Mistral opens its 1-trillion-parameter model for coding and cyber defense

Mistral has released the weights of Mistral Large 4, its 1-trillion-parameter multimodal model nicknamed Le Chonk. Organizations can now download, customize and run the model on their own infrastructure instead of depending on Mistral’s API.

The mixture-of-experts system activates about 52 billion parameters per token and supports a 1-million-token context window, text and image input, tool use and three reasoning modes. Mistral says additional reinforcement learning after the public preview improved agentic coding and defensive-security performance.

On Cognition’s FrontierCode 2.0, the model scored 19.7%, narrowly ahead of DeepSeek V4.1 Flash at 19.3%. Mistral also reports a 48% score on Artificial Analysis’s Coding Agent Index while using 47% fewer tokens per task than GLM-5.3, and 41% on Terminal-Bench 4.0. These are benchmark results supplied or cited by the vendor, not independent proof of performance in every workload.

Cybersecurity is the other main pitch. Mistral reports 51% on the Artificial Analysis Cyber Index and a 96.6% refusal rate for malicious cyber requests. The company says it has tested the model with organizations including Cloudflare, CrowdStrike and ENISA.

The weights use Mistral’s Ecosystem Open-Weight License, a modified MIT license that permits commercial and non-commercial use, modification and fine-tuning. Model-as-a-Service providers need a separate commercial license. The release can be self-hosted with vLLM or SGLang, giving companies a frontier-scale European model they can keep inside their own infrastructure.