Microsoft integrates Anthropic's Claude models into 365 Copilot

Microsoft has announced that it will add Anthropic's Claude Sonnet 4 and Claude Opus 4.1 as selectable models within 365 Copilot, expanding the set of models available to users while retaining OpenAI’s models as the engine for Copilot overall.

Microsoft said users will be able to select Anthropic models, Claude Sonnet 4 and Claude Opus 4.1, in Copilot’s AI-powered reasoning agent "Researcher," and when developing agents in Microsoft Copilot Studio. "Starting Wednesday, users who opt in to try Claude can switch between OpenAI and Anthropic models in Researcher," said Charles Lamanna, president of Microsoft's business and industry Copilot operations." Copilot itself will continue to be powered by OpenAI’s latest models by default.

What this means in practice

For end users and administrators, the immediate change is choice: within the specific Copilot features named by Microsoft, people who opt in can select which vendor’s model to use. The update also extends to Copilot Studio, where agents are configured and tested, giving developers an option to build or tune agents against Anthropic’s newer Claude variants as an alternative to OpenAI models.

Microsoft has been diversifying its AI sourcing and hosting approach: it is developing its own models, has integrated models from other providers into Azure, and previously said it would host models from companies such as xAI and Meta in its own data centres. Anthropic’s models are primarily hosted on Amazon Web Services, which Microsoft notes is a rival to its cloud business. The move to offer Anthropic models inside Copilot reflects a multi-supplier posture that allows Microsoft to offer customers model choice while managing its own product and hosting strategies.

Implications and trade-offs

Implications and trade-offs Whether to permit several vendor models within a single assistant product is an empirical dilemma to both the client and to Microsoft:

• Choice vs. consistency: organisations need to balance between the advantages of the vendor choice and the possible differences in output, cost, latency, and compliance among the models.
• Cloud and hosting dynamics: The models used by Anthropic are mostly hosted on AWS, which will introduce cross-cloud concerns whereby a Microsoft product calls metrics on a rival cloud provider that Microsoft is also hosting other third-party metrics in their data facilities.
• Product positioning: Leaving OpenAI models as the default with the addition of Anthropic options is an indicator of the platform transforming to one supporting a variety of suppliers as opposed to one supplier dependency.

Anthropic Claude Sonnet 4 and Claude Opus 4.1 are updated by Microsoft. They can be accessed by users in established Copilot functionalities, with the models of OpenAI at the core of Copilot. The transformation enhances the vendor selection within the assistant ecosystem of Microsoft and highlights the multi-faceted nature of the company in terms of AI sourcing and hosting.