ميسترال تنشر أوزان Mistral Large 4: نموذج بحوالي تريليون معامل تقدر الشركات تشغّله على سيرفراتهاMistral releases the weights of Mistral Large 4, a model of about 1 trillion parameters that companies can run on their own servers

Mistral has released the weights of its flagship model Mistral Large 4, nicknamed "Le Chonk", after putting it in public preview earlier this month. It is a natively multimodal model with about 1 trillion parameters, and Mistral says it is the best open-weight model for coding.

According to Mistral, it leads Cognition's private FrontierCode 2.0 benchmark with 19.7% and scores 48% on the Artificial Analysis Coding Agent Index, effectively tied with GLM-5.3 while using 47% fewer tokens per task on average. It offers three reasoning modes, low, medium and high, to balance precision and cost.

On cybersecurity, Mistral says it is strong at defensive tasks and refused 96.6% of malicious cyber requests in its tests. But the company admits it sees potential for uplift in real attacks and says it is adjusting the guardrails. A specialised cybersecurity variant is going only to selected enterprise partners such as Orange Cyberdefense.

The weights are on Hugging Face, and the licence allows use, modification and commercial use, though companies offering the model as a service need a separate commercial licence. Mistral says the model was trained from scratch on European infrastructure and is fluent in more than 160 languages. API pricing is $0.70 per million input tokens and $2.10 per million output tokens.