Mistral Medium 3 is a proprietary mid-sized language model released by Mistral AI on May 7, 2025. It is designed to balance frontier-class reasoning and knowledge with high cost-efficiency and flexible deployment options for enterprise environments. The model features multimodal capabilities, enabling it to process and understand both text and visual inputs, such as images, charts, and long documents.
The model is optimized for professional use cases, demonstrating particular strength in coding, STEM-related reasoning, and technical data analysis. Mistral AI reports that Medium 3 delivers performance comparable to significantly larger frontier models, such as Claude 3.7 Sonnet, while operating at a substantially lower cost and infrastructure footprint. It is engineered for efficient single-node inference, allowing it to be deployed in private environments using as few as four H100-class GPUs.
With a 128,000-token context window, Mistral Medium 3 supports long-context applications including document summarization and complex data extraction. While other models in the Mistral 3 family, such as Mistral Large 3, were released with open weights, Mistral Medium 3 is a closed-source offering available via managed APIs and enterprise cloud platforms.