Mistral AI has launched a public preview of Mistral Large 4, a 1 trillion-parameter natively multimodal model with 52 billion active parameters, and said it will release the open weights by the end of October, the company said on 6 October 2026.

The preview API is available on Mistral Studio. Mistral said the model was trained from scratch on 3,800 NVIDIA Grace Blackwell GPUs in its own European datacentres, and that the preview is served on the same infrastructure. It will be offered in multiple regions, including a European deployment that Mistral operates end to end under European law.

Until the weights are released, Mistral said it is red-teaming the model with cybersecurity leaders, vetted partners and state authorities, who will access it with reduced moderation and expanded cyber capabilities.

On the Artificial Analysis Cyber Index the company said the model ranks among the top five globally, scores 82% on a test that asks a model to reproduce and then patch a real open-source vulnerability, and solves 93% of the challenges in Cybench. Mistral said several closed models score near zero on the reproduction test because they refuse the task.

A significant share of the training data was multilingual, spanning more than 160 languages, according to the company. Mistral said it would share further details on architecture, benchmarks and post-training methodology as it works towards the weights release. The benchmark figures are the company's own.


Share this post
The link has been copied!