Mistral AI has released the open weights for Mistral Large 4, the 1 trillion parameter model it teased as "Le Chonk" earlier this month. The model runs 49 billion active parameters, is natively multimodal with text and image input, and carries a 512k token context window. Mistral says it was forged in Europe end to end and can be deployed from Europe through its own Mistral Cloud infrastructure.
On benchmarks, Mistral Large 4 scores 38 on the Artificial Analysis Intelligence Index, roughly level with GPT-6 Luna (max, 38) and just behind DeepSeek V4.1 Flash (max, 39). That makes it the most intelligent model from outside the United States and China, ahead of challengers from South Korea and the UAE. It also scores 50 on the Artificial Analysis Cyber Index, tied with GLM-5.3-Flash and behind MiMo-V2.6-Pro at 56, and hits 82 percent on CyberGym-E2E-AA. Mistral is pitching it at cyber defense, manufacturing and finance workloads, and claims it beats closed frontier models on visual grounding.
The model has been served on Mistral's API as a Research Public Preview at USD 1.36 per million input tokens and USD 4.18 per million output tokens, with cached input at USD 0.14 and a 50 percent launch discount for the first two weeks. The one caveat is efficiency: at USD 1.13 per Intelligence Index task, Mistral Large 4 costs over four times as much per task as open weights rivals such as GLM-5.3-Flash and DeepSeek V4.1 Flash. For teams that value weight access and European hosting over raw price per token, it is now the strongest downloadable option on the continent.