Mistral Releases the Weights of Its Trillion-Parameter Model, With One Condition
Mistral released the weights of Mistral Large 4 on Oct. 29, 2026, which means the 1-trillion-parameter model can now be downloaded and run on hardware its users own rather than reached through Mistral's servers. It comes under what the company calls the Mistral Ecosystem Open-Weight license, which Mistral describes as a modified MIT license.
The company states that the model is free to use, modify, fine-tune and commercialize, with a single condition: a business that operates a model-as-a-service offering must obtain a separate commercial license from Mistral first. The weights are handed out through Hugging Face, and the model is also served through Mistral's own interface, OpenRouter and Microsoft Foundry.
The model released is not the one Mistral put into public preview earlier in October 2026. On its release page the company says three further weeks of reinforcement learning, a trial-and-error training stage, lifted its own score on Terminal-Bench 4.0 by 13 percentage points to 41% while cutting output tokens by 30%. It also reports 19.7% on FrontierCode 2.0, a private benchmark run by Cognition, and 48% on the Artificial Analysis Coding Agent Index, where it says the model is effectively tied with GLM-5.3 while using 47% fewer tokens per task. Those are the company's own figures, and no independent evaluation of the released weights has been published.
The release is aimed at security teams, and those numbers are the company's too. Mistral reports 51% on the Artificial Analysis Cyber Index, which it says measures whether a model can find flaws in a codebase, reproduce them and patch them without breaking what already works, and an average refusal rate of 96.6% on malicious cyber requests across three public safety benchmarks. The company adds that, watching its live traffic since the preview, it sees a potential for the model to help carry out real attacks, and that it is changing its guardrails.
Mistral prices the hosted version at $0.70 per million input tokens and $2.10 per million output tokens. A full technical report has not been published.
