Claude is now inside Copilot Chat.
Microsoft's July 2026 release notes contain a line that would have sounded absurd two years ago: Anthropic's Claude is now a model you can pick inside Microsoft 365 Copilot Chat — on Windows, Mac, web, iOS, and Android. Same Copilot window, same tenant, different brain. Here's what actually changed, how to try it, and when picking Claude is the right call.
01 What changed
Copilot Chat now has a model option. Alongside the default OpenAI models, you can select Claude for a conversation. Microsoft's own guidance positions Claude for complex analysis, document understanding, and structured content generation — and that matches what Claude has been known for all along: long documents, careful reasoning, structured writing.
This isn't a side app or a plugin. It's the same Copilot Chat that ships with your Microsoft 365 license — your files, your permissions, Microsoft's governance — with a choice of which frontier model does the thinking. Microsoft has been heading here for a while: Copilot Studio already lets builders choose Anthropic and xAI models per agent. Now the everyday chat surface gets the same flexibility.
And it keeps widening: Microsoft announced on July 2 that Anthropic's newest model, Claude Sonnet 5, is rolling out in Microsoft 365 Copilot — including Copilot Cowork and PowerPoint — positioned for agentic, multi-step work across documents, spreadsheets, and presentations. Rollout varies by tenant and region, so check your model selector.
Updated August 2026 — the picker follows you into Word. Microsoft’s Microsoft 365 Copilot release notes dated August 11, 2026 added the ability to choose Anthropic models when editing a Word document with Copilot (web). Until then, picking a model was a Copilot Chat thing; now the choice sits in the model selection menu inside the document itself, alongside the existing OpenAI models. It is governed by the same admin switch in section 04 — if Claude is missing in Chat, it will be missing in Word too. Walk through it in Copilot in Word.
02 How to try it
Open Copilot Chat (the app or microsoft365.com), start a new chat, and look for the model selector. If your admin has enabled Anthropic models, Claude appears as an option. Then run Microsoft's own three suggested tests:
- Hand it a long document and ask for an analysis — structure, gaps, risks.
- Ask it to break down a complex topic you actually need to explain to someone this week.
- Request a multi-step plan or structured outline for a real project.
03 When to pick Claude (honestly)
| The job | Reach for… |
|---|---|
| Long-document analysis — contracts, reports, policies | Claude — this is its home turf |
| Structured writing — outlines, plans, specs, careful prose | Claude — noticeably steadier structure |
| Breaking down genuinely complex topics | Claude — Microsoft's own recommendation |
| Everyday drafts, quick answers, M365-grounded lookups | Default OpenAI models — fast and deeply wired into Graph |
| Agents that need live X / web spice | Grok in Copilot Studio — separate lesson |
The honest caveat: model choice inside Copilot is new, and Microsoft will keep tuning which model handles what behind the scenes. Treat the selector as a tool for the jobs above, not a religion. Run your own A/B on your real work — the difference shows up most on long inputs and structured outputs.
04 For admins: turning it on
IT admins can enable Anthropic as a model provider for specific users and groups from the admin center (xAI models are US-only for now, via Copilot Studio). Practical rollout pattern:
Admin quick list
- Review the Anthropic model provider setting in the Microsoft 365 admin center
- Enable for a pilot group first — the analysts and writers who benefit most
- Update your internal AI usage documentation to reflect the new model options
- Note: data governance still runs through your tenant — the model changes, the controls don't
Update — July 24, 2026: a second model-provider switch flipped on by default. Microsoft onboarded OpenAI as a subprocessor for Microsoft 365 Copilot and Copilot Studio, and for eligible tenants the setting was auto-enabled on July 24 unless admins had selected “No users” first (Message Center MC1422074). It covers OpenAI-operated models only — starting with GPT-5.6 — running on OpenAI infrastructure; OpenAI models operated by Microsoft through Azure OpenAI are unaffected. Prompts and responses keep enterprise data protection either way, and admins can review or change the setting anytime in the Microsoft 365 admin center.
05 Why this matters
Microsoft bundling Copilot into Business plans (from July 1) while opening the model menu to its biggest AI rival tells you where this market is going: the assistant is the product, the model is a component. For you, that means the skill that compounds isn't loyalty to one model — it's knowing which brain to hand which job. That's exactly the skill this track teaches.
This week's challenge
Pick one real document over 20 pages. Run the same three prompts — summary, risk list, action plan — once on the default model and once on Claude. Compare side by side. You'll know in 15 minutes which jobs deserve the selector click.