Magistral Small 1.2: An Enhanced Multimodal Reasoning Model

Magistral Small 1.2, developed by Mistral AI, is a 24 billion parameter model that significantly upgrades its predecessor, Magistral Small 1.1. This version introduces multimodality with a vision encoder, allowing it to process and reason based on both text and image inputs. It is specifically designed for long chains of reasoning traces before providing an answer, making it suitable for complex problem-solving.

Key Capabilities & Features

Multimodal Reasoning: Integrates vision capabilities to analyze images and reason from visual content.
Enhanced Reasoning: Capable of detailed, step-by-step reasoning processes, encapsulated by [THINK] and [/THINK] special tokens.
Multilingual Support: Supports dozens of languages, including English, French, German, Japanese, Chinese, and Arabic.
Optimized Performance: Delivers significantly better performance than Magistral Small 1.1 across various benchmarks, including AIME24, AIME25, GPQA Diamond, and Livecodebench.
Extended Context Window: Features a 128k context window, with good performance expected up to 40k tokens.
Improved Output Quality: Offers better LaTeX and Markdown formatting, shorter answers for easy prompts, and reduced likelihood of infinite generation loops.
Apache 2.0 License: Provides an open license for both commercial and non-commercial use.

When to Use This Model

Magistral Small 1.2 is ideal for applications requiring robust reasoning, especially those involving multimodal inputs. Its ability to handle complex logical steps and its broad language support make it versatile for various tasks. The model's efficient design allows for local deployment on hardware like an RTX 4090 or a 32GB RAM MacBook once quantized, offering flexibility for developers.