Mistral启动万亿参数模型ML4公开预览,计划10月27日发布权重
法国AI公司Mistral于10月6日启动Mistral Large 4(代号“Le Chonk”,简称ML4)的公开预览。该模型总参数量达1万亿,激活参数为490亿,是一个多模态模型,接受多模态输入并生成文本输出,重点覆盖编程、网络安全、金融、制造与视觉定位等场景。
据Mistral介绍,ML4从零开始训练,使用4000块Nvidia Grace Blackwell GPU,在其位于欧洲的自有数据中心内耗时约两个月完成,训练数据覆盖160多种语言,包括欧盟所有官方语言。Mistral表示,ML4采用稀疏架构,推理时仅激活490亿参数,整体设计延续了大型开放模型在扩大总容量的同时降低推理成本的方向。
公司计划通过Mistral API提供该模型,并将在约三周测试期后,于10月27日公开发布模型权重,采用定制Mistral许可证。Mistral未在相关材料中公布API定价。公司称,预览期间将继续进行强化学习并调优最终检查点,以便在权重发布前进一步提升模型能力。
Mistral联合创始人兼首席科学家Guillaume Lample表示,ML4处于开放权重模型的前沿。Mistral还提供了ML4在DeepSWE、Harvey Legal Agent Benchmark等基准上的初步成绩,但相关排名仍需等待独立评测验证。
- 2026-10-07 20:03 | 36氪:海外开源模型重新提速:Mistral、“美版DeepSeek”同时亮牌阅读原文
西方开放模型阵营开始提速。 当地时间 10 月 5 日,被称作“美版 DeepSeek”的美国创业公司 Reflection 发布首款开放权重模型 Beam;一天之后,法国 Mistral 推出迄今最大的 Mistral Large 4。两家公司发布模型时选择的主要比较对象高度一致:GLM、Kimi、DeepSeek、Qwen。 Hugging Face 数据显示,过去一年,中国模型已经占到平台模型下载量的约 41%,月度下载量和累计下载量均超过美国。DeepSeek、阿里 Qwen、月之暗面 Kimi、智谱 GLM、MiniMax、小米 MiMo 等模型轮番刷新开放模型能力上限,中国...
- 2026-10-07 00:23 | X:RT Surge AI: For their Mistral Large 4 release, @MistralAI ran a blind Surge human eval, with our expert software engineers evaluating five frontier m...阅读原文
RT Surge AI For their Mistral Large 4 release, @MistralAI ran a blind Surge human eval, with our expert software engineers evaluating five frontier models on coding quality. Le Chonk ranked #1 among open-weight models and #2 overall, behind only Opus 5. Unit tests ask: does it work? Human code revie...
- 2026-10-06 22:33 | TechCrunch:Mistral’s new 1T model aims to leapfrog closed and open rivals阅读原文
The race between open and closed AI is on, and Europe is still in the mix. On Tuesday, French AI lab Mistral AI released Mistral Large 4 (ML4), a new large multimodal model aiming to leapfrog both American and Chinese rivals — following what French president Macron described as “a third way in AI.” ...
- 2026-10-06 22:02 | X:More from @MistralAI:阅读原文
Meet Mistral Large 4, aka Le Chonk. • 1T parameters, natively multimodal. 49B active.It is the best open weights model from US or Europe on aggregated benchmarks.• State-of-the-art on critical workloads, including cyber defense, manufacturing and finance and it surpasses closed frontier models on vi...
- 2026-10-06 22:02 | X:Le Chonk has arrived 🐈 Mistral Large 4 is now on OpenRouter in public preview: 1T params (49B acti...阅读原文
Le Chonk has arrived 🐈Mistral Large 4 is now on OpenRouter in public preview: 1T params (49B active), natively multimodal, 512K context, up to 256K output.50% off for the first two weeks: $0.68 in / $2.09 out per 1M tokens ($0.07 cached).
- 2026-10-06 21:56 | IT之家:宣称“欧美最强开源模型”:Mistral AI 发布 Mistral Large 4 公开预览版,月底开放权重阅读原文
感谢IT之家网友 fuhao 的线索投递! IT之家 10 月 6 日消息,当地时间 10 月 6 日,法国公司 Mistral AI 正式推出 Mistral Large 4 模型的公开预览版(Public Preview),简称 ML4,代号 le Chonk。即日起,用户可以在 Mistral Studio 上体验预览版 API,模型权重将于本月底正式开放下载。目前,官方正联合顶尖网络安全机构、受信任的合作伙伴及政府监管部门,在真实业务环境中对模型展开红队测试(Red-Teaming)。相关合作方将获得该模型的专项测试访问权限;该测试版本适当放宽了内容安全审查策略...
- 2026-10-06 21:00 | venturebeat.com:Mistral debuts Large 4 ‘Le Chonk', a 1-trillion parameter text output model with high benchmarks planned for open weights release阅读原文
Mistral is launching a public preview of Mistral Large 4, a one-trillion-parameter multimodal model code-named “Le Chonk” that the Paris AI company says pushes it back toward the global open-weight frontier — with a particular focus on coding, cybersecurity, finance, manufacturing and visual groundi...