Mistral releases a pair of AI reasoning models

French AI Lab Mysterial is entering the logic AI Model Game.

Tuesday morning, Mistral Declaration Magistral, the first family of its logic models. Like other reasonable models – such as OpenA O3 And Google Gemini 2.5 Pro -Works through step -by -step problems for advanced continuity and reliability on topics such as Magistical Mathematics and Physics.

Magistral comes in two flavors: magistral small and magistral medium. Magistrals are 24 billion parameters in small size, and the hug of the AI ​​Dev platform is available for downloads under a permit Apache 2.0 license. (Parameters are the internal components of a model that guide its behavior) Magistral medium, more capable model, Mistral’s LE chat chatboat platform and company API, as well as third -party partner cloud.

“[Magistral is] Structural calculations and programmatic arguments are suitable for the use of extensive enterprise on trees and rules-based systems, “Mistral wrote in a blog post.”[The models are] Fine tune for multi-step log

Established in 2021, Myster is a Frontier Model Lab that creates a range of AI-driven services with the aforementioned LE chat and mobile applications. It is supported by zealous investors such as the General Catalist and has collected € 1.1 billion (about $ 24 billion) today.

Despite its powerful resources, the Myster is lagging behind in other top AI labs in certain cases, such as rational models. The magistral does not seem to be a particularly competitive release, is, judging by Mistral’s own criteria.

In GPQA Diamond and AIME, tests that do the physics, mathematics, and science skills, magistral medium under performing Gemini 2.5 Pro and anthropic’s clad oopas 4. Magistical Medium also fails to exceed Gemie 2.5 Pro in LiveCodbench, Benmark, a popular programming Benmark.

Perhaps that is why Mistral touches the magistral energy in his blog post. Magistral Le Chat, Mistral claims the speed of the contestants “10X” and supports various languages ​​including Italian, Arabic, Russian and simplified Chinese.

“Based on our flagship models, designed for magistial research, strategic plan, operational optimization and data-driven decisions,” the company wrote in its post, “Risk evaluation and modeling with multiple reasons, or calculating favorable distribution window under constraints.”

Mistral is a “Vibe Coding” client, the Mistral Code is released after the launch of the magistrate. A couple of weeks before, Mistral A number of coding-centric models have been launched and corporate-centric chattabot service has Roll Out out, providing equipment like AI agent manufacturer and integrates Mistral models with third-party services like Gmail and SharePoint.