Mistral releases the weights of Mistral Large 4 'Le Chonk', a 1-trillion-parameter multimodal model pitched at coding and cyber defence

French AI company Mistral AI has published the weights of Mistral Large 4, nicknamed “Le Chonk”, on Hugging Face (repository Mistral-Large-4-1T-A52B). It is a natively multimodal mixture-of-experts model with about 1 trillion parameters, first shown in public preview earlier this month. Mistral says it was trained from scratch on European infrastructure.

What Mistral reports

• After about three more weeks of reinforcement learning, it leads open-weight models on Cognition’s FrontierCode 2.0 (19.7%) and Terminal-Bench 4.0 (41%, up 13 points from the preview).
• 48% on the Artificial Analysis Coding Agent Index, roughly tied with GLM-5.3 while using 47% fewer tokens per task.
• 51% on the Artificial Analysis Cyber Index, and a 96.6% average refusal rate on malicious cyber prompts.
• Three reasoning modes (low, medium, high).

Safety note from Mistral

Mistral says that while monitoring live traffic it saw “a potential for uplift to conduct real attacks” and is tightening its guardrails. A specialised offensive-security variant is offered only to vetted partners such as Orange Cyberdefense.

Licence and price

The Mistral Ecosystem Open-Weight licence (a modified MIT licence) allows free use, fine-tuning and commercial use, but companies running a model-as-a-service business need a separate commercial licence. API price: $0.70 per million input tokens and $2.10 per million output tokens. It is also offered through Microsoft Foundry and NVIDIA NIM.

All benchmark figures are Mistral’s own; a full technical report is “coming soon”.