Mistral, a young tricolor shoot founded in 2023, revealed masterful, his first family of AI models specializing in reasoning. Like Gemini 2.5 Pro (Google) or Claude Opus 4 (Anthropic), these models are designed to solve complex problems following a step-by-step logic, especially in mathematics and physics.
An AI that reflects step by step
Two variants are offered: Masterful Small, 24 billion parameters with free online on the Hugging Face platform under Apache 2.0 license; And masterful medium, more muscular version but still in the test phase, accessible via the Mistral cat interface, its API, as well as via partner clouds.
Subscribe to WorldOfSoftware
In its announcement, the startup insists on the usefulness of its new models for cases of concrete use in business: “ structured calculations, programmatic logic, decision trees or even rules based on rules “, With an ability to” Provide an explicit trace of reasoning, in the language of the user ». Magistral wants to be particularly suitable for multilingual environments, with a supported support for Arabic, Russian, Italian or even simplified Chinese.
Despite his ambitions, masterful Medium is struggling to compete with the best in the sector. On several reference benchmarks (GPQA Diamond, AIME or LiveCODEBENCH), the model remains behind its American competitors. But Mistral advances another argument: speed. “” Ten times faster than the others in the cat “, Ensures the company.
The launch of masterful is part of a broader movement of acceleration of AI capabilities in Europe. During a common event at the Living Room in Paris, Jensen Huang, the CEO of Nvidia, revealed a major project: more than 20 “IA Factory” should be created in Europe within two years, some even exceeding a gigawatt of power.
Among the partnerships highlighted: the one knotted with Mistral. The two companies will create Mistral Compute together, a cloud infrastructure powered by 18,000 Grace Blackwell chips in Nvidia, hosted in a Data Center in Essonne. This sovereign cloud will then be extended elsewhere in Europe.
An initiative which obviously received the supported support of Emmanuel Macron, which sees it as a centerpiece of the French strategy of technological autonomy. “” I want to encourage all large companies and startups to join this team “He said.
Nvidia is currently pushing a global strategy: providing each country with the means to develop a local AI, based on its own data. “” AI begins with the data. Do not export them, Buy a machine instead “Launched Jensen Huang. And if possible an Nvidia machine, of course! Its objective is to democratize access to advanced AI, including for the smallest structures.
🟣 To not miss any news on the Geek newspaper, subscribe to Google News and on our WhatsApp. And if you love us, .
