429 that should be a 500 in chat endpoint (GPT-4)


When using the chat endpoint, occasionally I receive HTTP 429 errors with the following error body:

{"error":{"message":"The server had an error while processing your request. Sorry about that!","type":"server_error","param":null,"code":null}}

I have verified that all the rate-limit headers are fine (I am not going over the rate limit), so it feels to me this should be a 500 instead of a 429. I have also read the rate limit guide and confirmed that I am not going over GPT-4-8k’s current limits of 40k TPM or 200 RPM; they are both significantly lower.

I am using the GPT-4 model for these requests, if that matters.

It appears others have experienced similar issues with 429 and this specific error message: Error: status code 429, Too Many Requests - #3 by tomashby


We are experiencing the same, we are getting this error very frequently, even when testing with 5 requests per minute which is way lower than the allowed 200 per minute.


We are having the same issue. It’s very slow. So we put 10 second gap between two successive calls. Still we get 429 error. Would love someone from OpenAi to chime in.

We are running into the same issue. We are probably sending 2-3 requests each minute (1000 token range for each request) and are encountering the 429 error.

I’m going to send this thread to OpenAI support so we get some insight into this issue. In the meantime, I have added exponential backoff retry logic to API requests and it seems good as a workaround so far.


If u have a free account, know this: after 3 mo all your credits expire so u will get 429 regardless of usage. This was the SO answer that helped me: a/75898717/5298555 (I can’t include links)