From the archive
Mistral releases its first 7B language model
Mistral 7B brings downloadable base and chat-oriented weights under Apache 2.0, with local and cloud deployment options.
Mistral AI released Mistral 7B, a 7.3-billion-parameter language model under the Apache 2.0 license. The release includes a base model and an instruction-tuned version for chat.
A smaller model to run and adapt
The downloadable weights give developers a route to use the model locally or on their own cloud infrastructure. Mistral provides a reference implementation and identifies vLLM as an option for serving requests.
Its attention design aims to reduce the memory and processing overhead of handling longer sequences. The model can also be fine-tuned for a more specific task.
Base weights and chat weights
The base release provides a starting point for adaptation. The instruction-tuned version demonstrates chat behavior, but the launch notes say it does not include a moderation mechanism. That matters when choosing how to expose it in an application.
The Mistral AI profile covers its model families and access routes.