French AI startup Mistral has officially entered the reasoning model arena with the launch of Magistral, Europe's first AI system capable of step-by-step logical problem-solving across multiple languages.
Unveiled on June 10, 2025, Magistral comes in two distinct versions. Magistral Small is a 24-billion-parameter model released under the permissive Apache 2.0 license, making it freely available for developers on Hugging Face. The more powerful Magistral Medium is available through Mistral's Le Chat platform and API services, targeting enterprise customers with premium performance.
"Historically, we've seen U.S. models reason in English and Chinese models reason in Chinese," explained Mistral CEO Arthur Mensch during London Tech Week. What sets Magistral apart is its ability to maintain high-fidelity reasoning across numerous European languages, including French, Spanish, German, Italian, as well as Arabic, Russian, and Simplified Chinese.
The model excels at mathematics, coding, and structured thinking tasks, making it suitable for applications in law, finance, healthcare, and engineering. Mistral claims Magistral Medium delivers answers at "10x" the speed of competitors through its new "Think" button and Flash Reasoning feature, processing up to 1,000 tokens per second.
Despite impressive capabilities, Magistral Medium does face stiff competition. On benchmarks like GPQA Diamond and AIME, it underperforms compared to Google's Gemini 2.5 Pro and Anthropic's Claude Opus 4. However, Mistral has positioned Magistral competitively on pricing, with output costs substantially undercutting OpenAI and Google while being significantly cheaper than Anthropic's flagship models.
The launch reinforces Mistral's dual strategy of supporting open-source development while monetizing premium enterprise offerings. Valued at $6.2 billion and having raised over $1 billion in funding, Mistral is widely considered Europe's best shot at developing a homegrown AI competitor to American and Chinese tech giants. With Magistral, the company founded by former DeepMind and Meta researchers continues its rapid pace of innovation, having released seven AI model upgrades this year alone.