Claude Is Now Inside Microsoft 365 Copilot: How to Switch Models and When To

Buried in Microsoft’s release notes for Microsoft 365 Copilot, dated June 16, 2026, is a line that would have sounded absurd two years ago: Anthropic’s Claude is now available in Microsoft 365 Copilot, in Copilot Chat. Microsoft, the company that bankrolled OpenAI, is now offering a rival lab’s model inside its flagship office AI. The same notes tucked in a second change, Black Forest Labs’ FLUX.2 Flex image model coming to PowerPoint.

If you spend your day in Outlook, Word, and Teams, this is not industry gossip. It means the Copilot button your company already pays for can now run on a different brain, and the brain you pick changes what comes out.

What the release notes say, and what they do not

Worth being precise here, because a lot of coverage of these things rounds up. Microsoft’s June 16 notes confirm two facts: Claude is available in Microsoft 365 Copilot via Copilot Chat, and FLUX.2 Flex from Black Forest Labs is now an image generation option in PowerPoint. That is the documented core.

What the notes do not spell out matters just as much. They do not say which Claude model version you get. They do not publish a benchmark showing Claude beating or trailing the default models inside Copilot. They do not lay out data-handling terms in the release note itself. Release notes are announcements, not contracts, and the details live in Microsoft’s admin documentation, which your IT team can read with their tenant settings open.

So treat everything in the next few sections accordingly: the availability is documented fact, and the guidance on when to use which model is advice, drawn from how these models behave in their own native apps.

Why a model picker inside Copilot matters

Until now, Copilot Chat was a single-vendor experience. You typed, Microsoft’s chosen OpenAI-based models answered, and there was no second opinion. A model choice turns Copilot into something closer to a panel than a single adviser.

Different models genuinely produce different work. Ask two frontier models to summarize the same 40-page contract and you get different structures, different emphasis, and sometimes a different judgment about what mattered. One will catch the clause the other skims. We covered the personality gap between these two labs’ products in Claude vs Copilot, and the short version holds: these are not interchangeable engines with different logos.

For the average office worker, the practical upside is a free second draft. Same prompt, different model, thirty seconds, and you keep the better output.

First hurdle: your admin

Here is the part that determines whether any of this reaches your screen this week. Model availability in a Microsoft 365 tenant is controlled by administrators, and features Microsoft ships routinely arrive at different times for different organizations. If you open Copilot Chat tomorrow and see no way to pick Claude, the feature is probably not broken. It is either not enabled for your organization or not rolled out to your tenant yet.

Three steps, in order:

1. Look for a model selector in Copilot Chat. If Claude appears as an option, you are done, skip to the next section.
2. If not, ask your IT admin whether Anthropic models are enabled for your tenant. Third-party model access is the kind of switch admins reasonably leave off until someone asks, and “someone asked” is often the entire approval process.
3. If your admin has questions about data handling, point them at Microsoft’s Copilot admin documentation rather than guessing. Claude comes from Anthropic, a separate company, and your compliance folks will want the actual terms, not a summary of a summary.

That conversation is worth having even if you are not the decision maker. Admins enable what people request, and a specific request (“can we turn on the Claude option in Copilot Chat?”) beats a vague one.

When to pick Claude, when to keep the default

No published head-to-head exists for the two models inside Copilot specifically, so what follows is judgment, not measurement. It comes from how these models perform in their own apps, where the differences are well established.

Reach for Claude when the task is long-document work. Summarizing a lengthy report, pulling obligations out of a contract, holding a consistent thread across a big pile of text: this is the territory where Claude built its reputation, and it is the first thing worth testing once the option shows up in your tenant. Drafting is the other candidate. Claude’s writing tends to run less template-shaped, which matters when the output goes to a client under your name rather than into a folder.

Keep the default models as your everyday driver, at least at first. They are the incumbent inside Copilot, Microsoft has tuned the product around them, and quick factual questions or routine summaries do not need a second opinion.

And when it matters, use both. The genuinely new capability here is not that Claude is better or worse. It is that a second opinion now costs one click. Before an important document goes out, running the same prompt through both models and comparing is the cheapest quality check available to you. Most of the time they will agree. The times they disagree are exactly the times you wanted to know.

Here is a concrete way to run that test without turning it into a project. Pick three tasks from your actual week: one summary of a long document, one draft you would normally write from scratch, and one messy question with no clean answer, the kind you would ask a smart coworker. Run each through both models with the identical prompt. Do not read the outputs side by side at first; read each as if it arrived alone, and mark whether you would send it. Then compare. After three tasks you will have a clearer opinion than any review can give you, because it will be an opinion about your work.

Teams should also think about consistency. If five people on a project use Copilot and two of them quietly switch to Claude, the drafts that come back will have subtly different shapes. That is not a problem, but it is worth knowing about before someone asks why the Tuesday report suddenly reads differently. A one-line team norm (“use whichever model, but say so in shared docs”) costs nothing.

One habit transfers regardless of model: a specific prompt beats a vague one everywhere. Our guide to writing AI prompts applies to Copilot Chat verbatim.

The PowerPoint footnote that isn’t one

The same June 16 release notes added FLUX.2 Flex, an image model from Black Forest Labs, to PowerPoint. Read those two changes together and a pattern appears: Microsoft is assembling models from multiple labs inside Microsoft 365 and letting the products mix and match. Anthropic for chat. Black Forest Labs for slides. Whoever ships the best component gets the slot.

For users, that is quietly great news. It means the AI inside your office suite is no longer capped at what one lab shipped this quarter. For anyone who makes slides, it is also worth a concrete look: generating a usable image inside PowerPoint beats hunting a stock photo site for one more handshake picture.

What to actually do this week

Check Copilot Chat for a model selector. If Claude is there, give it your most annoying current document task, the summary or draft you have been putting off, and compare it against the default model on the same prompt. If it is not there, send your admin the release notes link and ask whether it can be enabled.

Then form your own opinion. The release notes tell you Claude is in Copilot. They cannot tell you whether it writes a better version of your Thursday status report. That experiment takes ten minutes, and it is the only benchmark that counts for your job.