Unbearably slow request speeds

Bug Reports
1

I’ve been using Cursor Pro for about a month now and have really enjoyed the fast request speeds—it’s been a great experience overall. When I ran out of fast requests, I switched to slow requests, and they initially worked fine, with the longest wait time being around 15–20 seconds. However, over the past few days, I’ve experienced significantly increased wait times—often 5 minutes or more—for a single request to go through. This has been very frustrating and has made it difficult to continue using Cursor effectively.

Just wondering whats happening and is this a cause of overuse?

4 Likes
2

I asked the same question and @danperks answered to me that if I use the slow request a lot from the 100th or more, the speed will start slower, but at that level, it is only a few dozen seconds, having to achieve the large number of Slow Request can make you fall into the “minute” queue. However, I have not used 100 times the slow request, I have to wait 5 minutes for 1 response (Claude 3.7 Sonnet Thinking and Gemini 2.5 Pro)

4

Hi,

Me too, for a few days now, it’s mega slow, several minutes to get an answer from the Agent/Chat. It’s not usable.

  • I pay for a pro plan
  • It’s happening even if I enable usage-based pricing and even before reaching the 500 prompts limit
  • I tried with VPN
  • I tried Agent / Ask / Manual
  • also by reinstalling Cursor from scratch…

Nothing changes.

1 Like
5

Ive used 300% of my “requests” so almost 1500. Don’t know how many others have used so im really not sure.

6

I was around 450 requests when the problem started.
Now I am slightly above 500, it still takes several minutes for any question in the Agent/Chat.

Auto-complete seems to work normally.

Currently looking for alternatives, as Cline, Windsurf.

7

I downgraded to 0.49.6, and it has fixed the issue!

Running and Macbook Silicon by the way.

2 Likes
8

we are on the same page mine is slow as well

9

Just downgraded however still extremely slow, im on Windows 10

10

Yeah, not sure. I cranked out about 6900 requests or so in April and my slow requests are about the same wait time as yours.

11

Yeah i’m just hoping they can fix this or else i’ll have to move to another ide like co pilot

12

Have you found an alternative solution?
I think it’s because the cursor official has restricted it, so I’m also looking for alternatives.
This slow request is already a decoration

13

same problem

14

I don’t think this is going to be fixed. I think its just by design now. it downgrades to the point where is not usable.

15

i left my slow request running over night and it was generating a response, came back and it just quit generating. so i tried agian came back an hour later and it agian just quit generating. using claude 3.5 sonnet

16

ok i tried auto aswell and it just stops generating. cursor is unuseable right now

17

only solution i found was just purchasing fast requests.

1 Like
18

And let’s mention the fact that you cannot choose when you need a fast request and when you can accept a slow request. So that when your month starts, you can eat up the fast requests in days.

So for me, I basically chose cursor not because of the fast request, but for the unlimited slow request. I would have no fast request at all I’d still keep cursor.

But if I have to wait many minutes for slow requests for an otherwise free and instantaneous model online, that’s gonna be a problem for me.

I have already used for many months in a satisfactory manner, but times to times I have to go to the forum and check why it’s getting really slow. And right now, It’s too slow, I lose too much productivity.

That’s what we mean by a service not reliable: OK sometimes, and just degrading too much sometimes.

On the bright side, while waiting for the answer, I have time to complain in the forum…

1 Like
19

■■■■!
It takes a year to ask one question in cursor while select claude-3.7-sonnet model.

1 Like
20

claude-3.5 sonnet too. Gemini too

21

I am experiencing similar with Google Gemini 2.5 Pro in Cursor. Since the latest update, the model takes between 5 to 10 minutes to provide a response. This has effectively made Gemini unusable for me, even though it is my preferred model. I am on the Paid pro plan.

While I do not mind waiting for slower requests, the current response time is beyond what I would consider slow; it seems as though the integration is broken, or cursor changed slow requests behind the scene rendering this entire app useless, and its pay to play now…

Wish this OFFICIAL Forum would get response from… THE OPFFICIAL TEAM.

this was their ai bots reply to me:
"thanks for reporting this issue!

There is currently some minor instability with the Gemini 2.5 model that can cause slow responses or tool calls to fail. Google is actively working to fix this on their end, and we expect this to be fully resolved soon.

In the meantime, you can either:

Retry your prompt
Try using Auto-select mode, which will pick a reliable model for you
Switch to another model like Claude 3.7 Sonnet for your current tasks
Let me know if you need help with any of these alternatives!"

So I replied:

"Thank you for the update. I understand that the issue stems from instability with Gemini 2.5, and that you’re relying on Google to resolve this. However, the current wait times — often exceeding 10 minutes per response — have rendered the tool virtually unusable for my workflow.

I’m aware there are alternative models like GPT-4 or Claude, but frankly, they are not viable replacements in this context. They lack the deep code awareness, project-wide context, and direct file editing capabilities that make Cursor valuable in the first place. When Gemini 2.5 is down or lagging, the other models simply don’t deliver usable results — they fail at implementation, misinterpret instructions, or produce broken outputs. So suggesting alternate models is not a functional solution unless those alternatives can match Gemini’s integration quality within Cursor.

Given the severity of this issue, I’d like to request the following:

A realistic timeline for when full Gemini 2.5 stability is expected.

An explanation of whether these delays are caused by Cursor-side infrastructure (e.g., shared API keys, rate limits) or are purely on Google’s end.

I value what Cursor offers when everything is working correctly, but the current state prevents me from relying on it professionally. I hope to hear back with some actionable updates soon."

and got back another “Pass the Buck” Response:

"
Hi Jeff,

Thanks for the detailed follow-up. The delays you’re experiencing are due to instability on Google’s end - this isn’t related to Cursor’s infrastructure, shared API keys, or rate limits. While Google is actively working on fixing these issues, we don’t have a specific timeline from them yet. We’re monitoring the situation closely and will update our users as soon as we have concrete information from Google.

I understand your frustration about the alternative models not matching Gemini’s integration quality for your specific needs. For now, if you want to continue using Gemini 2.5, you can try asking it to execute the actions it planned to take when it seems unresponsive, or retry your prompt. We’ll keep pushing for updates from Google on the resolution timeline.

Let me know if you need anything else!

Best,"

None of this fixes anything and its clear from the other responses in this thread is is not just gemini…