Mistral AI
French company developing open-weight and commercial large language models for text, code, vision, and audio tasks.
📍 France 🇫🇷, Paris
Product overview
Mistral AI develops general-purpose and specialized large language models from Paris. Founded in 2023 by former DeepMind and Meta AI researchers, the company raised over €2.8 billion, reaching a €12 billion valuation at its September 2025 Series C led by ASML and NVIDIA. The model lineup spans multiple sizes and tasks. Mistral 7B and Mixtral 8x7B (mixture-of-experts) are released under Apache 2.0. Mistral Large (123B parameters), Mistral Medium 3, and Mistral Small 3.1 (24B) serve commercial workloads via API. Pixtral handles image understanding, Codestral targets code generation, and Voxtral processes audio. Mistral Large 3, released December 2025 with 675B total parameters, ranks among the top open-source models on LMArena. Le Chat provides a consumer and enterprise chat interface with web search, image generation, canvas editing, and deep research. La Plateforme offers developers API access, fine-tuning, and agent-building tools. Mistral Vibe is a CLI tool for AI-assisted coding. Enterprise clients include HSBC, CMA CGM (€100 million partnership), and Safran. Models are also available through Azure, AWS, Google Cloud, and NVIDIA NIM. Mistral Compute, planned for 2026, will add sovereign European GPU infrastructure with 18,000 NVIDIA Blackwell chips. KEY FEATURES: - Open-weight models (Apache 2.0) alongside commercial API across text, code, vision, and audio - La Plateforme API with free experimentation tier, fine-tuning, and agent-building tools - Le Chat assistant with search, image generation, canvas editing, and deep research - Models distributed on Azure, AWS, Google Cloud, NVIDIA NIM, and Hugging Face - Sovereign European compute infrastructure (Mistral Compute) planned for 2026