Mistral has launched a public preview of Mistral Large 4 — nicknamed "Le Chonk" — a natively multimodal AI model with one trillion total parameters and 49 billion active parameters using a mixture-of-experts architecture, marking Europe's biggest swing yet in the global open-model race.
The model was trained from scratch over roughly two months on about 3,800 to 4,000 Nvidia Grace Blackwell GPUs in Mistral's own European data centres, and covers more than 160 languages including every official EU language. An API is available now through Mistral Studio, with open weights scheduled for release on October 27 under a custom Mistral licence, following about three weeks of red-teaming with cybersecurity leaders, vetted partners and state authorities.
Mistral claims the model significantly outperforms any open-weight model developed in the United States or Europe on aggregated benchmarks, with state-of-the-art results on enterprise workloads — including cybersecurity benchmarks where it scored 82 per cent on the Artificial Analysis Cyber Index and resisted 93.3 per cent of attacks on Lakera's B3 safety benchmark.
The launch lands one day after U.S. rival Reflection unveiled its "Beam" model, positioning Mistral — fresh off a €3 billion Series D funding round — as Europe's answer to the American and Chinese open-model push.
