What it does

Microsoft has opened Copilot to more than one underlying model, including Anthropic's Claude alongside OpenAI models. In parts of the Copilot experience — notably Researcher and Copilot Studio — you can pick which model does the reasoning.

Key facts

• Model choice appears in specific surfaces (such as Researcher and agent building), not everywhere in Copilot at once.
• Anthropic Claude models are offered as an option in those surfaces.
• Admins can control whether non-Microsoft-hosted models are available to users.
• Different models have different strengths; there's no single "best" for every task.

When to use / skip

Worth caring about when a team has a genuine preference or a task that one model handles better. Skip agonising over it for everyday chat — the default is fine for most people, most of the time.

Configuration decisions

• Whether to allow third-party-hosted models like Claude, and for whom.
• The data-residency and processing implications of a given model choice.
• How you communicate the choice so users aren't paralysed by it.

Gotchas

• Where a model is hosted and how data is handled can differ by option — a real compliance point, not a detail.
• Availability of model choice varies by surface and moves quickly; don't assume it's everywhere.
• Users given a model dropdown with no guidance will either ignore it or cargo-cult a choice.

Consultant notes

• The headline is optionality, but the substance is governance: know the data-handling terms for each model before you enable it.
• Give users a simple default and a one-line "pick X when Y" steer, rather than a menu they can't reason about.
• This space is shifting fast. Re-check which models are available in the tenant rather than trusting a demo from last quarter.

---

*Worth revisiting after th