Mistral Launches Magistral: First Open-Source Reasoning Model from Europe

Mistral AI has announced Magistral, its first reasoning model and the first open-weight reasoning model from a major European AI company. Released on June 10, Magistral comes in two sizes: Magistral Small, a 24B parameter model available under the Apache 2.0 license, and Magistral Medium, a larger version accessible through Mistral's API and La Plateforme. The model was trained using Mistral's own scalable reinforcement learning pipeline and excels in domain-specific, transparent, and multilingual reasoning. According to Mistral, Magistral generates output up to 10x faster than some proprietary models, though it acknowledges the model lags behind frontier competitors on certain standard benchmarks. The open-source release is significant for the AI community because reasoning models have been largely proprietary — OpenAI's o-series and Google's thinking models are closed-source. Magistral gives researchers and developers the ability to inspect, modify, and deploy a reasoning model on their own infrastructure. For MENA developers, the Apache 2.0 licensing means Magistral Small can be used commercially without restrictions, deployed on-premise for data-sensitive applications, and fine-tuned for Arabic language tasks.