Forum Discussion
Has model selection been removed from Microsoft 365 Copilot?
I noticed that the option to manually select GPT models in Microsoft 365 Copilot seems to be gone, or at least no longer visible. I have also seen similar concerns discussed on Reddit.
Could Microsoft clarify the current status?
- Has manual model selection been removed?
- How can users know which model Copilot is currently using?
- Is Copilot still using stronger models such as GPT-5.5, or only automatic routing now?
- Why does “Deep Analysis” / “Think Deeper” appear to switch back to Auto mode?
For corporate use, this lack of transparency is concerning. If I cannot see or control which model is being used, it reduces trust in the quality of the output and makes me more cautious about what tasks I rely on Copilot for.
A clear statement from Microsoft would be appreciated.
3 Replies
Hi, I understand the concern. For business use, it is much easier to trust the output when you know what kind of model behavior you are getting.
From what I have seen, Microsoft 365 Copilot generally behaves more like a managed service with automatic routing than a plain model picker. Features such as "Think Deeper" or "Deep Analysis" may select a different reasoning path for that request, but that does not always mean the UI will keep showing a fixed model name afterward.
I would not rely on Reddit screenshots or model-name guesses for governance decisions. The best places to watch are Message Center, the Microsoft 365 roadmap, and official Copilot release notes. In the meantime, I would treat Copilot output as automatically routed and validate it based on task criticality, especially for legal, financial, security, or customer-facing work. A clear Microsoft statement on current model-selection behavior would definitely help admins set expectations.
It appears that GPT 5.5 has returned to the Copilot Chat model selector, with two visible options: GPT 5.5 Quick response and GPT 5.5 Deep response.
Microsoft Learn confirms that Copilot Chat offers Auto, Quick response, and Think deeper modes. Quick response prioritizes speed, while Think deeper uses deeper reasoning for more complex requests. Auto lets Copilot route the request dynamically.
However, Microsoft Learn does not yet document the exact technical mapping behind these GPT 5.5 labels in Copilot Chat.
Depending on admin settings, Anthropic models can also be enabled so models such as Claude Opus appear in the selector.
This is different in Copilot Studio, where model selection is more explicit: makers can choose the primary AI model for agent orchestration, with model categories.
big issue for me as well. I have been trying to find info but to no avail
