Mistral made a significant announcement by unveiling its newest artificial intelligence (AI) breakthroughs on a Tuesday, introducing Magistral, a revolutionary large language model (LLM), in two distinctive variants – Small and Medium. It is worth noting that while Magistral Small is an openly accessible model, the Medium variant caters to enterprise projects with its proprietary closed setup. This blog discusses about Mistral Releases Magistral.

Multi-step Reasoning

Both models showcase advanced capabilities in multi-step reasoning and exhibit a transparent chain-of-thought (CoT) process. Emphasizing its versatile applications, the French AI company positioned Magistral as an essential tool for various domains like research, strategic planning, and decision-making based on data analytics. For those eager to explore its potential, Mistral offers a preview version of this powerful new reasoning model through its Le Chat platform.

Magistral Models

Additionally, Mistral’s Magistral Small model is now conveniently available for access under the Apache 2.0 license. The company disclosed this information via an informative newsroom publication where they officially presented the distinctive characteristics of both Magistral models. Users can easily obtain Magistral Small, an open-weight LLM, by downloading it from Mistral’s Hugging Face repository seamlessly. This model has been designed to cater to a wide array of applications, spanning from academic research to commercial projects. This article discusses about Mistral Releases Magistral.

Magistral Medium

On the flip side, Magistral Medium represents a more advanced and proprietary version of the model, accessible through platforms such as Amazon SageMaker. Furthermore, it is set to be rolled out on renowned cloud platforms like IBM WatsonX, Azure AI, and Google Cloud Marketplace in the near future. Interested users can also familiarize themselves with a preview of the model through Le Chat or by leveraging its API on La Plateforme.

Magistral Small Variant

Delving deeper into technical specifications, the Magistral Small variant boasts an impressive 24 billion parameters, although the company has reserved the disclosure of the enterprise edition’s parameters. Moving on to Magistral Medium’s performance metrics, it notably achieved a commendable score of 73.6 percent on the AIME2024 evaluation, drawing parallels to the esteemed DeepSeek-R1. Conversely, the Small variant did not fall short either, earning a respectable score of 70.7 percent, as confirmed by the representatives at Mistral.

Both models flaunt native reasoning capabilities across a spectrum of languages, facilitating coherent chain-of-thought processes across global languages including English, French, Spanish, German, Italian, Arabic, Russian, and Simplified Chinese. Moreover, Mistral underscores the utility of Magistral in handling structured calculations, programmatic logic, decision trees, and rule-based systems, aligning with the standard functionalities of contemporary reasoning models.

Summary

Mistral remains dedicated to ensuring that experts and enterprises in pivotal sectors such as finance, healthcare, government, and law are equipped with traceable reasoning within the models, enabling them to scrutinize the logical pathways taken to produce responses. This scrutiny can be particularly critical for auditing responses that are distinguished by their sensitivity, underscoring Mistral’s commitment to transparency and reliability in its groundbreaking AI offerings.

LEAD GENERATION SERVICES APPSREAD