'official open weight model release 2026 NVIDIA Google Thinking Machines'
)
WHERE SEM_MATCH(content, 'is an official developer page or first-party model card that confirms a 2026 public model release')

{
 "show_preview": false
}

tool result (6311 chars)

result_set_id: rc3e82e8cb59
total_rows: 4
columns[4]: published_at,release,title,url
rows[4]:
  - url: "https://huggingface.co/Qwen/Qwen3.5-0.8B"
    title: Qwen/Qwen3.5-0.8B · Hugging Face
    published_at: "2026-03-09T14:18:49"
    release:
      organization: Qwen Team
      model: Qwen3.5
      release_date: 2026-02
      parameters: 0.8B
      context_window: "262,144"
      license: null
      summary: "prototyping, task-specific fine-tuning, and other research or development purposes"
      organization_evidence[8]: # Qwen3.5-0.8B,### Citation,"@misc{qwen3.5,","    title  = {{Qwen3.5}: Towards Native Multimodal Agents},","    author = {{Qwen Team}},","    month  = {February},","    year   = {2026},","    url    = {https://qwen.ai/blog?id=qwen3.5}"
      model_evidence[3]: # Qwen3.5-0.8B,"Over recent months, we have intensified our focus on developing foundation models that deliver exceptional utility and performance.","Qwen3.5 represents a significant leap forward, integrating breakthroughs in multimodal learning, architectural efficiency, reinforcement learning scale, and global accessibility to empower developers and enterprises with unprecedented capability and efficiency."
      release_date_evidence[8]: # Qwen3.5-0.8B,### Citation,"@misc{qwen3.5,","    title  = {{Qwen3.5}: Towards Native Multimodal Agents},","    author = {{Qwen Team}},","    month  = {February},","    year   = {2026},","    url    = {https://qwen.ai/blog?id=qwen3.5}"
      parameters_evidence[3]: # Qwen3.5-0.8B,## Model Overview,"- Number of Parameters: 0.8B"
      context_window_evidence[3]: # Qwen3.5-0.8B,## Model Overview,"- Context Length: 262,144 natively"
      license_evidence: null
      summary_evidence[2]: # Qwen3.5-0.8B,"> In light of its parameter scale, the intended use cases are prototyping, task-specific fine-tuning, and other research or development purposes."
  - url: "https://huggingface.co/Qwen/Qwen3.5-4B"
    title: Qwen/Qwen3.5-4B · Hugging Face
    published_at: "2026-03-09T14:18:49"
    release:
      organization: Qwen Team
      model: Qwen3.5-4B
      release_date: 2026-02
      parameters: 4B
      context_window: "262,144 tokens natively and extensible up to 1,010,000 tokens"
      license: null
      summary: delivering exceptional utility and performance
      organization_evidence[6]: # Qwen3.5-4B,Citation,"@misc{qwen3.5,","    title  = {{Qwen3.5}: Towards Native Multimodal Agents},","    author = {{Qwen Team}},","    month  = {February},"
      model_evidence[3]: # Qwen3.5-4B,Model Overview,"- Number of Parameters: 4B"
      release_date_evidence[7]: # Qwen3.5-4B,Citation,"@misc{qwen3.5,","    title  = {{Qwen3.5}: Towards Native Multimodal Agents},","    author = {{Qwen Team}},","    month  = {February},","    year   = {2026},"
      parameters_evidence[3]: # Qwen3.5-4B,Model Overview,"- Number of Parameters: 4B"
      context_window_evidence[3]: # Qwen3.5-4B,Model Overview,"- Context Length: 262,144 natively and extensible up to 1,010,000 tokens."
      license_evidence: null
      summary_evidence[2]: # Qwen3.5-4B,"Over recent months, we have intensified our focus on developing foundation models that deliver exceptional utility and performance."
  - url: "https://huggingface.co/Qwen/Qwen3.6-35B-A3B?local-app=docker-model-runner"
    title: Qwen/Qwen3.6-35B-A3B · Hugging Face
    published_at: "2026-04-22T11:30:29"
    release:
      organization: Qwen
      model: Qwen3.6-35B-A3B
      release_date: 2026-04
      parameters: 35B in total and 3B activated
      context_window: "262,144 natively and extensible up to 1,010,000 tokens"
      license: null
      summary: "prioritizes stability and real-world utility, offering developers a more intuitive, responsive, and genuinely productive coding experience"
      organization_evidence[2]: # Qwen3.6-35B-A3B,"@misc{qwen36_35b_a3b, title = {{Qwen3.6-35B-A3B}: Agentic Coding Power, Now Open to All}, url = {https://qwen.ai/blog?id=qwen3.6-35b-a3b}, author = {{Qwen Team}}, month = {April}, year = {2026} }"
      model_evidence[2]: # Qwen3.6-35B-A3B,"Following the February release of the Qwen3.5 series, we're pleased to share the first open-weight variant of Qwen3.6."
      release_date_evidence[2]: # Qwen3.6-35B-A3B,"@misc{qwen36_35b_a3b, title = {{Qwen3.6-35B-A3B}: Agentic Coding Power, Now Open to All}, url = {https://qwen.ai/blog?id=qwen3.6-35b-a3b}, author = {{Qwen Team}}, month = {April}, year = {2026} }"
      parameters_evidence[3]: # Qwen3.6-35B-A3B,## Model Overview,"- Number of Parameters: 35B in total and 3B activated"
      context_window_evidence[3]: # Qwen3.6-35B-A3B,## Model Overview,"- Context Length: 262,144 natively and extensible up to 1,010,000 tokens."
      license_evidence: null
      summary_evidence[2]: # Qwen3.6-35B-A3B,"Following the February release of the Qwen3.5 series, we're pleased to share the first open-weight variant of Qwen3.6."
  - url: "https://huggingface.co/Qwen/Qwen3.5-397B-A17B/blame/refs%2Fpr%2F9/README.md"
    title: README.md · Qwen/Qwen3.5-397B-A17B at refs/pr/9
    published_at: null
    release:
      organization: Qwen
      model: Qwen3.5-397B-A17B
      release_date: null
      parameters: 397B in total and 17B activated
      context_window: "262,144 natively and extensible up to 1,010,000 tokens"
      license: apache-2.0
      summary: delivering exceptional utility and performance
      organization_evidence[2]: # Qwen3.5-397B-A17B,"Qwen3.5 features the following enhancement:"
      model_evidence[2]: # Qwen3.5-397B-A17B,"Qwen3.5 features the following enhancement:"
      release_date_evidence: null
      parameters_evidence[3]: # Qwen3.5-397B-A17B,## Model Overview,"- Number of Parameters: 397B in total and 17B activated"
      context_window_evidence[3]: # Qwen3.5-397B-A17B,## Model Overview,"- Context Length: 262,144 natively and extensible up to 1,010,000 tokens."
      license_evidence[2]: # Qwen3.5-397B-A17B,"license: apache-2.0"
      summary_evidence[2]: # Qwen3.5-397B-A17B,"Over recent months, we have intensified our focus on developing foundation models that deliver exceptional utility and performance."
notes[2,]{text}:
  Merged 8 queries (1299 total) → 1156 unique results
  "Search result set: rf8e01d178e7 (1156 rows), reference it as FROM rf8e01d178e7 in follow-up queries"

result set rc3e82e8cb59 · 4 rows

url	title	published_at	release
https://huggingface.co/Qwen/Qwen3.5-0.8B	Qwen/Qwen3.5-0.8B · Hugging Face	2026-03-09T14:18:49	{"organization": "Qwen Team", "model": "Qwen3.5", "release_date": "2026-02", "parameters": "0.8B", "context_window": "262,144", "license": null, "summary": "pr…
https://huggingface.co/Qwen/Qwen3.5-4B	Qwen/Qwen3.5-4B · Hugging Face	2026-03-09T14:18:49	{"organization": "Qwen Team", "model": "Qwen3.5-4B", "release_date": "2026-02", "parameters": "4B", "context_window": "262,144 tokens natively and extensible u…
https://huggingface.co/Qwen/Qwen3.6-35B-A3B?local-app=docker-model-runner	Qwen/Qwen3.6-35B-A3B · Hugging Face	2026-04-22T11:30:29	{"organization": "Qwen", "model": "Qwen3.6-35B-A3B", "release_date": "2026-04", "parameters": "35B in total and 3B activated", "context_window": "262,144 nativ…
https://huggingface.co/Qwen/Qwen3.5-397B-A17B/blame/refs%2Fpr%2F9/README.md	README.md · Qwen/Qwen3.5-397B-A17B at refs/pr/9	null	{"organization": "Qwen", "model": "Qwen3.5-397B-A17B", "release_date": null, "parameters": "397B in total and 17B activated", "context_window": "262,144 native…

Download all 4 rows · 7 KB

tool result (6660 chars)

result_set_id: r9dc0747a006
total_rows: 3
columns[4]: published_at,release,title,url
rows[3]:
  - url: "https://z.ai/blog/glm-5.1"
    title: "GLM-5.1: Towards Long-Horizon Tasks"
    published_at: null
    release:
      organization: null
      model: GLM-5.1
      release_date: 2026-04-07
      parameters: null
      context_window: null
      license: MIT License
      summary: "next-generation flagship model for agentic engineering, with significantly stronger coding capabilities than its predecessor"
      organization_evidence: null
      model_evidence[2]: "GLM-5.1: Towards Long-Horizon Tasks","GLM-5.1 is our next-generation flagship model for agentic engineering, with significantly stronger coding capabilities than its predecessor."
      release_date_evidence[2]: 2026-04-07 · Research,"GLM-5.1: Towards Long-Horizon Tasks"
      parameters_evidence: null
      context_window_evidence: null
      license_evidence[2]: "GLM-5.1: Towards Long-Horizon Tasks",GLM-5.1 is released as open source under the MIT License.
      summary_evidence[2]: "GLM-5.1: Towards Long-Horizon Tasks","GLM-5.1 is our next-generation flagship model for agentic engineering, with significantly stronger coding capabilities than its predecessor."
  - url: "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro?show_file_info=model.safetensors.index.json"
    title: deepseek-ai/DeepSeek-V4-Pro · Hugging Face
    published_at: "2026-04-24T02:53:41"
    release:
      organization: DeepSeek-AI
      model: DeepSeek-V4
      release_date: "2026"
      parameters: "1.6T total (49B activated) for DeepSeek-V4-Pro, 284B total (13B activated) for DeepSeek-V4-Flash"
      context_window: "1,000,000 tokens"
      license: MIT
      summary: " Mixture-of-Experts language models designed for efficient million-token context intelligence, coding, reasoning, and agentic tasks."
      organization_evidence[2]: "# DeepSeek-V4: Towards Highly Efficient Million-Token Context Intelligence","@misc{deepseekai2026deepseekv4, title={DeepSeek-V4: Towards Highly Efficient Million-Token Context Intelligence}, author={DeepSeek-AI}, year={2026}, }"
      model_evidence[2]: "# DeepSeek-V4: Towards Highly Efficient Million-Token Context Intelligence","We present a preview version of **DeepSeek-V4** series, including two strong Mixture-of-Experts (MoE) language models — **DeepSeek-V4-Pro** with 1.6T parameters (49B activated) and **DeepSeek-V4-Flash** with 284B parameters (13B activated) — both supporting a context length of **one million tokens**."
      release_date_evidence[2]: "# DeepSeek-V4: Towards Highly Efficient Million-Token Context Intelligence","@misc{deepseekai2026deepseekv4, title={DeepSeek-V4: Towards Highly Efficient Million-Token Context Intelligence}, author={DeepSeek-AI}, year={2026}, }"
      parameters_evidence[2]: "# DeepSeek-V4: Towards Highly Efficient Million-Token Context Intelligence","We present a preview version of **DeepSeek-V4** series, including two strong Mixture-of-Experts (MoE) language models — **DeepSeek-V4-Pro** with 1.6T parameters (49B activated) and **DeepSeek-V4-Flash** with 284B parameters (13B activated) — both supporting a context length of **one million tokens**."
      context_window_evidence[2]: "# DeepSeek-V4: Towards Highly Efficient Million-Token Context Intelligence","We present a preview version of **DeepSeek-V4** series, including two strong Mixture-of-Experts (MoE) language models — **DeepSeek-V4-Pro** with 1.6T parameters (49B activated) and **DeepSeek-V4-Flash** with 284B parameters (13B activated) — both supporting a context length of **one million tokens**."
      license_evidence[2]: "# DeepSeek-V4: Towards Highly Efficient Million-Token Context Intelligence","This repository and the model weights are licensed under the [MIT License](/deepseek-ai/DeepSeek-V4-Pro/blob/main/LICENSE)."
      summary_evidence[2]: "# DeepSeek-V4: Towards Highly Efficient Million-Token Context Intelligence","We present a preview version of **DeepSeek-V4** series, including two strong Mixture-of-Experts (MoE) language models — **DeepSeek-V4-Pro** with 1.6T parameters (49B activated) and **DeepSeek-V4-Flash** with 284B parameters (13B activated) — both supporting a context length of **one million tokens**."
  - url: "https://huggingface.co/zai-org/GLM-4.5/discussions/20/files"
    title: "zai-org/GLM-4.5 · Improve model card with detailed quick start, system requirements, and paper link clarification"
    published_at: "2025-08-11T13:14:01"
    release:
      organization: Zhipu AI
      model: GLM-4.5
      release_date: 2025-08-11
      parameters: 355B total parameters with 32B active parameters
      context_window: 128K
      license: MIT
      summary: foundation models designed for intelligent agents
      organization_evidence[2]: # GLM-4.5,"📖 Check out the GLM-4.5 <a href=\"https://z.ai/blog/glm-4.5\" target=\"_blank\">technical blog</a>, <a href=\"https://arxiv.org/abs/2508.06471\" target=\"_blank\">technical report</a>, and <a href=\"https://zhipu-ai.feishu.cn/wiki/Gv3swM0Yci7w7Zke9E0crhU7n7D\" target=\"_blank\">Zhipu AI technical documentation</a>."
      model_evidence[2]: # GLM-4.5,The **GLM-4.5** series models are foundation models designed for intelligent agents.
      release_date_evidence[2]: # GLM-4.5,"📖 Check out the GLM-4.5 <a href=\"https://z.ai/blog/glm-4.5\" target=\"_blank\">technical blog</a>, <a href=\"https://arxiv.org/abs/2508.06471\" target=\"_blank\">technical report</a>, and <a href=\"https://zhipu-ai.feishu.cn/wiki/Gv3swM0Yci7w7Zke9E0crhU7n7D\" target=\"_blank\">Zhipu AI technical documentation</a>."
      parameters_evidence[2]: # GLM-4.5,"GLM-4.5 has **355** billion total parameters with **32** billion active parameters, while GLM-4.5-Air adopts a more compact design with **106** billion total parameters and **12** billion active parameters."
      context_window_evidence[3]: # GLM-4.5,"Under the configurations in the table below, the models can utilize their full 128K context length:",| Model | Precision | GPU Type and Count | Test Framework |
      license_evidence[2]: # GLM-4.5,"We have open-sourced the base models, hybrid reasoning models, and FP8 versions of the hybrid reasoning models for both GLM-4.5 and GLM-4.5-Air. They are released under the MIT open-source license and can be used commercially and for secondary development."
      summary_evidence[2]: # GLM-4.5,The **GLM-4.5** series models are foundation models designed for intelligent agents.
notes[2,]{text}:
  Merged 8 queries (1279 total) → 1277 unique results
  "Search result set: rfd33f114311 (1277 rows), reference it as FROM rfd33f114311 in follow-up queries"

result set r9dc0747a006 · 3 rows

url	title	published_at	release
https://z.ai/blog/glm-5.1	GLM-5.1: Towards Long-Horizon Tasks	null	{"organization": null, "model": "GLM-5.1", "release_date": "2026-04-07", "parameters": null, "context_window": null, "license": "MIT License", "summary": "next…
https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro?show_file_info=model.safetensors.index.json	deepseek-ai/DeepSeek-V4-Pro · Hugging Face	2026-04-24T02:53:41	{"organization": "DeepSeek-AI", "model": "DeepSeek-V4", "release_date": "2026", "parameters": "1.6T total (49B activated) for DeepSeek-V4-Pro, 284B total (13B …
https://huggingface.co/zai-org/GLM-4.5/discussions/20/files	zai-org/GLM-4.5 · Improve model card with detailed quick start, system requirements, and paper link clarification	2025-08-11T13:14:01	{"organization": "Zhipu AI", "model": "GLM-4.5", "release_date": "2025-08-11", "parameters": "355B total parameters with 32B active parameters", "context_windo…

Download all 3 rows · 7 KB

tool result (7229 chars)

result_set_id: rdb4928b59ce
total_rows: 1
columns[4]: published_at,release,title,url
rows[1]:
  - url: "https://huggingface.co/moonshotai/Kimi-K2.5/blob/refs%2Fpr%2F66/README.md"
    title: README.md · moonshotai/Kimi-K2.5 at refs/pr/66
    published_at: "2026-02-05T03:49:51"
    release:
      organization: Moonshot AI
      model: Kimi K2.5
      release_date: 2026-02-02
      parameters: "Total Parameters: 1T, Activated Parameters: 32B"
      context_window: 256K
      license: Modified MIT
      summary: Native multimodal agentic model built through continual pretraining on mixed visual and text tokens
      organization_evidence[2]: ## 1. Model Introduction,"Kimi K2.5 is an open-source, native multimodal agentic model built through continual pretraining on approximately 15 trillion mixed visual and text tokens atop Kimi-K2-Base."
      model_evidence[2]: ## 1. Model Introduction,"Kimi K2.5 is an open-source, native multimodal agentic model built through continual pretraining on approximately 15 trillion mixed visual and text tokens atop Kimi-K2-Base."
      release_date_evidence[3]: ```," tags: - compressed-tensors license: other license_name: modified-mit library_name: transformers pipeline_tag: image-text-to-text paper: arxiv.org/abs/2602.02276","@misc{kimiteam2026kimik25visualagentic, title={Kimi K2.5: Visual Agentic Intelligence}, author={Kimi Team and Tongtong Bai and Yifan Bai and Yiping Bao and S. H. Cai and Yuan Cao and Y. Charles and H. S. Che and Cheng Chen and Guanduo Chen and Huarong Chen and Jia Chen and Jiahao Chen and Jianlong Chen and Jun Chen and Kefan Chen and Liang Chen and Ruijue Chen and Xinhao Chen and Yanru Chen and Yanxu Chen and Yicun Chen and Yimin Chen and Yingjiang Chen and Yuankun Chen and Yujie Chen and Yutian Chen and Zhirong Chen and Ziwei Chen and Dazhi Cheng and Minghan Chu and Jialei Cui and Jiaqi Deng and Muxi Diao and Hao Ding and Mengfan Dong and Mengnan Dong and Yuxin Dong and Yuhao Dong and Angang Du and Chenzhuang Du and Dikang Du and Lingxiao Du and Yulun Du and Yu Fan and Shengjun Fang and Qiulin Feng and Yichen Feng and Garimugai Fu and Kelin Fu and Hongcheng Gao and Tong Gao and Yuyao Ge and Shangyi Geng and Chengyang Gong and Xiaochen Gong and Zhuoma Gongque and Qizheng Gu and Xinran Gu and Yicheng Gu and Longyu Guan and Yuanying Guo and Xiaoru Hao and Weiran He and Wenyang He and Yunjia He and Chao Hong and Hao Hu and Jiaxi Hu and Yangyang Hu and Zhenxing Hu and Ke Huang and Ruiyuan Huang and Weixiao Huang and Zhiqi Huang and Tao Jiang and Zhejun Jiang and Xinyi Jin and Yu Jing and Guokun Lai and Aidi Li and C. Li and Cheng Li and Fang Li and Guanghe Li and Guanyu Li and Haitao Li and Haoyang Li and Jia Li and Jingwei Li and Junxiong Li and Lincan Li and Mo Li and Weihong Li and Wentao Li and Xinhang Li and Xinhao Li and Yang Li and Yanhao Li and Yiwei Li and Yuxiao Li and Zhaowei Li and Zheming Li and Weilong Liao and Jiawei Lin and Xiaohan Lin and Zhishan Lin and Zichao Lin and Cheng Liu and Chenyu Liu and Hongzhang Liu and Liang Liu and Shaowei Liu and Shudong Liu and Shuran Liu and Tianwei Liu and Tianyu Liu and Weizhou Liu and Xiangyan Liu and Yangyang Liu and Yanming Liu and Yibo Liu and Yuanxin Liu and Yue Liu and Zhengying Liu and Zhongnuo Liu and Enzhe Lu and Haoyu Lu and Zhiyuan Lu and Junyu Luo and Tongxu Luo and Yashuo Luo and Long Ma and Yingwei Ma and Shaoguang Mao and Yuan Mei and Xin Men and Fanqing Meng and Zhiyong Meng and Yibo Miao and Minqing Ni and Kun Ouyang and Siyuan Pan and Bo Pang and Yuchao Qian and Ruoyu Qin and Zeyu Qin and Jiezhong Qiu and Bowen Qu and Zeyu Shang and Youbo Shao and Tianxiao Shen and Zhennan Shen and Juanfeng Shi and Lidong Shi and Shengyuan Shi and Feifan Song and Pengwei Song and Tianhui Song and Xiaoxi Song and Hongjin Su and Jianlin Su and Zhaochen Su and Lin Sui and Jinsong Sun and Junyao Sun and Tongyu Sun and Flood Sung and Yunpeng Tai and Chuning Tang and Heyi Tang and Xiaojuan Tang and Zhengyang Tang and Jiawen Tao and Shiyuan Teng and Chaoran Tian and Pengfei Tian and Ao Wang and Bowen Wang and Chensi Wang and Chuang Wang and Congcong Wang and Dingkun Wang and Dinglu Wang and Dongliang Wang and Feng Wang and Hailong Wang and Haiming Wang and Hengzhi Wang and Huaqing Wang and Hui Wang and Jiahao Wang and Jinhong Wang and Jiuzheng Wang and Kaixin Wang and Linian Wang and Qibin Wang and Shengjie Wang and Shuyi Wang and Si Wang and Wei Wang and Xiaochen Wang and Xinyuan Wang and Yao Wang and Yejie Wang and Yipu Wang and Yiqin Wang and Yucheng Wang and Yuzhi Wang and Zhaoji Wang and Zhaowei Wang and Zhengtao Wang and Zhexu Wang and Zihan Wang and Zizhe Wang and Chu Wei and Ming Wei and Chuan Wen and Zichen Wen and Chengjie Wu and Haoning Wu and Junyan Wu and Rucong Wu and Wenhao Wu and Yuefeng Wu and Yuhao Wu and Yuxin Wu and Zijian Wu and Chenjun Xiao and Jin Xie and Xiaotong Xie and Yuchong Xie and Yifei Xin and Bowei Xing and Boyu Xu and Jianfan Xu and Jing Xu and Jinjing Xu and L. H. Xu and Lin Xu and Suting Xu and Weixin Xu and Xinbo Xu and Xinran Xu and Yangchuan Xu and Yichang Xu and Yuemeng Xu and Zelai Xu and Ziyao Xu and Junjie Yan and Yuzi Yan and Guangyao Yang and Hao Yang and Junwei Yang and Kai Yang and Ningyuan Yang and Ruihan Yang and Xiaofei Yang and Xinlong Yang and Ying Yang and Yi Yang and Yi Yang and Zhen Yang and Zhilin Yang and Zonghan Yang and Haotian Yao and Dan Ye and Wenjie Ye and Zhuorui Ye and Bohong Yin and Chengzhen Yu and Longhui Yu and Tao Yu and Tianxiang Yu and Enming Yuan and Mengjie Yuan and Xiaokun Yuan and Yang Yue and Weihao Zeng and Dunyuan Zha and Haobing Zhan and Dehao Zhang and Hao Zhang and Jin Zhang and Puqi Zhang and Qiao Zhang and Rui Zhang and Xiaobin Zhang and Y. Zhang and Yadong Zhang and Yangkun Zhang and Yichi Zhang and Yizhi Zhang and Yongting Zhang and Yu Zhang and Yushun Zhang and Yutao Zhang and Yutong Zhang and Zheng Zhang and Chenguang Zhao and Feifan Zhao and Jinxiang Zhao and Shuai Zhao and Xiangyu Zhao and Yikai Zhao and Zijia Zhao and Huabin Zheng and Ruihan Zheng and Shaojie Zheng and Tengyang Zheng and Junfeng Zhong and Longguang Zhong and Weiming Zhong and M. Zhou and Runjie Zhou and Xinyu Zhou and Zaida Zhou and Jinguo Zhu and Liya Zhu and Xinhao Zhu and Yuxuan Zhu and Zhen Zhu and Jingze Zhuang and Weiyu Zhuang and Ying Zou and Xinxing Zu}, year={2026}, eprint={2602.02276}, archivePrefix={arXiv}, primaryClass={cs.CL}, url={https://arxiv.org/abs/2602.02276}, }"
      parameters_evidence[3]: ## 2. Model Summary,| **Total Parameters** | 1T |,| **Activated Parameters** | 32B |
      context_window_evidence[2]: ## 2. Model Summary,| **Context Length** | 256K |
      license_evidence[2]: ## 7. License,"Both the code repository and the model weights are released under the [Modified MIT License](/moonshotai/Kimi-K2.5/blob/main/LICENSE)."
      summary_evidence[2]: ## 1. Model Introduction,"Kimi K2.5 is an open-source, native multimodal agentic model built through continual pretraining on approximately 15 trillion mixed visual and text tokens atop Kimi-K2-Base."
notes[2,]{text}:
  Merged 8 queries (1250 total) → 1098 unique results
  "Search result set: r3edf476ed5b (1098 rows), reference it as FROM r3edf476ed5b in follow-up queries"

result set rdb4928b59ce · 1 rows

url	title	published_at	release
https://huggingface.co/moonshotai/Kimi-K2.5/blob/refs%2Fpr%2F66/README.md	README.md · moonshotai/Kimi-K2.5 at refs/pr/66	2026-02-05T03:49:51	{"organization": "Moonshot AI", "model": "Kimi K2.5", "release_date": "2026-02-02", "parameters": "Total Parameters: 1T, Activated Parameters: 32B", "context_w…

Download all 1 rows · 7 KB

tool result (11581 chars)

result_set_id: r7d093d4eb4c
total_rows: 6
columns[4]: published_at,release,title,url
rows[6]:
  - url: "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16"
    title: nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16 · Hugging Face
    published_at: "2026-08-14T21:27:48"
    release:
      organization: NVIDIA
      model: NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16
      release_date: 2026-06-04
      parameters: 550B (55B active)
      context_window: 1M tokens
      license: OpenMDW-1.1
      summary: "Frontier-scale general purpose reasoning and chat model optimized for complex agentic workflows, long-context reasoning, and high-stakes analytical workloads."
      organization_evidence[3]: # NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16,## Model Overview,"**Model Developer:** NVIDIA Corporation"
      model_evidence[1]: # NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16
      release_date_evidence[3]: # NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16,## Model Summary,"| **Release Date** | June 4, 2026 |"
      parameters_evidence[3]: # NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16,## Model Summary,| **Total Parameters** | 550B (55B active) |
      context_window_evidence[3]: # NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16,## Model Summary,| **Context Length** | Up to 1M tokens |
      license_evidence[3]: # NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16,## Model Summary,"| **License** | [OpenMDW License Agreement, version 1.1](https://raw.githubusercontent.com/OpenMDW/OpenMDW/refs/heads/main/1.1/LICENSE.OpenMDW-1.1) |"
      summary_evidence[3]: # NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16,### Use Case,"NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16 is a frontier-scale general purpose reasoning and chat model intended to be used in English, Code, and supported multilingual contexts."
  - url: "https://huggingface.co/google/gemma-4-31B-it-qat-q4_0-unquantized"
    title: google/gemma-4-31B-it-qat-q4_0-unquantized · Hugging Face
    published_at: "2026-07-21T21:12:15"
    release:
      organization: Google DeepMind
      model: Gemma 4
      release_date: 2026-07-02
      parameters: null
      context_window: 256K tokens
      license: Apache 2.0
      summary: "multimodal, handling text and image input (with audio supported on E2B, E4B, and 12B) and generating text output"
      organization_evidence[1]: Gemma is a family of open models built by Google DeepMind.
      model_evidence[1]: Gemma is a family of open models built by Google DeepMind.
      release_date_evidence[2]: Gemma is a family of open models built by Google DeepMind.,"[Technical Report](https://arxiv.org/abs/2607.02770)"
      parameters_evidence: null
      context_window_evidence[2]: Gemma is a family of open models built by Google DeepMind.,Gemma 4 features a context window of up to 256K tokens and maintains multilingual support in over 140 languages.
      license_evidence[2]: Gemma is a family of open models built by Google DeepMind.,"**License**: [Apache 2.0](https://ai.google.dev/gemma/docs/gemma_4_license) | **Authors**: [Google DeepMind](https://deepmind.google/models/gemma/)"
      summary_evidence[2]: Gemma is a family of open models built by Google DeepMind.,"Gemma 4 models are multimodal, handling text and image input (with audio supported on E2B, E4B, and 12B) and generating text output."
  - url: "https://developer.nvidia.com/blog/post-train-nvidia-cosmos-3-edge-for-on-device-robot-control"
    title: Post-Train NVIDIA Cosmos 3 Edge for On-Device Robot Control | NVIDIA Technical Blog
    published_at: "2026-08-19T16:00:00"
    release:
      organization: NVIDIA
      model: Cosmos 3 Edge
      release_date: 2026-08-19
      parameters: 4B
      context_window: null
      license: OpenMDW1.1
      summary: on-device robot manipulation policies and physical AI control
      organization_evidence[1]: Post-Train NVIDIA Cosmos 3 Edge for On-Device Robot Control
      model_evidence[1]: Post-Train NVIDIA Cosmos 3 Edge for On-Device Robot Control
      release_date_evidence[2]: Post-Train NVIDIA Cosmos 3 Edge for On-Device Robot Control,"Aug 19, 2026"
      parameters_evidence[1]: Post-Train NVIDIA Cosmos 3 Edge for On-Device Robot Control
      context_window_evidence: null
      license_evidence[2]: Post-Train NVIDIA Cosmos 3 Edge for On-Device Robot Control,"More broadly, Cosmos’s open weights and framework, with the OpenMDW1.1 license, make post-training a powerful, flexible, and easy way to create specialized, highly performant, and accurate custom models for physical AI."
      summary_evidence[2]: Post-Train NVIDIA Cosmos 3 Edge for On-Device Robot Control,"- [NVIDIA Jetson Thor](https://www.nvidia.com/en-us/autonomous-machines/embedded-systems/jetson-thor/) can run the 4B Cosmos 3 Edge omni-model natively for on-device robot manipulation policies."
  - url: "https://huggingface.co/nvidia/DeepSeek-V4-Flash-0731-NVFP4"
    title: nvidia/DeepSeek-V4-Flash-0731-NVFP4 · Hugging Face
    published_at: "2026-08-31T16:53:50"
    release:
      organization: NVIDIA
      model: DeepSeek-V4-Flash-0731-NVFP4
      release_date: 2026-08-31
      parameters: 304B in total and 13B activated
      context_window: 1 million tokens
      license: null
      summary: "well-suited for advanced reasoning, agentic AI applications, tool use scenarios, and complex problem-solving in domains such as mathematics, software engineering, and enterprise AI assistants"
      organization_evidence[3]: # Model Overview,"## Description:","The NVIDIA DeepSeek-V4-Flash-0731-NVFP4 model is the quantized version of DeepSeek AI's DeepSeek-V4-Flash-0731 model, an autoregressive Mixture-of-Experts language model that uses an optimized Transformer architecture with hybrid attention (Compressed Sparse Attention and Heavily Compressed Attention) and Manifold-Constrained Hyper-Connections."
      model_evidence[3]: # Model Overview,"## Description:","The NVIDIA DeepSeek-V4-Flash-0731-NVFP4 model is the quantized version of DeepSeek AI's DeepSeek-V4-Flash-0731 model, an autoregressive Mixture-of-Experts language model that uses an optimized Transformer architecture with hybrid attention (Compressed Sparse Attention and Heavily Compressed Attention) and Manifold-Constrained Hyper-Connections."
      release_date_evidence[3]: # Model Overview,"## Release Date:","Hugging Face 08/31/2026 via [https://huggingface.co/nvidia/DeepSeek-V4-Flash-0731-NVFP4](https://huggingface.co/nvidia/DeepSeek-V4-Flash-0731-NVFP4)"
      parameters_evidence[3]: # Model Overview,"## Model Architecture:","**Number of Model Parameters:** 304B in total and 13B activated"
      context_window_evidence[3]: # Model Overview,"## Input:",Maximum context length of 1 million tokens.
      license_evidence: null
      summary_evidence[3]: # Model Overview,"### Use Case:","DeepSeek V4 is well-suited for advanced reasoning, agentic AI applications, tool use scenarios, and complex problem-solving in domains such as mathematics, software engineering, and enterprise AI assistants."
  - url: "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-NVFP4"
    title: nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-NVFP4 · Hugging Face
    published_at: "2026-08-17T15:06:50"
    release:
      organization: NVIDIA
      model: Nemotron-3-Super-120B-A12B-NVFP4
      release_date: 2026-03-11
      parameters: 120B Total / 12B Active
      context_window: 1M tokens
      license: NVIDIA Nemotron Open Model License
      summary: general purpose reasoning and chat model optimized for collaborative agents and high-volume workloads
      organization_evidence[3]: # NVIDIA-Nemotron-3-Super-120B-A12B-NVFP4,## Description,"**Nemotron-3-Super-120B-A12B-NVFP4** is a large language model (LLM) trained by NVIDIA, designed to deliver strong agentic, reasoning, and conversational capabilities."
      model_evidence[3]: # NVIDIA-Nemotron-3-Super-120B-A12B-NVFP4,## Description,"**Nemotron-3-Super-120B-A12B-NVFP4** is a large language model (LLM) trained by NVIDIA, designed to deliver strong agentic, reasoning, and conversational capabilities."
      release_date_evidence[3]: # NVIDIA-Nemotron-3-Super-120B-A12B-NVFP4,## Model Summary,"| **Release Date** | March 11, 2026 |"
      parameters_evidence[3]: # NVIDIA-Nemotron-3-Super-120B-A12B-NVFP4,## Model Architecture,"- **Number of model parameters:** 120B Total / 12B Active"
      context_window_evidence[3]: # NVIDIA-Nemotron-3-Super-120B-A12B-NVFP4,## Model Summary,| **Context Length** | Up to 1M tokens |
      license_evidence[3]: # NVIDIA-Nemotron-3-Super-120B-A12B-NVFP4,## Model Summary,"| **License** | [NVIDIA Nemotron Open Model License](https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-nemotron-open-model-license/) |"
      summary_evidence[3]: # NVIDIA-Nemotron-3-Super-120B-A12B-NVFP4,### Use Case,"NVIDIA-Nemotron-3-Super-120B-A12B-NVFP4 is a general purpose reasoning and chat model intended to be used in English, Code, and supported multilingual contexts."
  - url: "https://build.nvidia.com/nvidia/nemotron-3-super-120b-a12b/modelcard?:~:text=Description_and%20120B%20parameters%20in%20total.&text=This%20model%20is%20ready%20for%20commercial%20use."
    title: nemotron-3-super-120b-a12b Model by NVIDIA | NVIDIA NIM
    published_at: null
    release:
      organization: NVIDIA
      model: Nemotron-3-Super-120B-A12B
      release_date: 2026-03-11
      parameters: 120B (12B active)
      context_window: 1M tokens
      license: NVIDIA Nemotron Open Model License
      summary: general purpose reasoning and chat model optimized for collaborative agents and high-volume workloads
      organization_evidence[1]: nemotron-3-super-120b-a12b Model by NVIDIA | NVIDIA NIM NVIDIA-Nemotron-3-Super-120B-A12B Model Summary
      model_evidence[2]: nemotron-3-super-120b-a12b Model by NVIDIA | NVIDIA NIM NVIDIA-Nemotron-3-Super-120B-A12B Model Summary,"NVIDIA-Nemotron-3-Super-120B-A12B Model Summary   Total Parameters 120B (12B active)  Architecture LatentMoE - Mamba-2 + MoE + Attention hybrid with Multi-Token Prediction (MTP)  Context Length Up to 1M tokens  Minimum GPU Requirement 8× H100-80GB  Supported Languages English, French, German, Italian, Japanese, Spanish, Chinese  Best For Agentic workflows, long-context reasoning, high-volume workloads (e.g. IT ticket automation), tool use, RAG  Reasoning Mode Configurable on/off via chat template (enable_thinking=True/False)  License NVIDIA Nemotron Open Model License  Release Date March 11, 2026"
      release_date_evidence[2]: nemotron-3-super-120b-a12b Model by NVIDIA | NVIDIA NIM NVIDIA-Nemotron-3-Super-120B-A12B Model Summary,"Release Date March 11, 2026"
      parameters_evidence[2]: nemotron-3-super-120b-a12b Model by NVIDIA | NVIDIA NIM NVIDIA-Nemotron-3-Super-120B-A12B Model Summary,Total Parameters 120B (12B active)
      context_window_evidence[2]: nemotron-3-super-120b-a12b Model by NVIDIA | NVIDIA NIM NVIDIA-Nemotron-3-Super-120B-A12B Model Summary,Context Length Up to 1M tokens
      license_evidence[2]: nemotron-3-super-120b-a12b Model by NVIDIA | NVIDIA NIM NVIDIA-Nemotron-3-Super-120B-A12B Model Summary,License NVIDIA Nemotron Open Model License
      summary_evidence[2]: nemotron-3-super-120b-a12b Model by NVIDIA | NVIDIA NIM NVIDIA-Nemotron-3-Super-120B-A12B Model Summary,"NVIDIA-Nemotron-3-Super-120B-A12B-BF16 is a general purpose reasoning and chat model intended to be used in English, Code, and supported multilingual contexts."
notes[2,]{text}:
  Merged 8 queries (1528 total) → 1524 unique results
  "Search result set: rbe28438d028 (1524 rows), reference it as FROM rbe28438d028 in follow-up queries"

result set r7d093d4eb4c · 6 rows

url	title	published_at	release
https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16	nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16 · Hugging Face	2026-08-14T21:27:48	{"organization": "NVIDIA", "model": "NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16", "release_date": "2026-06-04", "parameters": "550B (55B active)", "context_window"…
https://huggingface.co/google/gemma-4-31B-it-qat-q4_0-unquantized	google/gemma-4-31B-it-qat-q4_0-unquantized · Hugging Face	2026-07-21T21:12:15	{"organization": "Google DeepMind", "model": "Gemma 4", "release_date": "2026-07-02", "parameters": null, "context_window": "256K tokens", "license": "Apache 2…
https://developer.nvidia.com/blog/post-train-nvidia-cosmos-3-edge-for-on-device-robot-control	Post-Train NVIDIA Cosmos 3 Edge for On-Device Robot Control \| NVIDIA Technical Blog	2026-08-19T16:00:00	{"organization": "NVIDIA", "model": "Cosmos 3 Edge", "release_date": "2026-08-19", "parameters": "4B", "context_window": null, "license": "OpenMDW1.1", "summar…
https://huggingface.co/nvidia/DeepSeek-V4-Flash-0731-NVFP4	nvidia/DeepSeek-V4-Flash-0731-NVFP4 · Hugging Face	2026-08-31T16:53:50	{"organization": "NVIDIA", "model": "DeepSeek-V4-Flash-0731-NVFP4", "release_date": "2026-08-31", "parameters": "304B in total and 13B activated", "context_win…
https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-NVFP4	nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-NVFP4 · Hugging Face	2026-08-17T15:06:50	{"organization": "NVIDIA", "model": "Nemotron-3-Super-120B-A12B-NVFP4", "release_date": "2026-03-11", "parameters": "120B Total / 12B Active", "context_window"…

Download all 6 rows · 12 KB

assistant

tool call · select

WITH candidate AS (
  SELECT
    release_date,
    model,
    organization,
    parameters,
    context_window,
    license,
    significance,
    source_count,
    example_url AS source_url,
    'multi-source synthesis' AS source_basis
  FROM r8f4c29f3a52
  WHERE model IN (
    'GLM-5',
    'Qwen 3.5',
    'Nemotron 3 Super',
    'Mistral Small 4',
    'MiniMax M2.7',
    'Qwen 3.6',
    'Kimi K2.6',
    'Mistral Medium 3.5',
    'MiniMax M3',
    'GLM 5.2',
    'Inkling',
    'Kimi K3',
    'Muse Glimmer',
    'Qwen3.8-27B',
    'GLM-5.3-Flash'
  )
),
official_pages AS (
  SELECT
    release_date,
    model,
    organization,
    parameters,
    context_window,
    license,
    summary AS significance,
    1 AS source_count,
    url AS source_url,
    'first-party model card or announcement' AS source_basis
  FROM (
    SELECT
      url,
      UNNEST(release, recursive := true)
    FROM rdb4928b59ce
  )
  WHERE model = 'Kimi K2.5'
  UNION ALL
  SELECT
    release_date,
    model,
    COALESCE(organization, 'Z.ai') AS organization,
    parameters,
    context_window,
    license,
    summary AS significance,
    1 AS source_count,
    url AS source_url,
    'first-party model card or announcement' AS source_basis
  FROM (
    SELECT
      url,
      UNNEST(release, recursive := true)
    FROM r9dc0747a006
  )
  WHERE model IN ('GLM-5.1', 'DeepSeek-V4')
  UNION ALL
  SELECT
    release_date,
    CASE
      WHEN model LIKE 'NVIDIA-Nemotron-3-Ultra%' THEN 'Nemotron 3 Ultra'
      ELSE model
    END AS model,
    organization,
    parameters,
    context_window,
    license,
    summary AS significance,
    1 AS source_count,
    url AS source_url,
    'first-party model card or announcement' AS source_basis
  FROM (
    SELECT
      url,
      UNNEST(release, recursive := true)
    FROM r7d093d4eb4c
  )
  WHERE model IN (
    'NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16',
    'Gemma 4'
  )
)
SELECT
  release_date,
  SUBSTR(release_date, 1, 7) AS release_month,
  organization,
  model,
  parameters,
  context_window,
  license,
  CASE
    WHEN LOWER(COALESCE(license, '')) LIKE '%modified%'
      OR LOWER(COALESCE(license, '')) LIKE '%openmdw%'
      OR LOWER(COALESCE(license, '')) LIKE '%community%'
      OR LOWER(COALESCE(license, '')) LIKE '%nemotron%'
      OR LOWER(COALESCE(license, '')) LIKE '%kimi%'
      THEN 'open weights / custom terms'
    WHEN LOWER(COALESCE(license, '')) LIKE '%apache%'
      OR LOWER(COALESCE(license, '')) = 'mit'
      OR LOWER(COALESCE(license, '')) = 'mit license'
      THEN 'permissive open-source license'
    ELSE 'open model; verify terms'
  END AS openness_class,
  significance,
  source_count,
  source_basis,
  source_url
FROM (
  SELECT * FROM candidate
  UNION ALL
  SELECT * FROM official_pages
)
WHERE release_date LIKE '2026%'
ORDER BY
  CASE
    WHEN LENGTH(release_date) = 10 THEN TRY_CAST(release_date AS DATE)
    WHEN LENGTH(release_date) = 7 THEN TRY_CAST(release_date || '-01' AS DATE)
    ELSE TRY_CAST(release_date || '-01-01' AS DATE)
  END,
  organization,
  model

{
 "show_preview": false
}

tool result (7168 chars)

result_set_id: r7d085f93d9d
total_rows: 20
columns[12]: context_window,license,model,openness_class,organization,parameters,release_date,release_month,significance,source_basis,source_count,source_url
rows[20,]{release_date,release_month,organization,model,parameters,context_window,license,openness_class,significance,source_count,source_basis,source_url}:
  "2026","2026",DeepSeek-AI,DeepSeek-V4,"1.6T total (49B activated) for DeepSeek-V4-Pro, 284B total (13B activated) for DeepSeek-V4-Flash","1,000,000 tokens",MIT,permissive open-source license," Mixture-of-Experts language models designed for efficient million-token context intelligence, coding, reasoning, and agentic tasks.",1,first-party model card or announcement,"https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro?show_file_info=model.safetensors.index.json"
  2026-02-02,2026-02,Moonshot AI,Kimi K2.5,"Total Parameters: 1T, Activated Parameters: 32B",256K,Modified MIT,open weights / custom terms,Native multimodal agentic model built through continual pretraining on mixed visual and text tokens,1,first-party model card or announcement,"https://huggingface.co/moonshotai/Kimi-K2.5/blob/refs%2Fpr%2F66/README.md"
  2026-02-13,2026-02,Zhipu AI,GLM-5,744B (44B active),1 million-token,MIT,permissive open-source license,"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",10,multi-source synthesis,"https://z.ai/blog/glm-5"
  2026-02-16,2026-02,Alibaba,Qwen 3.5,17B active / 397B total,256K,Apache 2.0,permissive open-source license,Qwen3.5 (397B total / 17B active MoE); 256K native context; 256 experts,4,multi-source synthesis,"https://jaredwatkins.com/research/local-llm-finetuning/fundamentals/base-models"
  2026-03-11,2026-03,NVIDIA,Nemotron 3 Super,120.6B,Up to 1M,Open weights,open model; verify terms,36.0 on the Artificial Analysis Intelligence Index,4,multi-source synthesis,"https://tech-insider.org/ca/nvidia-nemotron-3-ultra-2026"
  2026-03-16,2026-03,Mistral AI,Mistral Small 4,119B / 6B,"256,000",Apache 2.0,permissive open-source license,"unifies reasoning, multimodal understanding, and agentic coding in one model",11,multi-source synthesis,"https://hidekazu-konishi.com/entry/open_weights_llm_release_history_and_timeline.html"
  2026-04-07,2026-04,Z.ai,GLM-5.1,null,null,MIT License,permissive open-source license,"next-generation flagship model for agentic engineering, with significantly stronger coding capabilities than its predecessor",1,first-party model card or announcement,"https://z.ai/blog/glm-5.1"
  2026-04-12,2026-04,MiniMax,MiniMax M2.7,"229B total parameters, 10B active per token",200K tokens,Modified MIT,open weights / custom terms,"Hits 56.22% on SWE-Bench Pro and 57.0% on Terminal Bench 2, with downloads already in the tens of thousands.",6,multi-source synthesis,"https://ai-tldr.dev/releases/minimax-m2-7-open-weights"
  2026-04-16,2026-04,Alibaba,Qwen 3.6,35B-A3B and 27B,1M,Apache 2.0,permissive open-source license,"building on Qwen3.5 while prioritizing stability, real-world coding utility, and direct community feedback",5,multi-source synthesis,"https://hidekazu-konishi.com/entry/open_weights_llm_release_history_and_timeline.html"
  2026-04-20,2026-04,Moonshot AI,Kimi K2.6,1T total / 32B active (MoE),256K,Modified MIT,open weights / custom terms,"Native multimodal agentic model. MIT-equivalent below 100M MAU / $20M monthly revenue, attribution clause above.",7,multi-source synthesis,"https://fazm.ai/t/latest-open-source-llm-releases-2026"
  2026-04-29,2026-04,Mistral,Mistral Medium 3.5,128B,256K,modified MIT license,open weights / custom terms,"merges instruction-following, reasoning, and coding in a single set of weights",10,multi-source synthesis,"https://hidekazu-konishi.com/entry/open_weights_llm_release_history_and_timeline.html"
  2026-06-01,2026-06,MiniMax,MiniMax M3,Undisclosed (MoE),1M,MiniMax Community License,open weights / custom terms,"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.",16,multi-source synthesis,"https://fazm.ai/t/latest-open-source-llm-releases-2026"
  2026-06-04,2026-06,NVIDIA,Nemotron 3 Ultra,550B (55B active),1M tokens,OpenMDW-1.1,open weights / custom terms,"Frontier-scale general purpose reasoning and chat model optimized for complex agentic workflows, long-context reasoning, and high-stakes analytical workloads.",1,first-party model card or announcement,"https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16"
  2026-06-13,2026-06,Zhipu AI,GLM 5.2,744B (40B active),1 million tokens,MIT,permissive open-source license,Successor to GLM-5.1 (vendor-reported 58.4% SWE-Bench Pro). Positioned for long-horizon agentic engineering.,20,multi-source synthesis,"https://fazm.ai/t/latest-open-source-llm-releases-2026"
  2026-07-02,2026-07,Google DeepMind,Gemma 4,null,256K tokens,Apache 2.0,permissive open-source license,"multimodal, handling text and image input (with audio supported on E2B, E4B, and 12B) and generating text output",1,first-party model card or announcement,"https://huggingface.co/google/gemma-4-31B-it-qat-q4_0-unquantized"
  2026-07-15,2026-07,Thinking Machines Lab,Inkling,975B-parameter,null,Apache 2.0,permissive open-source license,"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",3,multi-source synthesis,"https://best-ai.news/ai-model-releases-2026"
  2026-07-27,2026-07,Moonshot AI,Kimi K3,2.8-trillion-parameter,1-million-token,Kimi K3 License,open weights / custom terms,"matched or surpassed Anthropic's Fable 5 on key benchmarks at roughly half the cost, erasing the closed-source lead",28,multi-source synthesis,"https://edgen.tech/tr/news/post/kimi-k3-open-weight-release-draws-100k-downloads-erases-us-ai-lead"
  2026-08-10,2026-08,Meta,Muse Glimmer,30-billion-parameter,128K,Apache 2.0,permissive open-source license,a 30-billion-parameter model distilled from Muse Spark,4,multi-source synthesis,"https://absolutedigitalpublishers.com/articles/llama-mistral-deepseek-qwen-gemma-and-olmo-what-open-actually-means"
  2026-08-14,2026-08,Alibaba,Qwen3.8-27B,27B,262K,Apache 2.0,permissive open-source license,"Artificial Analysis independently scores it 52 on its Intelligence Index — 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–150B range.",2,multi-source synthesis,"https://codersera.com/blog/open-source-llms-landscape-2026"
  2026-08-26,2026-08,Z.ai,GLM-5.3-Flash,320-billion-parameter,above one million tokens,MIT license,permissive open-source license,"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",2,multi-source synthesis,"https://startupfortune.com/zhipus-glm-53-flash-undercuts-claude-and-gpt-on-price-not-on-hardware"
notes[0]:

result set r7d085f93d9d · 20 rows

release_date	release_month	organization	model	parameters	context_window	license	openness_class	significance	source_count	source_basis	source_url
2026	2026	DeepSeek-AI	DeepSeek-V4	1.6T total (49B activated) for DeepSeek-V4-Pro, 284B total (13B activated) for DeepSeek-V4-Flash	1,000,000 tokens	MIT	permissive open-source license	Mixture-of-Experts language models designed for efficient million-token context intelligence, coding, reasoning, and agentic tasks.	1	first-party model card or announcement	https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro?show_file_info=model.safetensors.index.json
2026-02-02	2026-02	Moonshot AI	Kimi K2.5	Total Parameters: 1T, Activated Parameters: 32B	256K	Modified MIT	open weights / custom terms	Native multimodal agentic model built through continual pretraining on mixed visual and text tokens	1	first-party model card or announcement	https://huggingface.co/moonshotai/Kimi-K2.5/blob/refs%2Fpr%2F66/README.md
2026-02-13	2026-02	Zhipu AI	GLM-5	744B (44B active)	1 million-token	MIT	permissive open-source license	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 age…	10	multi-source synthesis	https://z.ai/blog/glm-5
2026-02-16	2026-02	Alibaba	Qwen 3.5	17B active / 397B total	256K	Apache 2.0	permissive open-source license	Qwen3.5 (397B total / 17B active MoE); 256K native context; 256 experts	4	multi-source synthesis	https://jaredwatkins.com/research/local-llm-finetuning/fundamentals/base-models
2026-03-11	2026-03	NVIDIA	Nemotron 3 Super	120.6B	Up to 1M	Open weights	open model; verify terms	36.0 on the Artificial Analysis Intelligence Index	4	multi-source synthesis	https://tech-insider.org/ca/nvidia-nemotron-3-ultra-2026

Download all 20 rows · 12 KB

assistant

tool call · generate_html_report

{"brief":"The user asked verbatim: “assemble a release time line of major open source LLMs being released in 2026”. Build a beautiful, highly legible release timeline current through 2026-08-31. Lead with one large horizontal chronological timeline, grouped by month, with one event per model-family launch. Use r7d085f93d9d as the curated timeline: 20 major releases through August 31, with release date/month, developer, model, parameters, context window, license/openness classification, significance, evidence basis, source count, and a linkable source URL. Emphasize the acceleration from February into April and the concentration on multimodality, agentic coding, mixture-of-experts architectures, and million-token context. Make the crucial distinction between permissively licensed open-source releases (MIT/Apache 2.0) and open-weight models under custom or restricted terms; do not imply that all downloadable-weight models satisfy the OSI definition of open source. Put exact-date events at exact positions; DeepSeek V4 has only year-level date in the shaped row, but its linked first-party Hugging Face model card was published 2026-04-24, so display April 24 with a small note that the date is inferred from the first-party card publication. Month-only/less precise claims should be visually marked as such. Prefer first-party source rows where available; when source_basis is multi-source synthesis, retain source_count and avoid presenting vendor benchmark claims as independently proven. Include a compact comparison table after the main visual, preserving links. State scope: selective rather than exhaustive; ‘major’ means a new flagship/model-family generation or a consequential broadly released open-weight model, not every checkpoint, quantization, fine-tune, or patch. Research scanned broad 2026 web coverage plus first-party developer/model-card searches; current through 2026-08-31. The headline finding: at least 20 consequential open/open-weight LLM releases were identified through August, with April the densest month in the curated set and million-token context becoming common among frontier open-weight launches.","result_set_ids":["r7d085f93d9d"],"title":"Major Open & Open-Weight LLM Releases — 2026 Timeline","prev_report_token":null}