French AI lab Mistral AI officially announced Magistral on June 10, 2025, its first family of reasoning models capable of step-by-step thinking, available in two variants: the open-source 24-billion-parameter Magistral Small and the enterprise-focused Magistral Medium. Magistral Medium scored 73.6% accuracy on the AIME2024 mathematics benchmark, rising to 90% with majority voting. The new model excels in multilingual reasoning, supporting English, French, Spanish, German, Italian, Arabic, Russian, and Simplified Chinese, distinguishing itself from competitors that primarily reason in English or Chinese.
The Magistral model family is specifically fine-tuned for transparent reasoning, allowing users to follow and verify the thought process. With new "Think mode" and "Flash Answers" features in Le Chat, Magistral Medium reportedly achieves 10x faster token throughput than most competitors, enabling real-time reasoning. The model is suited for various enterprise applications, including legal research, financial forecasting, software development, and creative storytelling, where transparency and precision are critical.
Magistral Small is available for self-deployment under the Apache 2.0 license on Hugging Face, while Magistral Medium can be tried in preview on Le Chat or via API on La Plateforme. Magistral Medium is also available on Amazon SageMaker and will soon appear on IBM WatsonX, Azure AI, and Google Cloud Marketplace. Mistral AI, founded in 2023 and backed by over $1.24 billion in funding, enters the reasoning AI model market with this release, competing with similar offerings from OpenAI, Google, and DeepSeek.
Sources:
1.
Kyle Wiggers
2.
Carl Franzen
3.
4.
Ryan Browne
