Mistral AI first offered Large 4 in a limited public preview earlier this month. On Thursday, the French lab opened the improved model’s weights more broadly, branding the trillion-parameter system “Le Chonk” and positioning it as a serious challenger in open AI. ¹

The company’s own announcement was spare but emphatic: “Mistral Large 4 (Le Chonk) is now open weight.” ² That matters because open-weight models can be downloaded, run and modified by customers rather than accessed solely through a provider’s API.

Mistral’s central pitch is technical and geopolitical. It says Large 4 is the “best open-weight model for coding” and has improved cyber-related capabilities, including vulnerability reproduction and patch creation, while being more likely than rival open models to refuse malicious requests. ¹ Pierre Stock, Mistral’s vice president of science, called the release “a pivotal moment in the global AI race” and “a turning point for Europe.” ³

The lab says additional reinforcement-learning work lifted its Terminal-Bench 4.0 result from 28% in the preview to 41% for the release, tying China’s GLM-5.3 on that coding-agent test. It also says Large 4 uses 47% fewer tokens per task than GLM-5.3. ³

But the comparison has a clear limit. Even on Mistral’s cited benchmark, Anthropic’s proprietary Mythos 5.1 scored 60.9%, leaving closed frontier systems ahead overall. ³ Mistral’s argument is therefore not that it has surpassed every rival; it is that open weights offer businesses continuity and control when access to closed models can be restricted or cut off.