Mistral says it has released the weights for Mistral Large 4, a 1-trillion-parameter natively multimodal model that was previously in public preview. The company says three additional weeks of reinforcement learning improved coding and defensive cybersecurity, alongside gains in reasoning efficiency.
Mistral says the model scored 19.7% on Cognition’s FrontierCode 2.0 benchmark and 48% on the Artificial Analysis Coding Agent Index while using 47% fewer tokens per task on average than the leading open-weight model in that comparison. It also says the model supports Low, Medium and High reasoning modes.
On cybersecurity, Mistral reports a 51% score on the Artificial Analysis Cyber Index and an average 96.6% refusal rate for malicious cyber requests. The company says the model was trained on European infrastructure and can be customized and self-hosted; the page does not specify deployment terms or independently verify its benchmark claims.
Open weights give AI builders more scope to inspect, customize and host a capable coding and security model under their own control, potentially helping with data-sovereignty needs. The caveat is that the performance and safety figures are company-reported benchmark results, not a guarantee of real-world outcomes.