Mistral AI Magistral Release Notes

Mistral AI has announced Magistral, its first reasoning model designed for domain-specific, transparent, and multilingual reasoning. This release introduces a system capable of multi-step deliberation and analysis, addressing limitations in early thinking models such as lack of specialized depth and inconsistent reasoning across different languages.

Model Variants and Performance

Magistral is released in two distinct versions to balance accessibility and power:

  • Magistral Small: An open-source, 24B parameter model available under the Apache 2.0 license.
  • Magistral Medium: A more powerful enterprise-grade version.

In terms of benchmark performance on AIME2024, Magistral Medium achieved a score of 73.6%, rising to 90% with majority voting @64. Magistral Small followed with 70.7% and 83.3% respectively.

Technical Capabilities and Features

Transparent Reasoning and Chain-of-Thought

Magistral is fine-tuned for multi-step logic, providing a traceable thought process in the user's native language. Unlike general-purpose models, this approach improves interpretability and allows users to verify the logical steps leading to a conclusion.

Multilingual Dexterity

The model maintains high-fidelity reasoning across various global languages and alphabets. It is specifically optimized for languages including English, French, Spanish, German, Italian, Arabic, Russian, and Simplified Chinese.

High-Speed Inference in Le Chat

Through the "Think mode" and "Flash Answers" features in Le Chat, Magistral Medium achieves token throughput up to 10x faster than most competitors, enabling real-time reasoning and user feedback at scale.

Enterprise and Domain Applications

Magistral is designed for use cases where transparency and precision are critical, such as:

  • Regulated Industries: Providing auditability and traceable reasoning for legal, finance, healthcare, and government sectors to meet compliance requirements.
  • Systems and Software Engineering: Improving project planning, backend architecture, and data engineering through sequenced, multi-step actions involving APIs or external tools.
  • Business Strategy: Executing risk assessment, operational optimization, and data-driven decision making, including calculating optimal delivery windows under constraints.
  • Content Creation: Serving as a creative companion for coherent or eccentric storytelling and creative writing.

Availability and Deployment

Magistral Small is an open-weight model available for download via Hugging Face. Magistral Medium is available as a preview in Le Chat, via API on La Plateforme, and on Amazon SageMaker, with upcoming availability on IBM WatsonX, Azure AI, and Google Cloud Marketplace.

Sources

Related

  • Dispatch
    Mistral AI Le Chat Update July 2025

    Mistral AI has updated Le Chat with Deep Research, voice capabilities via Voxtral, multilingual reasoning via Magistral, Projects for organization, and advanced image editing.

  • Dispatch
    Mistral Small 4 release notes / what's new

    Mistral AI has released Mistral Small 4, a unified open-source model under Apache 2.0 that integrates reasoning, multimodal, and agentic coding capabilities into a single versatile architecture.

  • Dispatch
    Mistral Small 3.1 Release Notes

    Mistral AI has released Mistral Small 3.1, an Apache 2.0 licensed model featuring multimodal understanding, a 128k token context window, and performance that surpasses GPT-4o Mini and Gemma 3.

  • Dispatch
    Mistral Medium 3.5 and Vibe Remote Agents Release

    Mistral AI has released Mistral Medium 3.5, a 128B open-weight model, and introduced remote cloud-based coding agents in Mistral Vibe and a new Work mode in Le Chat for asynchronous, multi-step task execution.

  • Dispatch
    Mistral Large and Mistral Small Release

    Mistral AI has released Mistral Large, a flagship model with top-tier reasoning and multilingual capabilities, and Mistral Small, a model optimized for low-latency workloads.