Mistral Small 3.2

Mistral Small 3.2 is a 24B open-weight instruct model released in June 2025 under Apache 2.0. It is a refinement of Mistral Small 3.1, improving instruction following, reducing repetitive or infinite generations, and making function calling more robust while keeping the same efficient footprint. The model supports multimodal understanding through a built-in vision encoder, function calling for agentic workflows, and structured output for deterministic applications. Its 128K token context window enables document analysis and retrieval-augmented generation at production scale. Mistral Small 3.2 is well suited to virtual assistants, domain-specific fine-tuning, and lean deployments, running on a single high-end consumer GPU or a Mac with 32GB of memory. It serves as a capable open-weight alternative to compact proprietary models such as GPT-4o mini.