French AI startup Mistral has unveiled its Mistral 3 family of open-weight models, including a large frontier model with multimodal and multilingual capabilities and nine smaller models optimized for offline, cost-efficient deployment. The launch aims to provide enterprise clients with alternatives to large closed-source models, allowing customization that balances performance, efficiency, and accessibility.
Mistral Large 3 features a granular Mixture of Experts architecture with 41B active parameters and 675B total parameters, supporting a 256k context window for complex tasks such as document analysis, coding, content creation, and workflow automation. The nine smaller models, dubbed Ministral 3, range from 3B to 14B parameters and include Base, Instruct, and Reasoning variants. These models can run on a single GPU, enabling deployment on laptops, on-premise servers, robots, and edge devices.
The company has also begun integrating its models into physical systems, collaborating with Singapore’s HTX on robotics and cybersecurity, German startup Helsing on drones, and Stellantis on in-car AI assistants. Co-founder Guillaume Lample emphasized that reliability and independence from proprietary APIs are critical for enterprise adoption.
Founded by former DeepMind and Meta researchers, Mistral has raised $2.7 billion to date at a $13.7 billion valuation, positioning it as a key European contender in the open-weight AI market.
