Mistral Small 4 is positioned as the next major release in the Mistral Small family, designed as a single system that brings together capabilities previously split across separate flagship models. It blends the reasoning strengths of Magistral, the multimodal understanding of Pixtral, and the agentic coding abilities of Devstral, allowing one model to move between complex analysis, software development, and visual tasks within the same workflow. This consolidation reflects a broader trend in the Mistral lineup toward unified systems that can serve diverse agentic and analytical use cases without swapping models mid-pipeline.
In benchmark comparisons across Mistral's catalog, Small 4 lands as the third-most intelligent model on Artificial Analysis's index, behind only Mistral Medium 3.5 and the top entry, and pairs that capability with leading-class output speed. With reasoning enabled it scores around 77% on GPQA Diamond and roughly 49% on IFBench, while its coding index sits in the upper tier among peers; the non-reasoning configuration shows lower but still usable scores across the same evaluations. The model also supports function calling and structured JSON outputs, runs on open weights for self-hosting, and benefits from a 256K-class context window, making it well suited for long-horizon agentic coding, document-heavy reasoning, and multimodal workflows where consistency across text and image inputs matters.
