This is a very confusing business logic. When you upgrade from a monthly subscription to an annual one, once you’ve exceeded 500 requests, you’re no longer allowed to use the model normally. The support team keeps replying in different ways, suggesting that I use the 【0.04-per-call】 option instead.
I simply cannot understand this business model. Under the 【20】 plan, even if you exceed the 500 fast requests, you can still get a response after waiting 1–3 minutes. But under the 【192】 plan, it almost always gets stuck at “Generating…” , and after about 20 minutes, the session automatically closes.
I tried to analyze Cursor’s business logic: with the 【20】 plan, if users find the service unsatisfactory, they can simply choose not to renew the next month. There’s no obvious refund option, but at least the financial loss is small. However, with the 【192】 plan, even if users find it unusable, they can’t cancel until a year later.
So this business logic cleverly prioritizes acquiring 【new users】 , which makes user growth numbers look better. Other than that, I honestly can’t think of any reason why they would design such a 【u.s.e.r-h.o.s.t.i.l.e] system.
I used to use Max to change it. I spent a lot of money, and it was going to be successful. It was just 500 , I found a small problem. I went on (the default is to turn off usage billing) . As a result, I used some model automatically, and it was completely changed, so 500 is the watershed.
I’ve seen other posts where some people claim that using more than 500 requests is considered “abusing the API,” while others say they’ve never even reached 500 requests in months. Interestingly, these more “positive” posts seem to be ranked higher in visibility.
Here’s what I want to say:
Since Cursor offers this service to users, using 100, 200, or even 500 requests should not be considered “abuse.” From a user’s perspective, Cursor is simply a service provider. As long as the user is not breaking any laws, how they choose to use the service they paid for shouldn’t be judged or restricted as “abuse.”
The slow request behavior described on the official website worked just fine under the $20/month plan — it was slow (1–3 minutes), but it always responded. However, under the $192/year plan, I’ve experienced an almost 100% failure rate — the model doesn’t respond at all.
If there is such a hidden business logic, it should be clearly disclosed on the website. The lack of transparency is frustrating and misleading for paying users.
What they really care about is likely new user registration numbers — possibly to make their user growth metrics look better for financial reports or investors.
Since there’s no refund option anywhere on the official site, it feels like their mindset is:
“We’ve already got this person’s money — now let’s focus on getting more new users to sign up and pay. Once they pay, we shift resources to the next batch of newcomers.”
This cycle keeps repeating.
So honestly, my recommendation is:
Just register a new account, use the free 50 requests, and when that’s done — register another account.
That might actually be more reliable than paying for a plan that doesn’t deliver on what’s promised.
Indeed, you don’t know why it’s stuck on “Generating…”, and I don’t know either — even though many people are asking about this issue.
This situation is really interesting. Their way of handling it is not to solve the actual problem, but to solve the person who raised the problem. Interesting.
Yesterday I roughly sorted it out. The “Generating…” issue generally falls into the following categories:
Suddenly becomes unusable.
Becomes unusable after using up 500 requests post payment.
Becomes unusable after upgrading from monthly to annual plan and hitting 500 requests.
Hi, to be clear, you should still have access to the slow pool regardless of if you pay monthly or yearly - it sounds like you may have a bug which is stopping you from using this!
If you get stuck on Generating... again, can you open the console by running the Developer: Toggle Developer Tools command, checking the console tab and seeing if there are any errors there?
Also, starting a new chat and disabling buggy MCP server may get your requests working again!
Thanks again for getting back to me. Sadly, the “Developer: Toggle Developer Tools” suggestion didn’t resolve anything, and it’s a bit strange that the post was marked as “Solution” so quickly.
As I already detailed earlier, I’ve tested many methods, including the MCP-related suggestions. It’s starting to feel like quick recommendations are being made just to close tickets, which is understandable, but not helpful.
I fully uninstalled and reinstalled Cursor after seeing your reply, just in case.
The issue I’m facing looks like this:
1). The $0.04 per request option works flawlessly
2). After upgrading to the $192 annual plan, the model suddenly became completely unusable
3). Emails sent to [email protected] have only resulted in repeated suggestions to use the $0.04 option, in various forms
If I were able to resolve this issue on my own, I would never have come to a forum that hides users’ posts. The fact that I even had to find alternative ways just to get my issue seen makes me feel like a ridiculous fool who has wasted several days on this.
Please take a closer look — it seems to be affecting more users than just me.
Progress update:
Since May 20th, after reaching the 500 fast request limit under the annual plan, 【Claude-3.7-sonnet-thinking】 has been almost completely unusable — even now on May 26th, it’s still nearly 100% non-functional.
Apologies if I was not clear in my last message, the command I suggested was not a “solution” but would help us figure out why your requests are getting stuck.
From the screenshot of the Developer Tools, it does look like there is a bug here specific to Windows, where it is trying to use Git for some context and is unable to do so.
I have passed this to the team, but can you confirm the following for me:
Do you have Git installed on your machine?
Is there a file / folder as this path on your machine: C:\Users\Administrator\AppData\Local\Programs\cursor\.git\HEAD
Does this happen on every project, or just when you have not yet opened a project?
When I searched online or asked other AI models for solutions, the most common answers I got were:
Claude & Gemini have recently been rate-limited by Cursor. The only solution is to wait until Cursor increases the quota again, then slow requests might become usable again.
This issue is not caused by your local setup or network failure, but due to limited resources in the slow request queue. The resources in that queue are being allocated more to new users or other free-tier users. The Cursor client itself has no ability to speed up this part.
Even if you downgrade Cursor version, change timeout settings, switch between streaming/non-streaming mode — none of these solve the core issue, which is “backend throttling”. The only effective workaround is either “use a different model” or “pay for a higher-priority channel”.
Additionally, I’ve verified the following:
Git is installed on my machine.
This file does not exist on my machine: C:\Users\Administrator\AppData\Local\Programs\cursor\.git\HEAD
The issue is present regardless of project:
3.1) Whether there is a project or not: $0.04 requests always work.
3.2) Whether there is a project or not: Claude & Gemini slow requests are almost always 100% non-responsive.
3.3) Whether there is a project or not: Other models (like GPT or auto), though not ideal in my opinion, do work via slow requests.
If Cursor has changed backend behavior for annual users or deprioritized slow requests, that should be made transparent on the website.
The fact that you are stuck on ‘Generating…’ and not seeing the slow requests alert suggests this is some unexpected behaviour, as you would usually get an actual error if something was wrong, or you’d see a prompt telling you that your requests were in the slow pool and you may have to wait a bit for a response.
Your slow requests should be working correctly as expected, so this is a weird one!
To clarify, requests work fine when you are within your allowance?
1、Fast requests (within 500) — work fine.
2、$0.04 pay-per-use requests — work fine.
3、Slow requests (using auto or GPT) — work fine.
PS：Slow requests (using Claude or Gemini) — almost always fail (Stuck at ‘Generating…’).
Since May 30, Claude 3.7 slow requests occasionally (around 1–2 out of every 10) do not get stuck in “Generating…” if I wait for 5–10 minutes. I did not downgrade or reconfigure anything in Cursor; the responses just started coming back on their own. This might mean Cursor has adjusted the backend slow request threshold or increased the quota.
This thread is very frustrating for me for the following reasons:
1、After I upgraded to the annual plan and used up the 500 fast requests, I got completely stuck on “Generating…”
2、I saw many other posts reporting the same issue, so I realized the problem was not caused by upgrading to the annual plan.
3、Just as I was about to remove the phrase “after upgrading to the annual plan” and rewrite the post to focus only on the “stuck on Generating” issue, Cursor’s forum made my post title unsearchable. I also noticed similar posts being hidden.
4、As a result, I decided to keep my original tone and frustration intact, and responded elsewhere so more people could know what I’ve experienced.
When I came back to the forum on May 30, I found that my post could be searched again.
Although the post’s visibility had obviously dropped, and although the issue of being stuck on ‘Generating…’ hasn’t been fully resolved, but I can feel that the Cursor team is continuously upgrading, iterating, and improving.
Also, most of the answers I found online point to the limited resources in the slow request queue.
So for now, all I can do is patiently wait for the Cursor team to fix the “stuck on generating” issue.
1、After waiting for 5–7 minutes, about 1/5 of the requests work normally.
2、After waiting for 5–7 minutes, about 3/5 of the requests are interrupted with “Connection failed. If the problem persists, please check your internet connection or VPN”
3、After waiting for 5–7 minutes, about 1/5 of the requests partially generate a response but then get interrupted by “Connection failed. If the problem persists, please check your internet connection or VPN” After clicking “Try again” and waiting another 5–7 minutes, the same message appears again, and this cycle repeats endlessly.
Only by relying on the 1/5 successful attempts and the recent purchase of “Copilot” have I been able to continue the development of my project.
Claude & Gemini have recently been rate-limited by Cursor. The only solution is to wait until Cursor increases the quota again, then slow requests might become usable again.
