Mistral Unveils Magistral: Groundbreaking AI Models for Transparent Reasoning and Multi-Language Support

Mistral unveiled its initial reasoning-oriented artificial intelligence (AI) models on Tuesday. Named Magistral, this large language model (LLM) comes in two versions—Small and Medium. Magistral Small is an open-source model, while the Medium version is a closed model aimed at enterprise users. Both of these models excel in multi-step reasoning and provide a clear chain-of-thought (CoT). The French AI startup emphasized that Magistral is tailored for research, strategic planning, and data-informed decision making. Users can also test a preview version of the reasoning model through the company’s Le Chat platform.

Mistral’s Magistral Small Is Offered Under Apache 2.0 License

In a news release post, the Paris-based AI organization disclosed the introduction of the two Magistral models. The open-weight LLM, Magistral Small, is available for download and deployment via Mistral’s Hugging Face listing under the Apache 2.0 license. This model can be utilized for both research and commercial applications.

Conversely, Magistral Medium is a proprietary model that can be accessed through Amazon SageMaker. It is also set to be available on platforms like IBM WatsonX, Azure AI, and Google Cloud Marketplace shortly. A preview of this model can be experienced in Le Chat or through an API on La Plateforme.

Regarding technical specifications, Magistral Small consists of 24 billion parameters. The parameters for the enterprise version, however, have not been revealed by the company.

Turning to the capabilities of Magistral Medium, it reportedly achieved a score of 73.6 percent on the AIME2024, comparable to DeepSeek-R1. In contrast, the Small variant is claimed to have reached a score of 70.7 percent, as stated by Mistral.

Both models are capable of native reasoning, with the CoT functioning in several global languages. The company claims that Magistral can perform reasoning in English, French, Spanish, German, Italian, Arabic, Russian, and Simplified Chinese.

Like many reasoning models, Magistral is also equipped to manage structured calculations, programmatic logic, decision trees, and rule-based systems.

Mistral asserts that businesses and users operating in critical sectors such as finance, healthcare, government, and law will benefit from traceable reasoning, enabling them to verify the logical processes leading to specific responses. This feature permits users to audit responses that are particularly sensitive, as mentioned by the company.