Microsoft has added Anthropic models to Microsoft 365 Copilot. Claude Sonnet 4 and Claude Opus 4.1 are now available as model options inside Copilot. They appear first in the Researcher tool and in Copilot Studio for makers building agents.
The change expands the model choices that enterprise customers can use inside Copilot. Microsoft will continue to offer OpenAI models. Customers may now opt in to try Claude inside Researcher or select Anthropic models when building agents in Copilot Studio. The rollout begins through the Frontier program for eligible Copilot customers.
Anthropic designed Opus 4.1 for long-running workflows and for complex coding tasks. Sonnet 4 focuses on deep reasoning and planning with large outputs. Anthropic reports that Opus 4.1 improves coding benchmarks and that both models excel at multi-step and agentic tasks. These capabilities make them a fit for research and for automation inside business workflows.
Copilot Studio makers can now pick models from a catalog that includes Anthropic and OpenAI. Microsoft says the studio will let teams mix models for different agent steps. The platform supports connections to external knowledge sources and runtime tools so agents can act with context and data. These controls aim to let teams tune agents for accuracy, performance, and cost.
Anthropic models currently run on Amazon Web Services. Microsoft will access them by API as it does with other third-party models. Analysts note that this move diversifies model supply and reduces reliance on a single provider while also creating new cloud and governance choices for customers. Microsoft said it will keep adding options to serve business needs.
Why this matters
Enterprises gain choice and more options to match tasks to model strengths. Research teams may prefer Opus for coding tasks. Strategy teams may prefer Sonnet for deep analysis and planning. The mixed model approach can let an agent use one model for data gathering and another for final reasoning. That can improve results while letting customers balance cost and speed.
Privacy and governance
Microsoft says customers control which models they use and how data flows to external providers. Security teams must review where model data is sent and how outputs are stored. Organizations that need strict data residency should confirm hosting and contractual terms before they onboard new models. Microsoft provides tools and documentation to help with this review.
How to try Claude
If your organization has Microsoft 365 Copilot, you may see a Try Claude option in Researcher. Opting in will let you switch between OpenAI and Anthropic models for supported tasks. Copilot Studio users can also select Claude models when they build and test agents. Expect controls to manage cost usage and model selection while the pilot expands.
What changed
Microsoft added Anthropic Claude Sonnet 4 and Claude Opus 4.1 to Copilot. The models are live for Copilot business customers who opt in. Microsoft keeps OpenAI as a core provider while adding Anthropic for choice and specialization.