Last week I was able to work in cursor, using my own Gemini API key, with some errors from time to time. Mid-day on friday it just stopped working. Since then im getting 0 response from gemini API.
As a communty that pay for their software It’s a total lack of respect on their part and a lack of communication.
We have repeatedly requested an update on the status of this correction without receiving a response.
1 Like
Hey everyone,
When using your own API key (BYOK), you’re subject to Google’s rate limits based on your tier level. You can check the current limits at https://ai.google.dev/gemini-api/docs/rate-limits and monitor your usage at https://ai.dev/rate-limit.
These are limits imposed by Google’s infrastructure, not something we control on our end. Gemini does tend to experience more capacity constraints than other providers. This is also true even outside of BYOK scenarios.
If you’re hitting limits frequently, you can request a higher tier from Google, use Gemini during off-peak hours, or switch to a different model when Gemini is unavailable.
If you think there’s no way you’re bumping against a limit, I’d be happy to look at a recent Request ID and pass along the exact error message from the 429 we receive from Google’s APIs, which makes it clearer why this was returned.
Hi @Colin, please have a look at the earlier messages.
There are mulitple people reporting that they are in tier 3 on Gemini and far away from the limits.
There is also confirmation from @deanrie that this issue is the Cursor issue
The main cause is that a regional endpoint is being used instead of the global one for the Gemini API, which leads to problems with dynamic quotas.
Hey @RafalP
I’m doing my best to investigate with fresh eyes and with the knowledge we have today! Do you have a Request ID you can share?
Sure,
Thank you.
These 2 generated the error in the last few minutes:
1aba640a-8994-461f-8ba8-0500d3ad74b1
31605d4e-f039-4071-a981-5d92780945ad
I am using api key that is generated from Tier 3 account.
Thanks @RafalP!
For some reason, some of our logs that would typically show more logs are redacted. This is unrelated to Privacy Mode settings.
It’s still clear that the provider is returning a 429. It would be great if someone else facing the issue could provide a Request ID as well (and if those are redacted, maybe that points to the issue).
Checking Google’s Developer Forums, there are a fair number of users who report 429s when they don’t expect to receive them.
1 Like
Hi Colin,
The 429 errors for Tier 3 Gemini happen only in Cursor.
My production system work correctly using Gemini models without experiencing any issues related to hitting quotas for all Gemini models. Also Google’s dashboard show extensive available capacity for more API calls and the quotas are far from being reached.
My own investigation indicates that Cursor is hitting some regional endpoints offered as preview for free users that are subject to different restrictions instead of using global endpoints where the quota is calculated at account level for paid tiers.
Hope this helps.
1 Like
I can confirm that Cursor hits https://generativelanguage.googleapis.com/v1beta/openai/ in all cases where we see 429s returned (which is a funny way to phrase it, as that is the only URL we are hitting—we see lots of successful requests as well!).
Yes, the model starts returning a response and in the middle of it breaks with 429. Retries may immediately break with 429 or exhibit the same behavior, output a few tokens and then interrupt.
The fact the error returns OpenAI is also weird since the selected model is Gemini but probably that’s more related to your deployment and code.
1 Like
Some news or estimated date to resolve this. @Colin
Today already have the same problem i have a level 3 api key, it is a 0.04% of use and cursor keep saying ”User API Key Rate limit exceeded”. I’m not even close to the limit.
1 Like
Hey @Bumpymaze
Have a Request ID?
I’m still having this problem, but less so than in the previous month.
Except that with the new cursor update i don’t get a “Try again” button and i don’t get an error ID, i only see “User API Key Rate limit exceeded” alert
@Colin I’d love to do it, but since the last Cursor update, it’s no longer possible. We only receive this message as a result.
You’ve taken away the retry button. Great engineering! You’ve shown you can recognize 429 which is quite an achievement.
What about fixing the real problem instead.
1 Like
This doesn’t make any sense. I’m paying $60 for a product that doesn’t function properly. It’s ridiculous. If you’re going to offer this product, please remove the “Gemini” name from the homepage. You are not actually providing that model, or it only works half the time.
1 Like
You should be able to copy the Request ID from the chat!
Hi @Colin, I have the same problem. Some request id’s: 5bc153e2-a71e-419f-b290-edb336bc09a0, c2e21841-4007-41e0-8795-bd780e37d078, d7a81132-b854-4e51-babe-7bc8ee1cb4ed