Choose Rich Live
Tech

Mistral AI announces 1-trillion-parameter Mistral Large 4 with open weights set for late October

Mistral AI has introduced Mistral Large 4, a 1-trillion-parameter open-weight model nicknamed Le Chonk. Currently available through an API, its full model weights are scheduled for public release on October 27, 2026.

A promotional scene showing a fluffy cat facing a large pixelated orange number four inside an art gallery.
Image: @testingcatalog

Mistral AI has unveiled Mistral Large 4, an open-weight multimodal model with 1 trillion parameters nicknamed Le Chonk after a June internet cat meme. The model rolled out via API on October 6, 2026, with full model weights scheduled for release on October 27 so organizations such as banks and public institutions can host it independently. According to Wired, Mistral hopes the launch demonstrates it remains in the race to develop frontier-level artificial intelligence as part of a broader European sovereign AI push.[1][3][4][5][7][10][15]

The model features 49 billion active parameters and native multimodality, including visual grounding capabilities. Mistral trained the system from scratch on multilingual datasets spanning more than 160 languages. The training run took two months and was conducted on Mistral's own infrastructure across European data centers using between 3,800 and 4,000 Nvidia Grace Blackwell GPUs.[3][4][6][8][9][11][13][15]

Mistral is pitching the model directly against proprietary offerings from OpenAI and Anthropic, arguing that organizations using closed APIs have no guarantee those models will remain available. Mistral claims the model rivals leading Chinese open-weight architectures and achieves top results on enterprise workloads like cyber defense, manufacturing, and finance. Decrypt reported that Mistral's model topped GPT-6 Astra on one finance benchmark, though it trailed Claude on other tests.[1][2][4][6][9][12][14][15]

Key facts

  • Mistral AI unveiled Mistral Large 4, an open-weight model with 1 trillion total parameters and 49 billion active parameters, nicknamed Le Chonk.
  • Mistral Large 4 launched via API on October 6, 2026, with public model weights scheduled for release on October 27.
  • The model was trained over two months using between 3,800 and 4,000 Nvidia Grace Blackwell GPUs in Mistral's European data center infrastructure.
  • Mistral trained the model from scratch on multilingual data spanning more than 160 languages.
  • Mistral reported that Mistral Large 4 beat GPT-6 Astra on a finance evaluation but trailed Claude on other benchmarks.
  • Mistral positioned the open-weight release as an alternative to proprietary APIs from OpenAI and Anthropic, arguing closed models carry no guarantee of lasting availability.

Sources · 14 sources

  1. DE

    DecryptArticle ·

    Mistral AI Drops 'Le Chonk': A Massive AI Model Named After a Cat Meme Paris-based Mistral launched Large 4, a model nicknamed after a June internet joke. It tops GPT-6 Astra on one finance test, but trails Claude on others.

    Open source
  2. TD

    The Daily Block@thedailyblockPost on X ·

    🚨 BREAKING: Mistral Launches Large 4 AI Model Mistral says Large 4 outperforms several Chinese and open-weight rivals and is set to launch publicly on October 27.

    Open source
  3. TN

    TNW@thenextwebPost on X ·

    Just in: 4,000 GPUs and two months built a 1T model. Mistral is releasing the weights on 27 October, so a bank or a government can download it and run it themselves. https://t.co/kpAXgJxlS7 https://t.co/Gi1nMMmfpX

    Open source
  4. TA

    Techstrong.ai@TechstrongaiPost on X ·

    Mistral AI has unveiled Mistral Large 4, a 1-trillion-parameter open-weight model nicknamed Le Chonk, trained from scratch on more than 160 languages and aimed squarely at the closed APIs from OpenAI and Anthropic. Full weights arrive Oct. 27, and the pitch is blunt: if you use a closed model, there is no guarantee it will still be there tomorrow. Read the full article on Europe's bid for AI sovereignty and what Le Chonk can do: https://t.co/vgll7v9aji #AI #Mistral #OpenWeights #AISovereignty

    Open source
  5. WI

    WIRED@WIREDPost on X ·

    With the release of its new trillion-parameter model, Mistral is hoping to demonstrate it’s “still in the race” to build frontier-level artificial intelligence. https://t.co/GGMwuNsY19

    Open source
  6. 阳明

    阳明AI@x_autonomyPost on X ·

    【AI热点】20:00-22:00 更多详细信息→https://t.co/bbTrn8x0VI 1. Mistral 发布 Mistral Large 4：1T 参数、49B 激活的原生多模态模型 [影响大·可执行高]Mistral 发布 Mistral Large 4，1T 参数、49B 激活，原生多模态，称为美国或欧洲最佳开放权重模型。该模型在网络防御、制造和金融等关键负载上达到 SOTA，视觉 groundi 2. Google 将于 10 月 9 日起限制免费用户访问 Gemini Flash 和 Pro [影响大·可执行高]Google 宣布自 10 月 9 日起，Gemini 免费用户将只能使用 Flash Lite 模型，使用标准 Flash 模型需要 4.99 美元/月的 Google AI Plus 订阅。 3. Gamma 发布 Gamma 5：以智能体为中心重构演示文稿制作 [影响大·可执行高]Gamma 发布 Gamma 5，把智能体放在平台规划、设计和编辑演示文稿的核心位置。智能体会先询问目标和受众并展示计划再动手；支持导入 PowerPoint 并在保留原内容的情况下套用设计样式重做， 4. Meta 论文 RankEvolve：两个不同编码智能体互相审查补丁比加大单智能体预算更能发现隐性缺陷 [影响大·可执行高]Meta 论文《RankEvolve: A Reliable Multi-Agent Auto-Research Harness for Evolving Ranking Models》（arxiv. 5. Ant Group 发布 Ling 3.1 Flash，Artificial Analysis 智能指数得分从 20 升至 41 [影响大·可执行高]Artificial Analysis 评测 Ant Group 发布的 Ling 3.1 Flash，该推理模型总参数 560B、激活 25B、上下文窗口 1M token，智能指数 v4.3 得分 6. 多名美国议员推出法案限制联邦机构使用 Flock 摄像头 [影响大·可执行高]美国民主、共和两党及独立议员提出多项法案限制联邦机构访问 Flock 自动车牌识别摄像头。Sanders、Ocasio-Cortez 和 Merkley 提出 Ban Flock Act。 7. Eleven v4 与 Eleven v4 Turbo 登顶 Artificial Analysis 语音合成榜单前两名 [影响大·可执行高]ElevenLabs 宣布 Eleven v4 和 Eleven v4 Turbo 在 Artificial Analysis 文本转语音榜单上分列第一和第二。引用数据显示 Eleven v4 Tur 9. llama.cpp 支持 Qwen3.8-27B DFlash 加速 [影响中·可执行中]如果你在使用 Qwen3.8-27B + MTP，请务必升级到 DFlash 以获得额外加速： llama serve -hf ggml-org/Qwen3.8-27B-GGUF --spec-ty 10. PixVerse 插件一键生成讲解视频 [影响中·可执行中]用 PixVerse 插件为任何主题创建讲解视频 告诉你的 AI 智能体你想讲解什么 它会整理关键要点，并用 PixVerse 将其转化为清晰的视觉故事（补充：PixVerse 插件获取：获取插件 11. HLA-WM：面向长时程视频世界模型的混合线性注意力框架 [影响中·可执行中]HLA-WM 是一种免训练的混合线性注意力框架，将几何引导的粗粒度检索与细粒度循环线性状态计算结合，缓解 Gated DeltaNet 的长程遗忘问题。在 60 秒 SANA-WM-Bench 上，它 12. 人类是异步 API：多智能体协作架构思考 [影响小·可执行中] 13. Meta 广告 MCP server 上线 Databricks Marketplace，接入 Genie 用自然语言投放广告 [影响大·可执行高]Meta 的广告 MCP server 已在 Databricks Marketplace 上线，暴露覆盖广告活动全生命周期的 25+ 工具，可创建广告系列和广告组、调整预算与出价、管理受众和创意并拉 14. 电信运营商为何将 AI 战略押注在开放模型上 [影响中·可执行中]NVIDIA《State of AI in Telecommunications》报告显示，89% 受访者认为开源模型与软件对公司 AI 战略重要。NVIDIA 发布 300 亿参数的 Nemotro

    Open source
  7. MT

    MTS@MTSlivePost on X ·

    SITUATION DETECTED: Mistral announced the 1T Mistral Large 4, "Le Chonk", available via API today, open weights will be released at the end of the month.

    Open source
  8. TE

    Techmeme@TechmemePost on X ·

    Mistral releases Mistral Large 4, dubbed "le Chonk", a 1T-parameter open-weight model for general agentic capabilities, trained on 4,000 Grace Blackwell GPUs (@sabrinaa_ortiz / The Deep View) (Visit Techmeme dot com for the link and full context!)

    Open source
  9. AG

    AGTP@AGTPinsightsPost on X ·

    BREAKING: Mistral announces Mistral Large 4, an open-weight AI model it says beats some Chinese open-weight models on cyber. It goes fully public October 27, trained on its own infrastructure. https://t.co/XQPhwWKPSy

    Open source
  10. TS

    The Straits Times@straits_timesPost on X ·

    France’s Mistral unveils ‘Le Chonk’, its latest model in sovereign AI push https://t.co/h7b0DUT0zR

    Open source
  11. TE

    Techmeme@TechmemePost on X ·

    Mistral says ML4 was trained using 3,800 Nvidia Grace Blackwell GPUs in its own data centers in Europe and much of its training data was multilingual (Mistral Blog) (Visit Techmeme dot com for the link and full context!)

    Open source
  12. CN

    CNBC@CNBCPost on X ·

    Mistral unveils new AI model it says rivals best open systems from China https://t.co/ywlTh3h154

    Open source
  13. QU

    Quartz@qzPost on X ·

    Meet Le Chonk: The 1-Trillion-Parameter AI Giant Mistral AI just dropped its biggest model yet, and it claims nothing built outside China can touch it. Trained on 4,000 Nvidia GPUs across two months inside European https://t.co/qHMxfDJ6L4 #MistralLarge4 #OpenWeightAI #LeChonk https://t.co/L9skdorEX2

    Open source
  14. RE

    Reuters@ReutersPost on X ·

    France's Mistral launches AI model it says outperforms some Chinese rivals https://t.co/vY60sSZcII https://t.co/vY60sSZcII

    Open source
  15. �A

    🚨 AI News | TestingCatalog@testingcatalogPost on X ·

    BREAKING 🔥: Mistral announced Mistral Large 4 "Le Chonk", a new 1T-parameter open-weight model! > 49B active parameters, native multimodality. > Rolling out via APIs today; open-weight release is planned for the end of October. > SOTA on "critical workloads", including cyber defense. Le Chaton Fat "Le Chonk" is here 👀 https://t.co/FDh5mmvZxH

    Open source