French AI Lab Mistral is caught up in reasoning in AI model games.
On Tuesday morning, Mistral announced Magistral, the first family of the reasoning model. Like other inference models, such as Openai’s O3 and Google’s Gemini 2.5 Pro – work through problems step by step to improve consistency and reliability across topics such as mathematics and physics.
Guctistral comes in two flavors. The size of Gustistral Small is 24 billion parameters and can be downloaded from the AI Dev platform under the Apache 2.0 license to hug Face. (Parameters are internal components of the model that guide its behavior.) It can be found in previews of Mistral’s more capable model, the LE chat chatbot platform and company APIs, as well as third-party partner clouds.
“(Therapeutics) are suitable for a wide range of corporate use cases, from structured calculations and programmatic logic to decision trees and rule-based systems,” Mistral wrote in a blog post. “(The model) is fine-tuned for multi-step logic, improving interpretability and providing a thought process that can be tracked in the user’s language.”
Founded in 2023, Mistral is a frontier model lab that builds services equipped with a variety of AI, including the aforementioned LE chat and mobile apps. It is backed by venture investors like General Catalyst, and has raised more than 1.1 billion euros (approximately $1.24 billion) to date.
Despite its formidable resources, Mistral has lagged behind other major AI labs in certain areas, such as the development of inference models. Judging by Mistral’s own benchmarks, it appears Magistral is not a particularly competitive release either.
In GPQA Diamond and AIME, Gemistral Medium underperforms the Gemini 2.5 Pro and Claude Opus 4 in tests that evaluate the physics, mathematics and science skills of models.
Perhaps that’s why Mistral promotes Magistral’s other strengths in blog posts. Magistral responds to the speed of its competitors at “10x” with Le Chat and Mistral’s claims, supporting a wide range of languages, including Italian, Arabic, Russian and simplified Chinese.
“Based on our flagship model, Magistral is designed for research, strategic planning, operational optimization, and data-driven decision-making,” the company writes, “whether to perform risk assessment and modeling on multiple factors or calculate the optimal delivery window under constraints.”
The release of Mistral comes after Mistral debuted Mistral Code, a “vibe coding” client. A few weeks ago, Mistral launched several coding-centric models, deployed Le Chat Enterprise, a company-centric chatbot service that provides tools such as AI agent builders and integrates Mistral’s models with third-party services such as Gmail and SharePoint.