Mistral Small 4 – Fast, multimodal, reasoning AI in one open model

Summary: Mistral Small 4 is an open-weight AI model combining fast chat, deep reasoning, coding, and multimodal (text and image) capabilities. It uses a Mixture-of-Experts architecture with 119 billion parameters and a 256k context window, enabling configurable reasoning depth and efficient performance for diverse AI tasks.

What it does

Mistral Small 4 integrates instruction following, reasoning, coding, and multimodal understanding in a single system. Its Mixture-of-Experts design activates 4 of 128 experts per token to optimize compute efficiency.

Who it's for

It targets developers, enterprises, and researchers building assistants, coding agents, document analysis tools, or AI applications requiring reasoning and multimodal inputs.

Why it matters

It eliminates the need to switch between models by unifying multiple AI capabilities with configurable reasoning effort and long-context support.