Mistral releases Large 4 weights with a model-as-a-service license condition

What happened?

Mistral released the weights for Mistral Large 4, also called Le Chonk, on October 30. Mistral describes it as a roughly 1-trillion-parameter, natively multimodal mixture-of-experts model; its model card lists about 52 billion active parameters and support for text and image input, reasoning, coding and tool use.

The release is commercial but not unrestricted. Mistral's Ecosystem Open-Weight License allows use, modification, fine-tuning and commercialization, while businesses operating a Model-as-a-Service offering must obtain a separate commercial license. Mistral lists API pricing of $0.70 per million input tokens, $0.07 per million cached input tokens and $2.10 per million output tokens; those prices do not describe the cost of self-hosting the weights.

Mistral's model card reports a 48% score on the Artificial Analysis Coding Agent Index and says the model is effectively tied with GLM-5.3 while using fewer tokens per task. Those figures are vendor-published model-card claims here, not an independent LessCarbs evaluation.

Source: Mistral AI ↗ · Source: Mistral AI on Hugging Face ↗

Why does it matter?

For teams choosing between hosted APIs and local deployment, the important detail is the boundary between using weights and serving a model to other customers. The release offers more control over where the model runs, but a company building a hosted model service has an additional licensing step. Hardware, serving software, staffing and evaluation remain separate costs that the API price does not answer.