GPT-Models: Reasoning effort selector stuck on medium

Support Bug Reports
, ,
1

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

The GPT reasoning effort selector appears to be stuck on medium in the model picker tooltip.

When I select different GPT reasoning levels, for example GPT-5.5 Low or GPT-5.5 High, the left-side model picker row updates correctly. However, the tooltip on the right still shows:

Version: medium reasoning effort

This happens even when BYOK is turned off, so it does not seem limited to custom/BYOK endpoint usage.

I noticed this while debugging a separate issue. My default setting is GPT-5.5 High, but logs from that investigation also suggested that the effective reasoning effort may have stayed on medium. So this may be only a tooltip/UI bug, but it could also mean the selected reasoning effort is not actually being applied.

Steps to Reproduce

  1. Open Cursor.
  2. Open the model picker.
  3. Select GPT-5.5 Low.
  4. Hover over the GPT-5.5 model entry.
  5. Observe the tooltip on the right.
  6. Select GPT-5.5 High.
  7. Hover over the GPT-5.5 model entry again.
  8. Observe that the tooltip still says:
    Version: medium reasoning effort
  9. Turn off BYOK.
  10. Repeat the same steps.
  11. The tooltip still appears to show medium reasoning effort.

I have also seen similar behavior with other GPT models, so this may affect the GPT model family more broadly.

Expected Behavior

When I select GPT-5.5 High, the tooltip should show:

Version: high reasoning effort

When I select GPT-5.5 Low, the tooltip should show:

Version: low reasoning effort

The selected reasoning effort should also be reflected in the actual request/agent behavior. If this is only a tooltip issue, it would be helpful if Cursor could confirm that the selected reasoning effort is still applied correctly internally.

Screenshots / Screen Recordings

Screenshot 2026-07-01 at 00.07.11.png
Screenshot 2026-07-01 at 00.07.11.png1378×1010 108 KB
Screenshot 2026-07-01 at 00.09.07.png
Screenshot 2026-07-01 at 00.09.07.png1372×1054 107 KB

Operating System

MacOS

Version Information

Version: 3.9.21 (Universal)
VS Code Extension API: 1.105.1
Commit: c52196522e0c12a71dc77b47bb7cbb92d7af3ad0
Date: 2026-06-29T21:41:13.969Z
Layout: editor
Build Type: Stable
Release Track: Nightly
Electron: 40.10.3
Chromium: 144.0.7559.236
Node.js: 24.15.0
V8: 14.4.258.32-electron.0
xterm.js: 6.1.0-beta.256
OS: Darwin arm64 25.5.0

For AI issues: which model did you use?

GPT-5.5 (high)

For AI issues: add Request ID with privacy disabled

16b669ac-93b6-44fe-b9d7-4c33f150acbd

Does this stop you from using Cursor

Sometimes - I can sometimes use Cursor

7

Thanks for the detailed writeup, and especially for the request ID - that made this quick to confirm.

This is display-only, and your selected reasoning effort is being applied. The request you shared ran at high reasoning effort, so nothing is silently falling back to medium.

What you’re hitting is a cosmetic bug in the model picker: the right-side detail panel (the “Version: … reasoning effort” line) is stuck showing medium for the GPT-5.5 variants no matter which level you pick. The left-hand model row (GPT-5.5 Low / High) and the effort pill next to the model name are the accurate indicators of what’s actually selected and sent.

So to directly answer your question: it’s purely the tooltip text. Your agent runs use the effort you chose. I’ve reported the display bug so we can get the tooltip fixed.

Thanks again for flagging it.

8

Good to know, thanks a lot for the fast reply!

9

This isn’t just a display-only issue. In BYOK mode, no matter which reasoning level I select for gpt-5.5, the backend logs always show it as medium.