Mistral 3

About

Mistral 3 is announced as the next generation of open multimodal and multilingual AI models from Mistral AI. The release includes three state-of-the-art small, dense models (14B, 8B, and 3B) and Mistral Large 3, a highly capable sparse mixture-of-experts model with 41B active and 675B total parameters. All models are released under the Apache 2.0 license, emphasizing open-sourcing to empower the developer community. Mistral Large 3 is positioned as one of the best permissive open-weight models globally, demonstrating image understanding and strong multilingual performance, achieving a high rank on the LMArena leaderboard. The Ministral 3 series is designed for edge and local use cases, offering excellent cost-to-performance ratios and multimodal capabilities across base, instruct, and reasoning variants. The announcement also highlights strategic partnerships with NVIDIA, vLLM, and Red Hat to ensure faster and more accessible deployment of Mistral 3 models, optimized for various hardware from data centers to edge devices. Mistral AI also offers custom model training services for tailored enterprise solutions. The models are available on various platforms including Mistral AI Studio, Amazon Bedrock, Hugging Face, and more.