Since June 2026, Microsoft 365 Copilot has been rolling out a model picker across more and more apps. Claude or GPT: it's not a matter of preference, but a practical decision. This article looks at which model is the better fit for which tasks in everyday SME work, and what Swiss IT admins need to clarify first.

Why Did Microsoft Integrate Multiple AI Models in the First Place?

Microsoft added Anthropic's Claude to Copilot to give customers more choice and reduce dependency on a single provider. Both models are included in the existing Microsoft 365 Copilot subscription at no extra cost.

Since January 2026, Anthropic has been an official subprocessor for Microsoft. That means Anthropic is subject to the same data protection requirements as Microsoft and may not use customer data to train its models.

If you don't change anything, you're using "Auto" mode: Copilot automatically selects the most suitable model for each request. The manual model picker is there for users who want more direct control.

What Changed in Copilot in June 2026?

In May and June 2026, the model picker was significantly expanded: Claude Opus 4.8 is rolling out across Copilot Chat, Word, Excel, PowerPoint, and Copilot Studio. This is the largest expansion of model choice to date.

On June 16, 2026, Copilot Cowork became generally available worldwide. Cowork is Microsoft's agent for multi-step tasks: it can generate reports, presentations, and Excel files from a single prompt, drawing on your emails, meetings, and SharePoint content. Cowork is built on Anthropic technology and runs on Claude Opus 4.8 and Sonnet 4.6 by default.

The Researcher agent also offers two new modes: "Critique" (GPT drafts the research, Claude reviews and refines it) and "Council" (both models respond in parallel for direct comparison). This is particularly useful for decision papers and market analyses.

Which Tasks Is Claude Better Suited For?

Claude tends to deliver better results when depth matters more than speed. This is especially true for analyzing long documents, structured writing, and multi-step reasoning tasks.

Specific use cases:

• Contract analysis and document review: Claude reads long PDFs more thoroughly and delivers more nuanced summaries.
• Structured reports and meeting minutes: Well suited for content that needs to be logically organized and consistently worded.
• Data analysis in Excel: Claude follows multi-step instructions more reliably and explains its own calculations more transparently.
• Copilot Cowork: For delegated tasks like "create a monthly report from my recent meetings and emails," Opus 4.8 is the default.

When Is GPT the Better Choice?

For quick, creative, or conversational requests, GPT responds faster. Especially GPT-5.5 Instant in particular is built for short, everyday answers.

Typical scenarios: short email drafts, brainstorming, Teams meeting summaries, or simple follow-up questions. GPT remains the default model for most Copilot experiences; if you haven't changed anything, you're already working with it.

One practical note: "Auto" mode in the Researcher agent automatically activates Critique mode. GPT generates, Claude reviews – which often produces better results than either model alone.

What Do Swiss IT Admins Need to Check Before Rollout?

For Swiss companies, Anthropic models in Microsoft 365 Copilot are disabled by default, as an EFTA member, Switzerland falls under the regions requiring an explicit opt-in.

A Microsoft 365 administrator has to manually enable the toggle in the Admin Center under "Copilot Settings → Anthropic." Only then will all users with a Copilot license have access to Claude models.

Important for your data protection assessment: processing by Anthropic takes place outside the EU data protection boundary – on servers operated by Amazon Web Services and Google Cloud Platform, primarily in the US. For SMEs handling regulated data, for example, in finance or healthcare, it's advisable to consult your data protection officer before activation. Another immediately practical prerequisite for any Copilot rollout, especially once Cowork starts accessing your tenant data, is cleaning up SharePoint permissions beforehand.

Conclusion

Model choice in Microsoft 365 Copilot is a practical question, not a strategic one. "Auto" is enough for most day-to-day work. Claude pays off specifically for in-depth document analysis, structured reports, and multi-step tasks in Copilot Cowork.

For Swiss IT managers, the takeaway is this: enable Anthropic deliberately, after a brief data protection review, not automatically. Choosing the right model is a matter of nuance; preparing the rollout properly is not.