Mistral Large 4 (Le Chonk) is now Open Weight

Mistral Large 4, nicknamed “Le Chonk,” is a one-trillion-parameter multimodal model whose weights are now openly available. Mistral says it was trained from scratch on European infrastructure and is intended to let developers, businesses, and public institutions run, customize, and control AI on their own systems. Following a public preview, additional reinforcement learning improved its coding and defensive cybersecurity performance while making its reasoning more token-efficient. Developers can choose low, medium, or high reasoning: low for simple tasks, medium for everyday work, and high for demanding tasks requiring precision.

The model leads open-weight competitors on several coding measures. It scored 19.7% on Cognition’s FrontierCode 2.0 benchmark, reached 48% on the Artificial Analysis Coding Agent Index - matching the top open model while using 47% fewer tokens per task - and scored 41% on Terminal-Bench 4.0. Human evaluators also preferred its coding responses to several leading open models. For cybersecurity, it scored 51% on the Artificial Analysis Cyber Index, which tests vulnerability discovery, reproduction, and patching, while refusing 96.6% of malicious cyber requests. The release emphasizes that open weights enable security teams to inspect and adapt the model, keep sensitive data under their control, and deploy it in environments with strict sovereignty requirements.