“taking longer than expected...”. almost every conversation

Support Bug Reports
, ,
1

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

Basically, almost every conversation now runs into “taking longer than expected…”. When I send “continue” and wait for a while, the exact same issue happens again, making it impossible to make any progress. Meanwhile, the API usage keeps going up the whole time, which is extremely frustrating. I tried both Codex and Claude Code, and neither of them has this problem.

This issue existed before, but recently it has become extremely frequent, to the point where Cursor is basically unusable. I upgraded to the latest version and also tried older versions, but the same problem persists. It occurs on both my work and home computers, and one of my coworkers also reported encountering the same issue.

Steps to Reproduce

Normal Chat.

Screenshots / Screen Recordings

Snipaste_2026-05-11_23-10-57.png
Snipaste_2026-05-11_23-10-57.png1000×655 9.74 KB

Operating System

Windows 10/11

Version Information

version：2.5.20

For AI issues: which model did you use?

GPT5.5，Composer Fast

Does this stop you from using Cursor

Yes - Cursor is unusable

5

Hey, thanks for the report. “Taking longer than expected” in chat is a known class of bugs that we’re tracking. It’s a client-side indicator that shows up when the first token from the model doesn’t arrive before a timeout. There can be a few reasons, like slow provider TTFT, streaming stalls, or extension host issues. I can’t share an ETA for a fix yet.

To help us investigate your specific case, can you share:

  1. The Request ID from one of the stuck chats. Top-right corner of the chat > context menu > Copy Request ID.
  2. Does it also happen in new short chats, or mostly in long chats with a lot of context?
  3. If you can, try temporarily disabling extensions with cursor --disable-extensions in a terminal, then check if it still happens. Sometimes the extension host blocks the agent loop.

Related active threads, in case it helps to follow along: