Mistral releases the open weights of its trillion-parameter Large 4 model, pitched for cyber defense
French AI company Mistral on Thursday released the weights of Mistral Large 4, its 1-trillion-parameter model nicknamed Le Chonk, so businesses and governments can download it and run it on their own systems, TechCrunch reported. Mistral is pitching the model for coding and defensive cybersecurity. The performance figures below are Mistral’s own benchmark claims.
What it is: Mistral Large 4 is a multimodal model that first appeared in public preview earlier this month, Mistral said. The weights are available on Hugging Face, and the model is also offered through Mistral’s API and Microsoft Foundry.
The cyber pitch: Mistral said the model scores 51% on the Artificial Analysis Cyber Index, which tests finding, reproducing and patching software vulnerabilities. It said the model refused 96.6% of malicious cyber requests on average in its tests. Mistral also said it saw potential uplift for real attacks in live traffic and is adjusting its guardrails.
The coding claims: Mistral said the model scored 48% on the Artificial Analysis Coding Agent Index, about level with China’s GLM-5.3, while using 47% fewer tokens per task. TechCrunch reported the same comparison.
The license: Mistral said the model is free to use, modify and sell under a modified MIT license, but companies that run a model-as-a-service business must get a separate commercial license. API pricing starts at $0.70 per million input tokens and $2.10 per million output tokens.
Why it matters: Open weights let an organization host the model itself instead of relying on an outside provider. Pierre Stock, Mistral’s vice president of science, told TechCrunch that this means access to the model can’t be cut off.