French AI company Mistral released the weights for Mistral Large 4, a 1-trillion-parameter open-weight model nicknamed Le Chonk. The system targets coding and defensive cybersecurity for businesses and governments. It is now available to download and deploy on users' own infrastructure.
Performance and efficiency
Mistral said the model leads open-weight benchmarks in agentic coding and defensive security. It scored 51% on the Artificial Analysis Cyber Index and 41% on Terminal-Bench 4.0. The company added that it uses 47% fewer tokens per task than GLM-5.3 in coding agent tests.
Company context
Mistral is based in France and said the model was trained from scratch in Europe. The company argues that open weights give organizations control over their AI and prevent access from being cut off. The model is available through Hugging Face, Mistral's API, Microsoft Foundry, and NVIDIA-hosted endpoints.
What users get
Mistral Large 4 offers low, medium, and high reasoning modes for different task needs. Its built-in guardrails refused 96.6% of malicious cyber requests in company evaluations. Pricing starts at $0.70 per million input tokens and $2.10 per million output tokens.
Mistral said a full technical report on the model will be published soon. The company also plans to build specialized and optimized models on top of Mistral Large 4. Select enterprise partners can preview a cybersecurity-specific variant under vetting and contractual safeguards.