On Tuesday, French AI startup Mistralannounced the launch of its new model family, "Magistral."
These models are specifically engineered for complex, stepped logical reasoning, a move that places Mistral in direct competition with industry leaders like OpenAI and Google.
Speaking at London Tech Week, Mistral CEO Arthur Mensch stated that the new "Magistral" models possess a unique ability to "think in multiple languages."
With a special focus on European languages, the models stand apart from many current systems often specialized in English or Chinese.
The company also noted support for other languages, including Arabic, offering a powerful tool for linguistically diverse markets.
Two "Magistral" Releases to Meet Diverse Needs
Mistral has introduced two models within the "Magistral" family:
The first is "Magistral Small," an open-source model with 24 billion parameters, released under the Apache 2.0 license.
This approach allows developers and researchers to access and modify its advanced reasoning capabilities without prohibitive costs.
It also requires reasonable computing resources to run, such as a single RTX 4090 GPU or a MacBook with 32GB of RAM after appropriate quantization.
The second release, "Magistral Medium," is a more powerful model primarily aimed at the enterprise sector and commercial applications.
It is currently available for preview through the company's chat platform, "Le Chat," and the Mistral API, as well as via third-party cloud partners.
Magistral's Features and Capabilities
One of the standout features Mistral has emphasized in the "Magistral" models is the transparency of the reasoning process.
The models are designed to display the logical chain of steps they follow to reach a conclusion, allowing users to track, understand, and verify the validity of their outputs.
In terms of performance, "Magistral Medium" has shown promising results on standard benchmarks. It scored 73.6% on the 2024 AIME (American Invitational Mathematics Examination) with a single attempt, a figure that rises to 90% with majority voting over 64 attempts.
Meanwhile, "Magistral Small" scored 70.7% and 83.3% on the same tests, respectively.
The company also claims that "Magistral Medium" delivers responses up to ten times faster than competitors within the "Le Chat" platform, thanks to its "Flash Answers" technology.
Leveraging Magistral in Business and Industry
The "Magistral" models are designed for a wide spectrum of enterprise use cases. Their capabilities support structured calculations, programmatic logic, and even rule-based decision-support systems.
Mistral envisions its new models as a valuable tool for strategic research and planning, operational process optimization, and data-driven decision-making.
Potential applications extend to sensitive sectors such as law, finance, healthcare, and government, where the ability to trace logic and audit conclusions is critical.
Furthermore, the models are expected to provide strong support for software and systems engineering, as well as creative content development.
The launch of "Magistral" comes at a time when the reasoning model sector is seeing growing interest and significant investment from various companies.
Founded in April 2023, Mistral has garnered support from major firms like Microsoft and General Catalyst.
It enters the race aggressively, backed by rapid growth that has seen its sales contracts exceed $100 million in value just 15 months after launching its paid products, according to informed sources.
Mistral reaffirms its commitment to open AI development, believing that open-source models like "Magistral Small" encourage community-driven innovation and accelerate the pace of progress in this vital field.
