429 rate limit error without reaching rate limit

Hi,

On occasion I’m getting a rate limit error without being over my rate limit. I’m using the text completions endpoint on the paid api which has a rate limit of 3,000 requests per minute. I am using at most 3-4 requests per minute.

Sometimes I will get the following error from the api:
Status Code: 429 (Too Many Requests)
Open Ai error type: server_error
Open Ai error message: That model is currently overloaded with other requests. You can retry your request, or contact us through our help center at help.openai.com if the error persists.

Open ai documentation states that a 429 error indicates that you have exceeded your rate limit which clearly I have not.
https://help.openai.com/en/articles/6891829-error-code-429-rate-limit-reached-for-requests

The weird thing is the open ai error message is not stating that. It is giving the response I usually get from a 503 error (service unavailable).

I’d love to hear some thoughts on this, any theories, or if anyone else has been experiencing this.

Cheers

1 Like

Hi @provmusic

It looks like the problem is not on your end, It’s on OpenAI’s end, Server Error, meaning thet the model is getting too many requests.

That’s what it sounds like, but I’m curious why it’s sending a 429 error rather than a 503. I has sent 503s in the past.

1 Like


Why this limit?

429 is the standard “rate limit exceeded” error code, from HTTP

Typically for most services it’s “YOU sent too many requests”

For OpenAI, it seems, it’s “WE received too many requests from everyone”

I think they should rate limit individual users instead of everyone arbitrarily, but I don’t work for OpenAI :confused:

2 Likes

UPDATE: This StackOverflow answered it for me - your free credits expire after 3mo, so if you are on a free key and see 429 regardless of your usage, that’s why. I can’t include a link but this is the SO answer: a/75898717/5298555

The use of 429 for what should be a 503 is a mistake on the API developer’s part, clean and clear. 429 is for USER rate limit (“the user” in Mozilla docs, not “all users” as OpenAI has appeared to have interpreted).

Theirs is a fast moving high demand product, so i can appreciate they have to prioritize issues - but adhering to correct error semantics is (a) easy and (b) crucial to user satisfaction so - not sure why they havent addressed this issue yet. Seems from my reading like it’s been an issue for >1mo.

1 Like

I have a credit card registered. Can’t I make a payment to unblock this? (seems not)

For account related issues please use the support site at help.openai.com and use the Bot icon in the bottom right corner to give your issue and contact details.

1 Like