Issue:
Cursor frequently stops mid-completion, hanging indefinitely and never finishing the provided output. This forces repeated retries, significantly reducing productivity and rapidly consuming the available completion quota without yielding usable results.
Impact:
This issue severely affects workflow efficiency. Completions are being used up quickly, but without successfully generating complete responses. Users are required to repeatedly regenerate outputs, which wastes resources and disrupts the development process.
Steps to Reproduce:
Enter a prompt that requires a detailed or moderately long completion.
Trigger completion generation in Cursor.
Cursor begins output but randomly stops mid-sentence or mid-code and hangs indefinitely.
Expected Behavior:
Cursor should consistently complete all generated outputs without interruption, hanging, or indefinite loading. If an error occurs, it should fail gracefully and inform the user clearly, instead of silently hanging.
Recommended Action:
Investigate the underlying cause of Cursor hanging during completions. Implement timeout or error-handling mechanisms to prevent indefinite hanging, preserving the user’s completion quota and improving reliability.
Not sure what’s happened, but the cursor program has slowed to a crawl past couple days. I can’t do anything more at this point. I feel like I’ve updated the app or something, and I haven’t. It’s quite confusing what’s going on.
Edit: More specifically, it feels like I updated to a bad update or something, but I’ve not updated at all; it’s just suddenly changed for seemingly no reason.
I can second this on mac with 0.45 it’s hanging mid generation on claude 3.5 and then timing out and forcing me to try again or pretty often. Probably every other request it’s hanging for a really long time while doing generation.
Yes, sorry I didn’t come back to report. I had found all that. Reverted, but it didn’t work; same issue happens. So I decided to redownload 0.46, and then discovered I had 0.47. Still didn’t work. So it turns out it wasn’t necessarily version-specific. I did notice though that, when first initiating this reverting process, uninstalling 0.46 uninstalled who knows what, because it was still there, even though it went through an entire uninstall process. So I’m guessing, at some point, Cursor installed another version over itself automatically? Is that a thing? Because I always ignore all update messages habitually. So I have no idea this could’ve happened.
But the issue itself – it instantly hangs to the point of error whenever I type anything at all in the chat and click send – is still unresolved. I’m about to start removing cache files and stuff on Windows, I guess.
I second this. I’ve tried to isolate my issue as much as possible by using different networks, reloading the window often, different models (and also selecting “Default”) and even comparing between my Linux and macOS setup, but the issue is still happening often enough that I find myself weighting each question if it’s worth waiting the repeated retries or not…
Hey all, if you could ensure you have privacy mode disabled (at least temporarily), sending over request IDs when this happens will help us check what’s going on under the hood here!
I am experiencing this and yet I cannot provide you the request ID because unlike “usual” generation, once this mid-completion occurs and hangs all the generation up to that point for that particular request generation just vanishes. Leaving the previous request. I imagine providing previous request ID is useless. What’s worse is that it still counts as fast request.
It is fair to not count requests that fail! I feel wronged by paying for something that is being consumed with errors and failures. Be fair and return the requests that failed!
Same problem here, except when I message the agent to ask if it’s stuck, it sometimes says it already finished despite having been stuck in a “Generating…” loop with the spinning circle on the bottom-right of the screen indicating it’s still processing.
I haven’t figured out yet if it’s actually completing its tasks and then getting stuck in a self-check loop or if it’s only partially completing its intended task and then convincing itself it actually finished. Sometimes it introduces new errors and sometimes not.
It’s a series of trial and error, but on the plus side, it’s still faster than manually keying all of the code myself. I suspect this issue only occurs in “slow mode” after having used up all of the pro account fast requests for the month.
Have the same problem. Tried to install 0.47, 0.46, 0.45 versions, tried to change LLMs, tried to stop and continue generation, tried to return to previous checkpoint - always got infinite “Generating” after one or two successful generated files. I’ve got this issue on a purely new project with no old files at all.
@danperks just to clarify, that’s not accurate in my case.
I emailed Cursor support over 8 days ago with visual proof that failed/no-change completions were still consuming fast credits. The convo was quick and responsive until I asked for a refund. Once I showed screenshots of “no changes made” outputs, support went completely silent — no reply since.
I’ll follow up via email again, but wanted to confirm here that these failed requests are being charged, at least in my experience.
its frusttrating… every 3rd request it stucks… i am on pro plan and all things and knowledge thats build up is lost and i have to start again… its unfair if you pay for that service!