Mistral releases open weights for trillion-parameter Large 4 'Le Chonk' aimed at coding and cyber defense

Mistral Large 4, nicknamed 'Le Chonk', is the French lab's new open-weight flagship. It is a 1-trillion-parameter multimodal model that activates 52 billion parameters per forward pass and was trained on 3,800 NVIDIA Grace Hopper superchips. Mistral says it scores 41% on Terminal-Bench 4.0. On Artificial Analysis' Coding Agent Index it leads all open models, matching GLM-5.3 while consuming 47% fewer tokens.

The model ships with selectable reasoning presets tuned for defensive cybersecurity and enterprise code generation. Orange Cyberdefense is the first European managed-security design partner for a cyber-optimized variant. Mistral is pitching the release as a sovereign European alternative for security teams.

The positioning is explicitly geopolitical. CEO Arthur Mensch warned that US export controls could cut Europe off from advanced AI, and some coverage framed Le Chonk as surpassing Chinese models on cybersecurity tasks. That matters in a week when AWS CEO Andy Jassy said the West is 'very hungry' for Western open models even as Bedrock adds Kimi and GLM from China.

The caveats come from the license and from community reception. Hacker News commenters questioned whether the MEOW License is truly 'open' given its commercial carve-outs. One commenter rated Large 4 'fine, but certainly not up to the best Chinese or US open weight models'. Independent agentic benchmarks over the coming weeks will show whether the token-efficiency claim holds up in practice.