How can I revert to Cursor WITHOUT GPT-5?

Ideas Feedback
1

Describe the Bug

Since the rollout of GPT-5 in cursor, I have seen a DRAMATIC DECREASE in the usefulness of Cursor.

  • There is no explanation anymore to help understand and follow the thought process ← this makes it impossible to steer the model in the right direction anymore.
  • Cursor stops in the middle of a fix asking for confirmation for some obvious thing (“Shall I continue fixing the issue?” ← Obviously YES!!)
  • My productivity with Cursor has dropped MASSIVELY: things that used to take 15mn now can’t be done in less than 2 hours!!

This is is BAD!!!

Steps to Reproduce

Open Cursor
Try to do anything.

Expected Behavior

Open cursor
Get helpful information on Cursor thought process and action.
Cursor only asks for action when the action or confirmation is needed.
We have helpful summary how what was done after each action is done.

Operating System

Linux

Current Cursor Version (Menu → About Cursor → Copy)

Version: 1.2.1
VSCode Version: 1.99.3
Commit: 031e7e0ff1e2eda9c1a0f5df67d44053b059c5d0
Date: 2025-07-03T06:13:13.763Z
Electron: 34.5.1
Chromium: 132.0.6834.210
Node.js: 20.19.0
V8: 13.2.152.41-electron.0
OS: Linux x64 6.8.0-1033-gcp

Does this stop you from using Cursor

Yes - Cursor is unusable

2

Can you please update to the latest version (1.4.5) and see if the issue persists? The team has shipped a few improvements.

GPT 5 has been working great for a lot of people, but since it’s sometimes slower than other models, so I personally use sonnet 4 for some tasks. GPT 5 High has been great to create plans and brainstorm with though.

3

I am using Gemini 2.5 pro and claude 4 and still facing the same issue.

The LLM first looks at code , thinks for a long time and then again thinks , makes increments in 1 line. Can’t do major fixes in 1 file like earlier .

4

@VIVAN-RANGRA I’ve seen gemini 2.5 pro do that. Can you try claude 4 (the non-thinking version)? That works pretty fast and has almost similar performance in my personal testing. GPT 5 also works well.

I’ve also seen this happen when you try to do too many things in one request. Try to break down the things you need to do, solve a specific problem in one chat. You can open another chat for the next problem.

Hope this helps!