Mistral Medium 3.5 Model Card

This document is a model card published by Mistral AI governing the Mistral Medium 3.5 large language model (version v26.04), describing its technical specifications, intended use cases, safety posture, and licensing terms; it does not assert a bilateral contractual basis but instead functions as a technical disclosure and governance reference for deployers and developers. The card states the model is released under a Modified MIT License, is optimized for agentic and coding use cases with native multimodal (vision) capabilities, supports a 128k token context window, and is described as a frontier-class model available both via API and as downloadable open weights. The document discloses prohibited use categories including generation of malicious code, disinformation, content that facilitates violence or exploitation of minors, and use in weapons development, while noting the model incorporates safety training and system prompt override resistance; however, as open weights, downstream deployers bear primary responsibility for use-case-specific safety controls, creating a governance gap that the card does not fully resolve. The card engages with the EU AI Act framework by classifying the model as general-purpose AI and noting conformity assessment obligations that may apply to high-risk downstream deployments, while also referencing Mistral AI's usage policies at mistral.ai/terms; the open-weights release structure creates jurisdiction-dependent compliance obligations for deployers, particularly under the EU AI Act's provider and deployer distinction.