Mistral AI's groundbreaking reasoning model, trained with scalable reinforcement learning to push the boundaries of artificial intelligence.
Magistral pushes the limits of what's possible with pure reinforcement learning training of large language models.
Explores the limits of pure reinforcement learning training without relying on existing implementations.
Simple method to force the reasoning language of the model for enhanced control and performance.
Maintains or improves multimodal understanding, instruction following, and function calling capabilities.
Magistral is backed by cutting-edge research and developed by a world-class team of AI researchers.
Our work on Magistral has been submitted to arXiv and represents a significant advancement in AI reasoning capabilities.
Read Paper on arXivAccess Magistral models through HuggingFace and start exploring the future of AI reasoning today.
Experience Magistral's reasoning capabilities instantly through our interactive chat interface.
Start ChattingOpen-source model available under Apache 2.0 license with cold-start data from Magistral Medium.
Download Open SourceAdvanced reasoning model trained on top of Mistral Medium 3 with reinforcement learning.
Access on HuggingFace