Not sure if this is a bug, but since a few days now, the response times for Cursor have gotten crazy slow. I’m waiting minutes for a response sometimes and it’s really slowing down my dev cadence.
I’m mainly using Gemini Pro 2.5 but have tried on Sonnet 3.7 with similar issues.
It’s not context saturation as I fired up a fresh test project and it was still super slow.
Have other users noticed a similar issue with huge delays before Cursor responds to a prompt?
Is it a recognised issue?
For reference: I’m on a MacBook Air M1 8GB on Sequoia 15.4.
I’m going down a route of reducing the payload for running prompts first by consolidating code etc and just trying to slim down the project size… will let you know if that helps.
What’s the largest project size you’ve developed in Cursor and did that have any issues as it grew?
I’m experiencing the same thing with Gemini. I think they’ve introduced rate limiting as the responses are no longer instant; I’ve seen people on Reddit discussing the same issue, so it sounds like they’ve finally put some major rate limiting in place for Gemini 2.5.pro Sad times.
Gemini 2.5 Flash is working fast, but it’s nowhere near as good as the regular 2.5 Pro.
It’s also dog slow with 3.7. It’s hopeless like this. Literally no point in using it as I need LLMs with some grunt.
I’ve tried consulting with o3 on GPT and spent a few dollars on o3 in Cursor as well where it was deemed that there were project inefficiencies, but I can’t see how that’s the root cause as I fired up a fresh empty project and it was still chugging.
I’m still working on project efficiencies now as I think the health check scripts I put in place were not effective enough, but this really ■■■■■ as I’ve wasted a whole day trying to resolve and am still not clear on the actual issue.
While writing this I did a raft of changes and then restarted Cursor and ran a prompt - 225 seconds before it responded. Then just switched to auto and it responded immediately. Although, I’ve no idea what model it used for this. I’m going to have to trust auto mode for now as it’s the only way of getting anything from it and just hope it does not completely trash my project!
Hey Matt, if you’re not on privacy mode, next time a request takes this long please send over the request ID as a reply to this comment so I can take a look at what’s going on!
Thanks for the offer, but I’m developing stuff that’s too sensitive to share. It’s actually got a little better and a business partner has said he had the same issue.