Microsoft is expanding its strategic partnership with Mistral AI by adding the frontier-class Mistral Medium 3.5 and OCR 4 models to Microsoft Foundry and integrating Medium 3.5 into Microsoft Copilot Studio as of July 21, 2026. This move brings Mistral’s latest multimodal and document-understanding capabilities into Microsoft’s AI platform layer, giving enterprises and regulated industries additional options for building and governing advanced applications across Foundry, Copilot Studio, and Azure environments. By leveraging Connected Apps in Google AI Mode, Microsoft aims to enhance user experience across its own platforms.
By positioning Mistral’s open-weight models inside a managed Azure ecosystem, the companies aim to combine frontier performance with operational control, deployment flexibility, and consistent tooling from cloud to customer-controlled infrastructure.
Mistral Medium 3.5 is described as a frontier-class multimodal model optimized for agentic and coding use cases, consolidating instruction following, reasoning, and software development into a single flagship system. The model is released as open weights under a permissive license, enabling organizations to deploy it in self-hosted, sovereign, or cloud-managed environments while still benefiting from Azure governance and monitoring.
Its dense architecture is paired with a large context window, supporting extended conversations, complex codebases, and long-running workflows typical of enterprise scenarios. Medium 3.5 accepts both text and images, supports function calling and tool use, and is designed to operate as the reasoning backbone for agents, structured outputs, and batch processing pipelines across Foundry and Copilot Studio solutions.
OCR 4 is delivered through Mistral Document AI in Microsoft Foundry, targeting high-fidelity, structured understanding of complex documents rather than simple text extraction. In addition, OCR 4 brings its 170-language coverage into Foundry to support global and mixed-language document workflows at scale. The service preserves layout and structural elements, allowing outputs to be traced back to specific regions of the source files, which is important for retrieval-augmented generation, compliance review, and auditable automation.
Document AI with OCR 4 supports pipelines that ingest documents, perform optical character recognition, analyze layout and tables, and feed clean, structured representations into downstream extraction or decision systems. Within enterprise applications, OCR 4 is positioned as a core component for document-processing workflows and agentic scenarios where automation must respect original context, formatting, and evidentiary requirements.
Within Microsoft Foundry, Mistral Medium 3.5 and OCR 4 are exposed through unified tools, APIs, and workflows for model selection and deployment. Foundry supports real-time endpoints and serverless APIs so teams can scale these models into production without separately building hosting or orchestration layers.
Organizations can fine-tune non‑OpenAI partner models, including Mistral’s portfolio, using consistent training and rollout processes that align with existing Azure AI practices. Because the same models and workflows can run in cloud and local environments, enterprises in regulated sectors gain a path from experimentation to controlled, even disconnected, deployment while preserving governance and auditability.
Together, Medium 3.5 and OCR 4 extend Microsoft’s AI stack with additional choice, control, and flexibility for agentic, automation, and knowledge-management solutions.
