Mistral opens the weights of Mistral Large 4 "Le Chonk": a 1-trillion-parameter multimodal model under an MIT-based license
Prepared by Sintelligent with the help of SI. Published in SI: . Updated: .
An open trillion-parameter model trained in Europe reaches the level of the best open models in coding, by the company's own claim.
According to Mistral, the weights of the improved Mistral Large 4 (Le Chonk) are now open; earlier this month the model came out in public preview. It is a natively multimodal mixture-of-experts model with 1 trillion parameters, of which, per the Hugging Face description, 49 billion are active; the context is 1 million tokens, and the formats are FP8, NVFP4 and GGUF. The license is the Mistral Ecosystem Open-Weight License, derived from MIT, with restrictions for "model-as-a-service" offerings. The company calls the model the best open one for agentic coding and cybersecurity and says it is on par with GLM-5.3 on the Artificial Analysis Coding Agent Index. Companies whose testimonials appear on the release page include Cisco, Cloudflare, CrowdStrike, Fortinet and ENISA.
Sources 2