Mistral 发布万亿参数编码模型 Le Chonk，整体仍落后中国开源模型
原标题：Paris-based Mistral releases "Le Chonk," a strong coding model that still trails China overall
AI 摘要
法国 Mistral 发布其迄今最大模型 Mistral Large 4（代号 Le Chonk）的权重，约 1 万亿参数、每步激活 520 亿，采用修改版 MIT 许可，可在自有硬件上运行。该模型在编码基准上领先开源模型，但 Artificial Analysis 智能指数仅 26 分，落后于 DeepSeek V4.1 Flash、GLM-5.3 等中国开源模型，也远逊于 Anthropic 和 Google 的领先系统。其优势是 161 tokens/秒的速度、较低的 API 价格，以及对有害网络请求 96.6% 的拒绝率。
正文节选
Paris-based Mistral releases "Le Chonk," a strong coding model that still trails China overall Key Points - Mistral has released the weights for Large 4 ("Le Chonk"). The model has about one trillion parameters, 52 billion of them active, and can run on your own hardware under a modified MIT license. - In independent benchmarks, the model does well on coding. On Artificial Analysis' Intelligence Index, though, it scores 26 points, behind open Chinese models like DeepSeek V4.1 Flash and GLM-5.3 a