Copilot isn’t popular, but it still does a pretty good job of offering some ChatGPT features for free with censorship and restrictions. Right now, Copilot’s Think Deeper feature uses the o3-mini-high model with knowledge cut off in October 2023, but Microsoft now appears to be A/B testing the o4-mini-high (or the regular o4-mini) model.
By default, Copilot has two modes or three (Deep Research) if you pay for the $20 Copilot Pro subscription. One is Quick Response, which uses one of OpenAI’s regular models, and the other is “Think Deeper,” a reasoning model. In March 2025, Microsoft confirmed the “Think Deeper” toggle turns on o3-mini-high, one of the premium OpenAI models.
“Think Deeper’s reasoning model has been upgraded to OpenAI’s o3-mini (high). This means you’ll wait less to get answers,” Microsoft noted in the March 2025 release notes.
o3-mini-high has a knowledge cut off of October 2023, and it’s no longer offered with ChatGPT Plus, Pro or Teams/Enterprise. OpenAI has already replaced the model with o4-mini-high, and Microsoft is also following the move.
Microsoft Copilot briefly replaced the o3-mini-high with o4-mini-high (or o4-mini)
See the screenshot below when I asked Copilot’s Think Deeper what the knowledge cut-off date is, and it clearly says October 2023 with absolute confidence. This confirms it’s still on o3-mini-high in this particular session:
But when I used one of my Microsoft accounts and asked Copilot what its knowledge cut-off date is with Think Deeper turned on, it told me that the training data goes up until June 2024.
Based on our tests and other findings, it does appear to be the o4-mini-high model, which has a knowledge cut-off in June 2024. It could also be o4-mini, but I’m assuming it’s o4-mini-high because Microsoft has used the “-high” variant for Copilot’s o3-mini integration.
Knowledge cut-off dates we have for OpenAI models:
- o3-mini / o3-mini-high: October 1, 2023
- o3: June 1, 2024
- o4-mini / o4-mini-high: June 1, 2024
- GPT-4.1 family (GPT-4.1, GPT-4.1 mini, GPT-4.1 nano): June 2024
Is it possible that “Think Deeper” could be using o3, which also has a knowledge cutoff of June 2024? I don’t think so. It’s an expensive model, and it would not make sense to replace the 30-second “Think Deeper” with that, but it cannot be ruled out with 100% confidence.
According to OpenAI, o4-mini does significantly better than the predecessor, but it’s still cheaper. For example, on the AIME math tests, o3-mini scored about 87% on the 2024 exam and 86.5% on 2025, whereas o4-mini scored about 93.4% in 2024 and 92.7% in 2025.
As for the “-high” variant, compared to o3-mini-high, the o4-mini-high variant spends more time “thinking,” which improves its quality.
