This is less about a new model architecture and more about distribution and integration, which is often the more practically important story for developers already working inside a particular cloud ecosystem. Mistral's Medium 3.5, an open-weight large language model, and OCR 4, its document parsing and extraction model, are now available directly inside Microsoft Foundry, with Medium 3.5 additionally integrated into Copilot Studio. For a developer or team already building on Azure, this removes a meaningful piece of friction: instead of standing up a separate integration to call Mistral's models through their own API or a third-party hosting provider, these models are now available through the same tooling, billing, and access-control surface as the rest of an organization's Azure-based AI stack. That matters more than it might initially sound, because a huge amount of practical friction in adopting a new model isn't about the model's raw capability, it's about the operational overhead of authentication, network configuration, compliance review, and billing integration required to actually start using it inside an existing enterprise environment. Having Medium 3.5 and OCR 4 available as first-class options inside Foundry sidesteps most of that overhead for teams that have already standardized on Microsoft's AI tooling. It also reflects a broader and increasingly common pattern among major cloud providers: rather than only offering their own in-house models, they're incorporating strong third-party and open-weight alternatives directly into their platforms, giving customers more model choice, including options with different licensing terms and self-hostability, without requiring them to leave the cloud ecosystem they've already invested in.