On June 10, Mistral AI introduced Magistral, its first model tailored specifically for complex reasoning tasks. Built with enterprise needs in mind, the model aims to assist in everything from strategic forecasting and operational planning to multifactor analysis and decision automation.

Unlike general-purpose AI systems, Magistral is designed to deliver clear, verifiable, and multilingual logical reasoning. Mistral describes it as a tool that supports transparent decision-making processes and handles logic-based scenarios such as modeling under constraints, scenario evaluation, and rule-driven systems.

What makes Magistral stand out is its ability to perform deep, multi-step reasoning while remaining interpretable to users. Its fine-tuning enables users to follow its logic path in multiple languages, making it suitable for international operations. The model performs reliably across English, Spanish, German, French, Italian, Russian, Arabic, and Simplified Chinese.

Beyond reasoning, Magistral also brings improvements to technical workflows. It is particularly valuable in areas like software architecture, data processing, and interface design — significantly boosting planning accuracy and development efficiency over models not optimized for logical consistency.

Mistral AI is offering two editions of Magistral:

• Magistral Small: An open-source model with 24 billion parameters, available for download via Hugging Face.
• Magistral Medium: A more capable version designed for enterprise applications, accessible through API on La Plateforme, on Le Chat, and also available via Amazon SageMaker.

In addition to the model launch, Mistral AI has released a detailed research paper outlining the training framework, reinforcement learning approach, evaluation benchmarks, and insights into the process of building a reasoning-focused AI model.