Mistral Small is an open-weight model in the Mistral Small family that brings together the reasoning strength of Magistral, the multimodal understanding of Pixtral, and the agentic coding capabilities of Devstral into a unified system. This convergence allows one model to handle complex analysis, software development, and visual interpretation within a single workflow, making it well-suited for applications that need to move fluidly between text, images, and tool-driven tasks. The model accepts text and image inputs and produces text outputs, with native function calling and JSON output that support reliable agentic behavior.

Building on the lineage of Mistral Small 3.1, a 24-billion-parameter dense architecture that was recognized for fitting comfortably on a single high-end consumer GPU and for extending context handling to the cataloged API limit tokens, the newer Mistral Small 4 generation scales that foundation to a 119-billion-parameter design with a 262k context window. Practical strengths highlighted for the family include fast-response conversational agents, low-latency function calling, long document understanding, multilingual coverage across dozens of languages, and fine-tunable specialization for subject-matter domains. The combination of open weights, vision capability, and strong reasoning makes it a flexible fit for organizations that need a deployable, privacy-conscious model capable of both routine assistance and more demanding analytical work.