Paris-based Mistral AI has published the open weights for Mistral Large 4, the trillion-parameter flagship it nicknames “Le Chonk.” Announced on October 29, the release follows the model’s October 6 preview and adds three more weeks of reinforcement learning, which Mistral says improved coding, defensive cybersecurity, and reasoning efficiency.
Large 4 is a mixture-of-experts model with roughly 1 trillion total parameters and about 52 billion active per token, a 1-million-token context window, and native text and image input. The weights are available from Hugging Face, and the model is also served through Mistral Studio, the Mistral API, OpenRouter, NVIDIA-hosted endpoints, and Microsoft Foundry. Hosted pricing starts at $0.70 per million input tokens and $2.10 per million output tokens.
The release uses Mistral’s Ecosystem Open-Weight License, a modified MIT license permitting free use, fine-tuning, and commercialization, with one condition: businesses offering the model as a hosted service need a separate commercial license. Developers can choose between three reasoning modes — low, medium, and high — to trade performance against cost.
Mistral’s own benchmark figures put Large 4 at 19.7% on Cognition’s FrontierCode 2.0, 48% on the Artificial Analysis Coding Agent Index, effectively tied with GLM-5.3 while using 47% fewer tokens per task. On security, it reports 51% on the Artificial Analysis Cyber Index and a 96.6% average refusal rate on malicious cyber requests. These are company-reported numbers; independent validation will take time.
Mistral also disclosed that it is tightening guardrails after observing potential for real-attack uplift during the preview, and a specialized penetration-testing variant will stay restricted to vetted partners. For teams weighing open versus closed flagships, the pitch is control: run the weights under your own policies. Mistral’s assistant Le Chat runs on its models; rivals include Claude and DeepSeek. See our ChatGPT vs Claude comparison.
Source: Unite.AI
