- Magistral, from the French startup Mistral AI, is the first European AI model focused on step-by-step, multilingual reasoning.
- It is released in two versions: Small (open source, 24.000 billion parameters) and Medium (for businesses), with high performance in international benchmarks.
- Magistral stands out for its transparency, auditability, and support for multiple European languages, which sets it apart from American and Chinese models.
- The model seeks to place Europe at the forefront of global AI, promoting technological sovereignty and openness through public licenses.
In recent days, Europe has fully entered the international race for artificial intelligence thanks to the launch of Magistral, the first AI model focused on advanced reasoning developed on the continent. Behind this milestone is the French startup Mistral AI , which, with its proposal, seeks to offer a robust and open alternative to US dominance and China's growing strength in this technological field.
Until now, innovation in AI models with step-by-step reasoning capabilities had been led by companies like OpenAI, Google, and DeepSeek in China , leaving Europe in a clear position of technological dependence. With Magistral, Mistral AI is positioning itself as the European voice in this strategic sector, receiving support from both the French government and major investment funds, and valuing it at around $6.200 billion.
Magistral's announcement was made during London Tech Week, one of the leading international technology showcases, and represents a milestone for the startup, as well as sending a clear message about Europe's capabilities to develop next-generation artificial intelligence with values of transparency, collaboration and privacy.
What makes Magistral different from other models?
Magistral 's main feature is its focus on explicit reasoning , enabling it to solve complex problems through a logical thought process, similar to how humans think. Unlike traditional LLMs that rely solely on text prediction, this new model can follow rules, make structured decisions, and even justify its answers step by step.
There are two versions: Magistral Small , which has 24.000 billion parameters and has been released under an open-source license on Hugging Face; and Magistral Medium , geared towards the enterprise sector with greater computational capacity and available through APIs and cloud platforms such as IBM WatsonX, Azure AI, or Amazon SageMaker. Thanks to this flexibility, it can be integrated into both on-premises deployments and advanced cloud services.
In terms of performance , Magistral Medium achieved 73,6% accuracy in the international benchmark AIME2024 (and 90% using majority voting), while Small achieved 70,7%. These results put the European model on par with the powerful Chinese DeepSeek R1 and close to the leading global benchmarks released by OpenAI and Google, although it still falls short of their most recent versions.
Open and multilingual innovation
Magistral's strengths include its open nature : unlike the restrictive policies of OpenAI or Google, Magistral allows public access to the Small model's weights, making it easier for developers, institutions, and public bodies to audit, adapt, and even customize the AI according to their needs. This is especially relevant in regulated sectors or environments that require technological transparency.
Furthermore, Magistral has been trained to reason and generate responses directly in multiple European languages , such as French, Spanish, German, and Italian, without relying on English as an intermediary language. This breaks down the language barrier of traditional models and makes it a particularly versatile tool for adoption across the European Union, delivering native and accurate results in various alphabets and linguistic contexts.
On the other hand, the company emphasizes the auditability and traceability of the reasoning process: each decision can be followed line by line, a key feature in fields such as law, engineering, health or public administration.
Applications and competitive context
The French company designed Magistral with applications requiring advanced reasoning in mind: structured diagnostics, programmatic logic, decision trees, and rule-based systems. Thanks to its Flash Response mode, the model can deliver up to 10 times faster results on certain tasks using its chatbot, Le Chat, which represents the European alternative to ChatGPT.
In the international arena, Magistral joins the battle of so-called reasoning models , initially led by OpenAI in 2024 and soon followed by DeepSeek in China. Although it doesn't yet incorporate multimodal vision or audio capabilities like its more powerful rivals, its open-source approach makes it a very attractive option for developers and public or private companies seeking technological independence.
Behind this project are former Meta and Google engineers, such as Arthur Mensch and Guillaume Lample, who bring a proven track record of experience in cutting-edge artificial intelligence design.
The emergence of Magistral marks a new stage for European technological sovereignty in AI : a commitment to transparency, openness, and adaptation to the continent's linguistic and cultural nuances, offering a clear alternative to American and Asian giants. While there is still a way to go to match the world leaders in terms of absolute power, this advance places Europe at the forefront of developing AI reasoning models , opening up new possibilities for companies, public bodies, and citizens who demand more accessible and controllable technological solutions.
