After the first prompt, Cursor responds (sometimes with delay), but subsequent prompts consistently hang and never complete. The message “Taking longer than expected” appears and the response does not finish.
The issue happens every time after the first successful completion.
Steps to Reproduce
Open Cursor IDE
Start a new chat
Send a prompt → response comes (sometimes delayed)
Try a different model. Does the issue still happen with Claude Sonnet 4.5 or another model? This will help confirm whether it’s specific to Opus 4.6.
Disable HTTP/2. Open app settings with CMD + , search for “HTTP/2”, and turn on “Disable HTTP/2”. In some network setups, requests can hang on the second call.
Try Stable instead of Nightly. Since you’re on Nightly, can you switch to Stable and see if the issue still happens? (Cursor Settings > Beta > Update Access)
Run network diagnostics. Go to Cursor Settings > Network > Run Diagnostics and share the results here.
Hello, I’ve had the same problem for two days. I’m using Nightly, but it doesn’t work for the stable and beta versions either.
I can’t make any requests anymore, I keep getting the message “Taking longer than expected…”, sometimes I just get the loading message and then it crashes.
However, when I go to my dashboard, it has burned through my tokens, so I have to force Cursor to close to prevent it from burning through all my tokens.
Sometimes, if I close and reopen Cursor after the “Taking longer than expected…” message, the response is displayed in the history and the changes are applied.
This is quite urgent, as Cursor is currently completely unusable.
Hey, also the same issue here. But it is dir specific - I am running more cursor windows on vm through ssh, and in other dirs on the same vm, the cursor works just fine.
The Request ID from the stuck chat (chat context menu in the top right > Copy Request ID)
@vpanek, interesting that the issue is tied to a specific directory. How big is that project? Is there anything unusual about it like lots of files, large binaries, or a non standard structure?
@Picolas, about token usage on stuck requests, this is on our radar. Thanks for flagging it.
Let us know if the issue still happens on the latest Nightly or Stable.
I’m experiencing the same issue. It gets stuck on “taking longer than expected” after running properly for less than a minute. Happens across models as well.
I encountered a similar problem. There is a persistent text of “Taking longer than expected“ and the AI can’t do anything then.
although I can stop the AI by click the button at the bottom right, and prompt it to continue, this problem will appear again in the next round of conversation.
I also encountered this issue and resolved it successfully based on Dean Rie’s reply by using an ignore file. Here’s how: In settings, search for “Ignore Files in .cursorignore”—you’ll see the CodebaseIndexing status above. If it hasn’t reached 100%, first try stopping it. Then, in Chat, ask it to generate an ignore file that fits your project architecture. You can provide the URL to it, such as doc ignore-files. After that, edit the .cursorignore file to paste the generated content and re-index. This should fix the problem, and you’ll find the number of files requiring indexing is greatly reduced.
button at the bottom right, and prompt it to continue, this problem will appear again in the next round of conversation.