Mistral released Large 4 on Oct. 6, a 1 trillion parameter model that uses a mixture of experts design with 49 billion active parameters, the company said in its announcement. The model, nicknamed Le Chonk, is available now in public preview through the Mistral Studio API, with full open weights due by the end of October.
Preview access costs $1.36 per million input tokens and $4.18 per million output tokens, according to Mistral. The company said it trained Large 4 from scratch on 3,800 Nvidia Grace Blackwell GPUs in its own European data centers, with a single training run producing roughly 33 billion tokens a day.
Mistral said Large 4 solves 93% of challenges on the Cybench security benchmark and scores 82% on one cybersecurity index test, the highest of any model tested. On Artificial Analysis's composite benchmark, the model scores 38, just behind the 552 billion parameter DeepSeek 4.1 Flash and a sharp jump from the 9 that Mistral Large 3 scored last December, according to independent developer Simon Willison.
Mistral vice president of science Pierre Stock said the company wants outside partners and governments to vet the open weights before release, TechCrunch reported. "In the meantime, 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," Stock said, according to TechCrunch. Stock said the model took two to three times less compute to train than Chinese rivals' models and significantly less than closed-source competitors, TechCrunch reported.
A trillion-parameter model with pricing posted today and open weights pending gives developers a concrete option to benchmark against, not just a promise. Whether Large 4 holds up once outside testers get the weights will decide if Mistral closes the gap with frontier labs or just talks about it.