French AI lab Mistral AI has released Mistral Large 4 — nicknamed “Le Chonk” — a new large multimodal model packing 1 trillion parameters, in a bid to leapfrog both American closed models and Chinese open-weight rivals.
The model uses a sparse mixture-of-experts design: only 49 billion parameters activate for any given token, delivering frontier-level punch at inference costs closer to a 49-billion-parameter system. It handles more than 160 languages with a 1-million-token context window and native text-and-image multimodality.
For now, ML4 is available only through a guarded API preview on Mistral Studio; the company plans to release the weights by the end of October after safety testing with cybersecurity teams, developers and government partners. “We’ll work with trusted partners and governments to make sure that the open source weights can be used to defend, but not to [perform] malicious attacks,” VP of science Pierre Stock told TechCrunch.
Notably, Mistral trained the model from scratch in its own European data centres in about two months on roughly 4,000 Nvidia Grace Blackwell GPUs — two to three times less compute than Chinese competitors burned, Stock said. The release follows what French President Emmanuel Macron has called a “third way in AI” — Europe’s bid for AI sovereignty between locked-down American models and increasingly dominant Chinese open weights.
