Magistral Small 1.2

Magistral Small 1.2 is a 24-billion parameter reasoning model designed to balance high-level performance with hardware accessibility. Built with an integrated vision encoder, the model enables seamless analysis of both text and images, allowing it to handle complex, multi-step tasks. Its architecture incorporates specialized reasoning tokens that manage internal thought processes, ensuring that the model reflects on its outputs to improve accuracy before presenting a final response to the user.

The model demonstrates significant advancements in reasoning and utility, achieving notable performance gains on benchmarks such as AIME and LiveCodeBench. Through its design, it supports improved tool usage, including web search and code interpretation, while maintaining a natural and clear tone across multiple languages. Because of its efficient parameter count, the model is well-suited for local deployment on consumer hardware like high-end GPUs or laptops, offering a practical solution for developers who require powerful reasoning capabilities without relying on cloud-only infrastructure.