Microsoft Adds Anthropic Models to Copilot
Microsoft integrates Anthropic Claude models into Copilot, ending OpenAI exclusivity after customer complaints about Excel and PowerPoint.
The software maker brought Claude into Office apps after enterprise users pushed back on mistakes in Excel and PowerPoint.
Microsoft has begun integrating Anthropic models into Microsoft 365 Copilot, ending years of practical exclusivity for OpenAI across its core workplace software. The change surfaced in reporting from The Information and was confirmed through technical documentation published by Microsoft. It is a pragmatic turn: despite investing more than $13 billion in OpenAI, executive leadership chose to open its ecosystem to an outside competitor rather than leave business customers dealing with faulty spreadsheets and poorly structured slide decks.
Internal Tests and the Shift Away From Exclusivity
The pivot started gathering steam in mid-2025. Corporate accounts were logging frequent complaints about Copilot miscalculating complex formulas in Excel and stumbling over presentation layouts in PowerPoint. Facing chronic issues with OpenAI’s GPT models, Charles Lamanna, then running business application platforms and now executive vice president over Copilot and Office, set up side-by-side benchmark demonstrations for Chief Executive Satya Nadella.
In those head-to-head tests, Anthropic’s Claude family handled financial calculations and document layouts with visibly better precision. Selling the idea internally took work. Because Microsoft’s massive capital commitment gave it preferential pricing with OpenAI, proposing a paid deal with an outside competitor was something Lamanna described internally as borderline heresy at the time. Yet customer demand for reliable tools ultimately settled the debate.
Anthropic Models in Microsoft Copilot
The clearest public confirmation came through Microsoft’s enterprise governance records. The company added Anthropic as an approved subprocessor within its Microsoft Online Services documentation. That administrative shift activates Claude models by default across enterprise tenants using Microsoft 365, Copilot Studio, Researcher, and Power Platform. Initial exceptions are largely confined to jurisdictions with strict local data rules, such as the European Union and the United Kingdom.
On the security side, corporate data remains protected under Purview and Work IQ compliance boundaries. The key operational difference is that the orchestration layer can now route prompts to different model families depending on which system handles the specific job best.
RUMOR: Projected $2 Billion Annual Contract
People familiar with the contract talks told The Information that growing call volumes could see Microsoft pay Anthropic more than $2 billion each year.
That estimate must be treated as a RUMOR. Neither Microsoft nor Anthropic has disclosed contract values, token pricing terms, or the exact query split between OpenAI and Anthropic in their public financial statements.
Orchestration Over a Single Provider
The successful multi-model rollout cemented Lamanna’s position leading Microsoft’s core productivity applications. It also provides the foundation for Copilot Cowork, a workspace built for autonomous agents handling complex, multi-step business workflows.
For the enterprise software industry, the decision validates a point Satya Nadella has made repeatedly: underlying models are becoming interchangeable utilities. By choosing functional accuracy in Excel over strict fidelity to OpenAI, Microsoft demonstrated that customer distribution, data context, and software governance outweigh any attachment to a single model provider.