Copilot isn’t widely used, but it still manages to deliver a number of ChatGPT-like capabilities for free, albeit with some limitations and content filtering. At the moment, Copilot’s Think Deeper function runs on the o3-mini-high model, which has data knowledge up until October 2023. However, Microsoft seems to be running A/B tests using the o4-mini-high (or possibly just o4-mini) model.

By default, Copilot offers two modes, or three if you opt for the $20 Copilot Pro subscription that unlocks Deep Research. One mode is Quick Response, powered by one of OpenAI’s standard models, while the other is “Think Deeper,” which uses a reasoning-based model. Back in March 2025, Microsoft officially announced that the “Think Deeper” feature activates the o3-mini-high model from OpenAI.

“With the latest update, the reasoning engine behind Think Deeper now leverages OpenAI’s o3-mini-high model. This translates into shorter wait times for results,” Microsoft stated in its March 2025 changelog.

The o3-mini-high model stops at October 2023 for training data and is no longer available through ChatGPT Plus, Pro, or Teams/Enterprise subscriptions. OpenAI has already phased this out in favor of o4-mini-high, and Microsoft appears to be following suit.

Microsoft Copilot temporarily swapped o3-mini-high for o4-mini-high (or simply o4-mini)

In a recent test, I asked Copilot’s Think Deeper feature about its knowledge cutoff date, and it confidently replied with October 2023, as shown in the screenshot below. This strongly suggests that the system was still operating on o3-mini-high during that session:

However, when I posed the same question using another Microsoft account—specifically asking about the knowledge cutoff date while Think Deeper was enabled—the response indicated that the training data extended up to June 2024.

Based on our testing and supporting evidence, it looks like the backend might be using the o4-mini-high model, which also ends at June 2024. It could potentially be the standard o4-mini, though I suspect it's the “-high” variant since Microsoft previously opted for the “-high” version when integrating o3-mini into Copilot.

Here’s what we know regarding knowledge cutoff dates 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 there a chance that the Think Deeper feature is actually running on o3? While technically possible, it seems unlikely given that o3 is a more expensive model, and replacing the current 30-second “Think Deeper” process with it wouldn’t make much sense financially. Still, nothing can be ruled out with absolute certainty.

According to OpenAI, o4-mini performs significantly better than its predecessor, yet remains more affordable. For example, on the AIME math exams, o3-mini achieved around 87% on the 2024 test and 86.5% on the 2025 version, whereas o4-mini scored approximately 93.4% in 2024 and 92.7% in 2025.

As for the “-high” variant, compared to o3-mini-high, the o4-mini-high variant dedicates more time to processing, which enhances the quality of its output.

The above is the detailed content of Test hints Microsoft Copilot may offer ChatGPT's o4-mini-high for free. For more information, please follow other related articles on the PHP Chinese website!