Fast, open-source and secure language models. Facilitated specialisation of models on business use-cases, leveraging private data and usage feedback.

Our models strike an unmatched latency to performance ratio, and achieve top-tier reasoning performance on all common benchmarks. We designed our models to be as unbiased and useful as possible, providing full modular control over moderation. We have shipped the most capable open models to accelerate AI innovation. Through our own independence, our endpoints and platform are portable across clouds and infrastructures to guarantee the independence of our customers.

Our optimized models can be deployed and managed where you need them, where your data is, maintaining the level of application hermeticity you require. We offer unique levels of customisation and control, with full fine-tuning capacities and everything you need to connect our models to your business systems and data. 

We believe in the power of open technology to accelerate AI progress. That is why we started our journey by releasing the world’s most capable open-weights models, Mistral 7B and Mixtral 8×7B.