Mistral Small 3.2 is a 24-billion parameter multimodal model released by Mistral AI on June 20, 2025. As an iterative update to the Mistral Small 3.1 series, it is optimized for high-precision tasks and provides a balance between computational efficiency and advanced reasoning capabilities. The model is released under the Apache 2.0 license, facilitating open use across diverse applications.
A significant feature of the 3.2 version is its multimodal capability, allowing the model to process and reason over both images and text simultaneously. This enables the model to perform complex tasks such as document understanding, visual question-answering, and image-grounded content generation. It supports a context window of 128k tokens, which is suitable for handling long-form documents and extensive conversation histories.
Technical refinements in Mistral Small 3.2 focus on enhancing instruction following and significantly reducing repetition errors. According to internal benchmarks, the model demonstrates improved accuracy in following precise user prompts and a reduction in "infinite loop" generation issues. Additionally, the model utilizes a more robust function-calling template designed for reliable structured API interactions and enterprise tool integration.