Mistral's New Model Is Open to Try, but Not yet to Download
Mistral opened a public preview of Mistral Large 4 on Oct. 6, 2026, and the only figures published for the model so far are the company's own. The French lab says it carries 1 trillion parameters, of which 49 billion are active for any one input, and describes it as natively multimodal. The preview is reachable through Mistral's API on Mistral Studio.
The weights have not shipped. Mistral says it will publish them by the end of October 2026, so for now the preview is access to a hosted service rather than a model that can be downloaded and run. In its announcement, the company describes open weights and self-deployment as the purpose of the release, saying organizations will be able to run the model on private cloud or on their own machines under their own policies.
Mistral says the model was trained from scratch on 3,800 NVIDIA Grace Blackwell GPUs in the company's own European data centers, and that the preview runs on that same hardware. It also says the training data spanned more than 160 languages, including every official language of the European Union. The preview is listed at $1.36 per million input tokens and $4.18 per million output tokens.
Every performance claim in the announcement is Mistral's own, measured on benchmarks the company selected, and no independent evaluation of Mistral Large 4 has been published. Mistral says it will release further benchmarks, details of the model's architecture and its post-training method alongside the weights.
Until the weights are out, Mistral says it is red-teaming the model in real-world settings with cybersecurity leaders, vetted partners and state authorities, which it says will use the same model with reduced moderation and expanded cyber capabilities.
