The system features 1T total and 49B active parameters and was trained on 3,800 Nvidia Grace Blackwell GPUs in European datacenters. Now in public preview on Mistral's API and OpenRouter, the model includes a 512k token context window and can process 100 images per request. Mistral is continuing reinforcement learning on the preview version to improve performance before the full open-weight release at the end of October.
Tuesday, October 20, 2026
Story
Mistral Le Chonk Tops Intelligence Index for Non-US and China AI
AI15 days ago
Mistral Large 4 scored 38 on the Artificial Analysis Intelligence Index, the highest mark of any AI system produced outside the U.S. and China. The model, also known as "Le Chonk," is comparable to GPT-6 Luna and DeepSeek V4.1 Flash and ranks among the top three open-weight models on the Cyber Index with a 50% score. It led rivals MiMo-V2.6-Pro and GPT-6 Luna on the CyberGym-E2E-AA benchmark with an 82% result, though it is over 4x more costly per task than rivals such as GLM-5.3-Flash, costing $1.13 against $0.25.
Looking for more context or analysis?
Tilt chat provides reliable, tailored insights, alerts, and automations.