Mistral unveiled its inaugural reasoning-centric artificial intelligence (AI) models on Tuesday. Named Magistral, this large language model (LLM) comes in two versions—Small and Medium. The Magistral Small is an open-source model, while the Medium variant is tailored for enterprise applications and is a closed model. Both versions are designed for multi-step reasoning and exhibit a transparent chain-of-thought (CoT) capability. The French AI startup emphasized that Magistral is intended for purposes such as research, strategic planning, and data-driven decision making. Users can also preview the reasoning model through the company’s Le Chat platform.
Mistral’s Magistral Small Is Available With Apache 2.0 Licence
In a post on Mistral’s newsroom, the Paris-based AI company announced details regarding the two Magistral models. The open-weight Magistral Small can be downloaded and deployed through Mistral’s Hugging Face listing under the Apache 2.0 licence, making it suitable for both academic and commercial applications.
Conversely, Magistral Medium is a proprietary model available through Amazon SageMaker. The model is also set to be hosted on platforms such as IBM WatsonX, Azure AI, and the Google Cloud Marketplace in the near future. A preview of this model can be accessed via Le Chat or through an API on La Plateforme.
In terms of specifications, Magistral Small is equipped with 24 billion parameters, though the company has not revealed the parameter count for the enterprise version.
Regarding the capabilities of Magistral Medium, the model reportedly achieved a score of 73.6 percent on the AIME2024, which is competitive with DeepSeek-R1. In comparison, the Small variant attained a score of 70.7 percent, according to Mistral.
Both variants possess native reasoning abilities, with the chain-of-thought functionality supporting multiple global languages. Mistral confirmed that Magistral can reason effectively in English, French, Spanish, German, Italian, Arabic, Russian, and Simplified Chinese.
Like many reasoning models, Magistral is adept at performing structured calculations, programmatic logic, and managing decision trees and rule-based systems.
Mistral indicated that enterprises and users in critical sectors such as finance, healthcare, government, and law will benefit from the traceable reasoning feature of the models. This functionality allows users to audit the logical steps taken to arrive at a specific response, which is particularly important for sensitive issues.
