Magistral Small

Compact Transparent AI for Reasoning & Logic

What is Magistral Small?

‍Magistral Small is a 24-billion-parameter, open-source reasoning model from Mistral AI, focused on delivering clear, stepwise, and traceable outputs for technical, business, and regulated domains. Building on the Mistral Small 3.1 architecture, it incorporates supervised fine-tuning and reinforcement learning from its Medium sibling, enabling trusted, auditable logic chains in over 20 languages. Its efficient design allows for deployment on a single RTX 4090 or modern Mac, making it accessible for edge, local, or self-hosted use.

Key Features of Magistral Small

Stepwise & Transparent Reasoning

Produces interpretable, auditable outputs with explicit chain-of-thought logic using special “[THINK]...” tokens for exam-style or technical workflows.

Multilingual Dexterity

Excels in high-fidelity logical reasoning across dozens of languages, including English, French, Spanish, German, Italian, Russian, Chinese, Arabic, and more.

Efficient, Local & Edge Inference

Runs on accessible consumer hardware, single GPUs or 32GB RAM MacBooks, ideal for privacy-oriented or regulated sectors, as well as scalable cloud deployment.

Open & Fine-Tuned

Released under Apache 2.0 for easy integration, fine-tuning on enterprise traces, and continuous improvement by the community.

Reduced Repetition & Robust Outputs

Delivers concise, relevant, and stable responses with improved formatting for technical tasks like LaTeX or Markdown.

Use Cases of Magistral Small

Delivers clear, step-by-step logic for compliance, legal, and regulatory reviews, ensuring transparent, auditable conclusions for business and technical operations.

Performs complex data-backed scenario modeling, supply chain predictions, and risk analysis tasks using interpretable multistep logic.

Enhances system design, code review, planning, and debugging via readable reasoning traces and agentic integration with external tools.

Delivers logic-based workflow outputs for chatbots, automation, and cross-lingual use in business, government, healthcare, and education.

Powers technical writing, creative storytelling, and knowledge base expansion with explicit, interpretable chains of thought.

Magistral Small

vs

Comparable Models

Feature	Magistral Small	Magistral Medium	Mistral Small 3.1	Claude Sonnet 3.7
Reasoning Logic	Stepwise, traceable	Transparent, larger	Standard CoT	Standard CoT
Auditability	High, stepwise	High, enterprise	Basic	Basic
Speed	150 tokens/sec	10x typical models	~150 tokens/sec	~80 tokens/sec
Languages	20+ (high-fidelity)	20+	21+	30+
Platform	Local/cloud/edge	Cloud/API	Local/cloud/edge	Cloud
License	Apache 2.0 open	Enterprise	Apache 2.0 open	Proprietary