Paris-based Mistral AI has released a public preview of Mistral Large 4, a multimodal model that processes text and images. Nicknamed “le Chonk,” the model has one trillion parameters and aims to challenge the dominance of closed-source models from American tech giants.
Mistral trained the model from scratch over two months. Training used 4,000 of Nvidia’s new Grace Blackwell GPUs.
Mistral is positioning the model as a foundation for enterprises and governments. Its open-weight design lets users download the model and run it on their own servers, offering an alternative to competitors’ closed ecosystems.
The model weights are scheduled for full public release on October 27. Before then, developers, cybersecurity firms, and government authorities will take part in a testing period.