Introduction
Microsoft has expanded model choice inside Copilot Studio, adding Anthropic’s Claude Sonnet 4.5 and Claude Opus 4.1 alongside OpenAI models. The change — rolling out now to early release environments and slated for broader preview and production by year-end — gives businesses more flexibility to pick the model best suited for specific agent tasks, from deep reasoning to long-running automation.
Why it matters
Enterprises building agents want control over performance, cost, safety and compliance. By supporting multiple vendors in Copilot Studio, Microsoft aims to reduce vendor lock-in, let IT teams match models to workloads, and give admins the controls to enable or restrict which models run in their tenant. That matters for regulated industries and teams that need predictable outputs or tighter safety guardrails.
Anthropic models in Copilot Studio — what’s available and how it works
Microsoft says Anthropic models — including Claude Sonnet 4.5 and Claude Opus 4.1 — are now options for the Researcher agent and for agents built in Copilot Studio. Admins must opt in via the Microsoft 365 Admin Center; once enabled, Anthropic will appear in the model selector within Copilot Studio and Power Platform admin controls. If Anthropic is disabled, agents fallback automatically to Microsoft’s default OpenAI models.
How teams will use multiple models
Copilot Studio supports multi-agent orchestration, so teams can assign different models to different tasks — for example, one model for customer-facing chat and another for compliance checks. Microsoft’s interface includes a prompt-builder dropdown to simplify model choice, making it easier to test which model performs best for a given flow.
Business and safety trade-offs
Adding Anthropic responds to real customer demand for choice and redundancy. Anthropic’s Sonnet 4.5 is noted for extended agentic runs and stronger alignment features — reports suggest it can sustain long autonomous tasks (30+ hours in some tests), which is useful for complex workflows. But enterprises should weigh latency, cost, data residency, and model behavior differences before switching production workloads.
Microsoft emphasises admin controls and automatic fallback to OpenAI if Anthropic is disabled. That design lets security and compliance teams manage exposure: enable Anthropic where its strengths matter and restrict it where data-handling or certainties demand stricter controls.
Early reactions and analyst takeaways
Analysts and industry watchers see the move as Microsoft hedging its model strategy while encouraging competition. “Model choice is now a platform feature,” one analyst wrote, noting that enterprises can optimize across cost, latency and safety. Early adopter organisations say multi-model support simplifies A/B testing and reduces dependency on a single supplier. However, some IT teams caution about the extra governance work required to manage multiple model vendors.
How to get started
- Admins: opt in via Microsoft 365 Admin Center and set tenant policies.
- Developers: use Copilot Studio’s prompt builder to experiment with Claude and OpenAI models, and measure performance on representative tasks.
- Security teams: validate data flows, logging and fallback behaviors; update audit and compliance playbooks for multi-model usage.
Bottom line: Microsoft’s addition of Anthropic models to Copilot Studio gives enterprises practical model choice — useful for matching strengths to tasks and for reducing single-vendor risk. The feature introduces governance complexity, so IT and security teams should test models, set clear policies, and monitor outcomes before broad production use.
Frequently Asked Questions
Which Anthropic models are available in Copilot Studio?
Microsoft currently supports Claude Sonnet 4.5 and Claude Opus 4.1 for Copilot Studio and the Researcher agent. Anthropic models roll out to preview environments now and production by year-end.
How do I enable Anthropic models for my organisation?
Admins must opt in using the Microsoft 365 Admin Center. Once enabled, Anthropic appears in the model selector in Copilot Studio; admins can manage access via Power Platform admin controls.
Will agents fallback if Anthropic is disabled?
Yes — Microsoft configures automatic fallback to the tenant’s default model (typically OpenAI GPT-4o) when Anthropic is disabled, so agents remain operational without manual changes.
Why would I choose Anthropic models over OpenAI models?
Different models have strengths: Anthropic’s Sonnet 4.5 is reported to excel at sustained agentic tasks and safety-focused reasoning, while OpenAI models may offer different latency, cost or integration benefits. Test for your use case.
Are there privacy or compliance differences?
Models are hosted by their providers and may have different data-handling policies. Enterprises should review data residency, logging, and contractual terms before enabling a model for sensitive workflows.
Author note: I’m a tech reporter summarizing Microsoft’s Copilot Studio post and Anthropic’s announcement. I used Microsoft’s blog and Anthropic/press coverage to explain how enterprises can enable model choice, the admin controls involved, and why organizations should test models for safety, cost and compliance before broad rollout