Mistral Launches Public Preview of Large 4 'Le Chonk,' a One-Trillion-Parameter Model with Open Weights Planned
French AI company Mistral has entered Mistral Large 4 into public preview, with plans to release its one-trillion-parameter multimodal model's open weights on October 27 following a roughly three-week testing period with developers, cybersecurity leaders, and government authorities.
Mistral has launched a public preview of Mistral Large 4 — nicknamed "Le Chonk" — a one-trillion-parameter multimodal model the Paris-based company says puts it back at the open-weight frontier. The model uses a sparse architecture: although it holds one trillion parameters in total, only 49 billion are active during any single inference pass, keeping computation manageable at enormous scale. Training took roughly two months on 4,000 Nvidia Grace Blackwell GPUs inside Mistral's own European data centers, covering more than 160 languages including every official EU language.
On benchmarks, ML4 reached 62% on DeepSWE, a software-engineering evaluation — ahead of Qwen 3.8 Max at 51%, DeepSeek V4 Pro 0813 at 57%, and GLM-5.3 at 61%. It scored 67% on Finch, an enterprise finance-and-accounting benchmark accepted to ACL 2026, tied with DeepSeek V4 Pro 0813 and ahead of GLM-5.3 at 65%. On Harvey's Legal Agent Benchmark, ML4 achieved a 15% task-pass rate. Mistral's comparisons have not yet been independently verified.
Co-founder and chief scientist Guillaume Lample called ML4 "at the frontier of open weight models" and said capabilities should improve further as reinforcement learning concludes. Mistral is pitching the model to enterprises and governments that want to run AI on sovereign infrastructure, with zero-data-retention options. Weights are planned for public release on October 27 under a custom Mistral license; API pricing was not disclosed.
Mistral, founded in 2023, recently closed a €3 billion Series D at a post-money valuation above €21 billion — described as the largest equity fundraise ever by a European technology company.
