Microsoft Adds Anthropic AI Models to Copilot 365, Expanding Beyond OpenAI

• New Integration of Anthropic Models

  Claude Sonnet 4 and Claude Opus 4.1 are now options in Microsoft 365 Copilot. Availability starts in Researcher (deep reasoning over web, third-party data, and organizational content like emails, chats, meetings, files) and in Copilot Studio for building enterprise agents. Model choice is surfaced directly in the UI.

• Dual Model Flexibility for Users

  Users can switch between OpenAI and Anthropic within Researcher once their org opts in; a prominent “Try Claude” button appears in the Researcher header. In Copilot Studio, a dropdown exposes model options for orchestration and prompts; OpenAI remains default. Researcher sessions using Claude revert to the default Generative AI Model for Microsoft 365 after the session ends.

• Strategic Shift Beyond OpenAI

  The partnership reduces single-vendor dependence as Microsoft formalized OpenAI as a competitor (2024) and pursues a broader multi-model strategy; Microsoft has also signaled work with xAI and Meta models and continues developing its own models. Separately, GitHub Copilot in Visual Studio Code now primarily relies on Claude Sonnet 4 when automatic model selection is enabled—an indicator of Anthropic’s performance in coding tasks. Reports indicate Anthropic could power Excel and PowerPoint scenarios next.

• Corporate Statements

  Charles Lamanna highlighted model choice as part of bringing “the best AI innovation” to Copilot, tuned for work scenarios; Copilot Studio guidance details how Anthropic models join OpenAI across orchestration, chat, and deep reasoning use cases, another signal that Microsoft adds Anthropic aligns with agentic workflows.

• Cloud & Compliance Implications

  Anthropic models run outside Microsoft-managed environments, accessed via the Anthropic API, and are hosted on AWS (with some reporting that Anthropic models also run on Google Cloud). Opt-in use requires acceptance of Anthropic Commercial Terms and DPA; Microsoft’s Product Terms, DPA, data-residency commitments, SLAs, and the Customer Copyright Commitment do not apply when you choose Anthropic. Admin enablement is required in the Microsoft 365 admin center.

• Operational Behavior & Controls

  In Researcher, Claude is session-scoped and the system reverts to the default model when the session ends. In Copilot Studio, if Anthropic is later disabled, agents automatically fall back to OpenAI GPT-4o without extra configuration.

• Rollout Timeline

  • Researcher: Rolling out via the Frontier Program to Microsoft 365 Copilot-licensed customers who opt in (from Sept 24, 2025).
  • Copilot Studio: Available now in early release environments, preview in all environments within two weeks, production readiness by end of 2025.

• Multi-Model Orchestration

  Copilot Studio supports multi-agent systems and prompt tools, letting builders mix models from Anthropic, OpenAI, and others in the Azure Model Catalog per task.

• Adoption Context

  Enterprise adoption remains phased: industry checks cited by KeyBanc indicate ~60% of customers pilot Copilot for ~10% of their Microsoft 365 users, while ~4% deploy to all users, suggesting model choice may broaden use cases but rollouts are still cautious.

Why This Matters

Copilot shifts from a single-vendor setup to a multi-model enterprise platform. Organizations gain choice, leverage, and risk diversification while assuming new governance responsibilities (non-Microsoft hosting, separate terms, and compliance boundaries). The approach accelerates agentic workflows and codifies workload-based selection. As Microsoft adds Anthropic across key apps, pragmatic “best model for the job” routing becomes standard.

This article was drafted with the assistance of generative AI. All facts and details were reviewed and confirmed by an editor prior to publication