All AI models are completely unusable. When processing a task, the conversation abruptly stops midway through without completing. The status shows “Taking longer than expected” indefinitely, and the model never resumes or returns any response. This affects every available model — no model is functional.
Steps to Reproduce
Open Cursor IDE on macOS
Start a new chat or agent session with any model (tested across multiple models)
Submit a prompt or task for the model to process
The model begins responding but stops midway through the output
Status message displays: “Taking longer than expected”
The model never recovers — no further output is generated, and the session hangs indefinitely
Expected Behavior
The model should complete the task fully and return a complete response without interruption. If a timeout occurs, it should either retry automatically or display a clear error message with an option to retry.
For AI issues: add Request ID with privacy disabled
5d223ee6-a764-4b03-9d79-d50b1396117a
Additional Information
Affected models: All models (auto mode and manually selected)
Frequency: 100% reproducible — happens every time on every model
Impact: Cursor is completely unusable for any AI-assisted task
Request ID: 5d223ee6-a764-4b03-9d79-d50b1396117a
OS: macOS
Severity: Critical — blocks all usage of Cursor
Hey, this is a known issue where the backend generates the response, but the client fails to render it. A few things to try:
Resync your project index: Cursor Settings > Indexing > Resync Index. This has fixed it for several users.
Disable HTTP/2: Add "cursor.general.disableHttp2": true to your settings.json via Cmd+Shift+P > Preferences: Open User Settings (JSON), then restart Cursor.
Run network diagnostics: Cursor Settings > Network > Run Diagnostics, then share a screenshot of the results.
Test in an empty folder: File > Open Folder > pick an empty directory, start a new chat, type “hello”. Does the same thing happen?
Update to 2.6.20. You’re on 2.6.19, and the latest stable is 2.6.20.
Are you using a VPN or proxy? If yes, try without it.
was facing similar issues yesterday but then - magic - it just started working again without probs. Some minutes ago my agent stucks and leaves me with the idea to wait or change to auto model where I’m already with. Further UI says I cannot update (current version 2.6.19) what is puzzling since the latest stable is 2.6.20 ?!
Thank you for the suggestions. I’ve tried all of the steps one by one, but the issue still persists. Here’s a summary:
Re-sync Project Index Tried, issue not resolved
Disable HTTP/2 Added "cursor.general.disableHttp2": true to settings.json and restarted, issue not resolved
Network Diagnostics Done, please see the screenshot above
Test in an Empty Folder Tried, the same issue occurs
Update to 2.6.20 Updated, issue still persists
Additional context regarding my network:
I’m currently using a VPN, and due to work requirements, I’m unable to disable it.
However, the key thing to note is:
Before today, Cursor was working perfectly fine with the same VPN setup. This issue only started occurring today.
So the VPN itself doesn’t seem to be the root cause. Could there be something on Cursor’s end — such as a server-side change or an update rollout — that might have introduced this issue? Any further guidance would be greatly appreciated!
Yes, I went through the whole day without it working, tried all sorts of methods, but then it suddenly started working again. I hope you can resolve your issue soon as well!
Glad it’s working again. From the logs, the backend was actually generating responses - the issue was on the client side failing to render them. This is a known bug and the team is tracking it.
Disable HTTP/2: add "cursor.general.disableHttp2": true to your settings.json and restart Cursor
Resync Index: Cursor Settings > Indexing > Resync Index
Test in an empty folder
If none of that works, the issue can be transient and resolve on its own, like it did for @houshanren. Mega-thread for tracking: Taking longer than expected
Tried, issue not resolved