Notebookcheck Logo

Mistral AI launches Magistral, its first AI capable of reasoning

Mistral AI adds reasoning capabilities to its lastest Magistral model. (Image source: Mistral AI)
Mistral AI adds reasoning capabilities to its lastest Magistral model. (Image source: Mistral AI)
Mistral AI has launched Magistral, a reasoning AI model that takes more time to think of answers to complex prompts while offering businesses traceability for compliance in finance, law, and medicine. Magistral provides businesses with an alternative to leading US and Chinese AI models.

Mistral AI, a French company, has unveiled Magistral, its latest artificial intelligence model capable of reasoning through problems. The reasoning model takes more time to research the problem, find information, and craft thoughtful replies than non-reasoning AI. Magistral also provides businesses with traceability for its reasoning, helping them meet strict compliance laws and regulations, especially in the fields of finance, law, and medicine.

Magistral provides businesses with an alternative to the US (OpenAI, Perplexity, Anthropic, Meta, and Grok) and Chinese (DeepSeek) AI models topping the LMArena AI Leaderboard charts. French President Emmanuel Macron has recently promoted Mistral AI as an example of sovereign technology that can keep his country from being beholden to others in the emerging AI era. President Macron has also announced the creation of France's own AI data center, a 30 to 50 billion Euro project, to maintain the country's standing in the AI world.

Mistral AI has released Magistral in two versions - a smaller 24-billion-parameter, open-source AI model called Magistral Small, and a larger Magistral Medium model for businesses and individuals who pay for access to the API or subscriptions to its web-based Le Chat chatbot. Access to Magistral Medium will also be rolled out to AI cloud platforms such as Microsoft Azure AI.

Magistral Medium has performance similar to DeepSeek R1, a Chinese reasoning AI model, while Magistral Small has naturally poorer performance. Magistral defaults to its Flash Answers mode in Le Chat, a mode where it produces answers very quickly to prompts. The company suggests it can process tokens of information up to ten times faster than the competition, but has not provided hard data. Nonetheless, most AI chatbot services on the web produce a quick answer within a second or two after being asked.

The Flash Answers mode does not provide detailed links or information backing up its answers, so users should use Think mode to gain access to that. The Think mode forces the AI to take more time to answer, while detailing each step it takes so that users can review the AI's work.

Readers who want to conduct research and modify the Magistral Small model available from Hugging Face on their own desktop PCs should make sure they have a powerful graphics card, like this Nvidia 5090 GPU sold on Amazon.

Mistral AI's Magistral Medium AI model has performance close to DeepSeek R1 across some common AI benchmarks. (Image source: Mistral AI)
Mistral AI's Magistral Medium AI model has performance close to DeepSeek R1 across some common AI benchmarks. (Image source: Mistral AI)
Mistral AI's Magistral Medium AI model performs poorly on the tough Humanity's Last Exam AI benchmark. Other top, competing AI models also perform poorly on this challenging exam. (Image source: Mistral AI)
Mistral AI's Magistral Medium AI model performs poorly on the tough Humanity's Last Exam AI benchmark. Other top, competing AI models also perform poorly on this challenging exam. (Image source: Mistral AI)
static version load dynamic
Loading Comments
Comment on this article
Please share our article, every link counts!
Mail Logo
> Expert Reviews and News on Laptops, Smartphones and Tech Innovations > News > News Archive > Newsarchive 2025 06 > Mistral AI launches Magistral, its first AI capable of reasoning
David Chien, 2025-06-18 (Update: 2025-06-18)