Ministral 8B

Mistral's compact model built for on-device and edge deployment

Free Language
Visit Product Page →
128000 context tokens
8B parameters
Mistral Research License license
Oct 2024 released

Ministral 8B is an 8 billion parameter model that Mistral AI released in October 2024 as part of “Les Ministraux,” a pair of small models aimed at on-device and edge deployment alongside the smaller Ministral 3B. It uses interleaved sliding-window attention to support a 128,000-token context window despite its small size, and Mistral positioned it against similarly sized models like Llama 3.1 8B and Gemma 2 9B, claiming better results on knowledge, reasoning, and multilingual benchmarks within that class.

Mistral released it under the Mistral Research License, which allows free use for research and testing but requires a commercial license for production deployment, unlike the fully open Apache 2.0 terms of Mistral 7B. It targets developers building applications for phones, laptops, and other constrained hardware, such as local assistants, translation, and lightweight agents, where running a full-size model isn’t practical.