French AI lab Mistral has officially entered the competitive landscape of reasoning AI models with the launch of Magistral, its first family of reasoning models. These models are designed to handle step-by-step logic and improve consistency in tasks such as math, physics, and programmatic decision-making. This announcement positions Mistral alongside other major players like OpenAI’s o3 and Google’s Gemini 2.5 Pro in the reasoning AI space. Magistral is offered in two configurations: Magistral Small and Magistral Medium. Magistral Small is a 24-billion parameter model released under the permissive Apache 2.0 license and is available for download via Hugging Face. Magistral Medium, on the other hand, is a more powerful version available in preview through Mistral’s chatbot platform Le Chat, its API, and third-party partner cloud platforms. According to Mistral’s official blog, these models are tailored for complex enterprise use cases. This includes structured calculations, programmatic logic, rule-based systems, and multi-step reasoning tasks. Mistral notes that the models are fine-tuned to improve interpretability and traceability of logic steps in a user-friendly language. Mistral, founded in 2023, has quickly established itself as a formidable player in the AI sector. With backing from General Catalyst and over €1.1 billion (approximately $1.24 billion) raised to date, the company is focused on building innovative AI services, including its chatbot and mobile platforms. Despite its promising debut, Magistral Medium does not currently outperform competing models like Gemini 2.5 Pro or Claude Opus 4 on key benchmarks such as GPQA Diamond, AIME, and LiveCodeBench. These results indicate that while Mistral is making progress, the competition in reasoning models remains fierce. Nonetheless, the release of Magistral marks a strategic milestone in Mistral’s roadmap, signaling the company’s ambitions to play a major role in the future of AI reasoning and enterprise applications.