Mistral opens public API preview for Mistral Large 4; weights are planned for October

Mistral opened the public API preview of Mistral Large 4 on October 6, 2026, through Mistral Studio. The planned release of public model weights is a separate step, with a reported target date of October 27.

Mistral Large 4 launches in public API preview

The October 6 launch gave users access through Mistral Studio’s API. Mistral said the model’s weights would follow by the end of October; contemporaneous reports specified October 27 as the target. The preview and the planned weights release are distinct milestones.

Mistral calls the model “Le Chonk.” Its documented model identifier is mistral-large-4.

What Mistral documents about the model

Mistral describes Large 4 as a natively multimodal Mixture-of-Experts model. Multimodal models are designed to handle more than one type of input; a Mixture-of-Experts architecture distributes work among specialized components. Mistral’s documentation lists 1.05 trillion total parameters, a 1.6-billion-parameter vision encoder and a context window of 1 million tokens.

The documentation also lists API support for structured outputs, function calling, document question-answering, batching, agents, conversations and built-in tools. These are features available through the documented API, rather than a description of what the planned public weights release will include.

What the reported evaluations measure

Mistral reports 61.7% on DeepSWE v1.1, 59.4% on SWE-Atlas-QnA and 28.3% on Terminal-Bench 4. Its combined Coding Agent Index score is 49.8%. These are results on named coding evaluations, each with its own tasks and scoring.

For cybersecurity, Mistral reports 82% on a test asking a model to reproduce and patch a real open-source software vulnerability, and says Large 4 solved 93% of Cybench’s 40 challenges. On AutomationBench, which covers 657 workflows, the company reports 59.9%. For visual grounding, Mistral reports 42% on Dense 200, compared with 41% for GPT-6 Astra.

Mistral reported these results in its announcement. Each figure belongs to its named evaluation; the scores are not a single shared scale across coding, cybersecurity, workflows and vision.

API rates and the planned weights release

Mistral’s model documentation lists these API rates per million tokens: $0.68 for input, $0.07 for cached input and $2.09 for output. They are listed in US dollars as platform rates.

The public weights were planned for October 27, separate from the API preview that opened on October 6.