Mistral opens weights of 1-trillion-parameter Large 4: No. 1 open model for coding, $0.70 per million input tokens

French AI company Mistral AI released the weights of Mistral Large 4 ("Le Chonk"), a 1-trillion-parameter multimodal model, on Oct. 29. The Hugging Face model name (Mistral-Large-4-1T-A52B) points to 52 billion active parameters. Mistral's X post passed 4,800 likes and 420,000 views.

In its blog, Mistral says three more weeks of reinforcement learning after this month's preview put it first among open models on Cognition's private FrontierCode 2.0 at 19.7%, against 19.3% for DeepSeek V4.1 Flash, 17.8% for GLM-5.3 and 17.3% for Kimi K3. On the Artificial Analysis Coding Agent Index it scores 48%, effectively tied with GLM-5.3, using 47% fewer tokens per task. Terminal-Bench 4.0 rose 13 points to 41%. All scores are Mistral's own.

Weights are on Hugging Face for vLLM or SGLang, and the model is on Mistral's API, OpenRouter, Microsoft Foundry and NVIDIA NIM. API prices are $0.70 per million input tokens, $0.07 cached and $2.10 output. The license is a modified MIT: free to use, modify and commercialize, but model-as-a-service businesses need a separate license.

Mistral says it saw potential for uplift in real attacks during the preview, is adjusting guardrails, and offers a specialized security variant only to vetted partners.

Unsloth shipped GGUF files for local use (136 likes), and one user wrote it "beats DeepSeek V4.1 Flash 1017 and Kimi K3" (50). The news also drew a thread on r/LocalLLaMA.