I’m concerned about the limitations imposed when exceeding the TPM (tokens per minute) usage. If there is only a one-minute restriction at that time, it should be fine. However, I’m curious if there are limitations extended beyond one day, one week, or any other duration. I’m planning to run a special promotion for my service for about four days, and it’s an urgent situation for me. I would appreciate it if someone who knows could provide an answer.
The usage of tokens per minute or requests per minute doesn’t have a memory of over a few minutes, and that “few minutes” is only if you go over the limit by having multiple parallel requests that consume more output tokens than anticipated.
If you have high usage, and do your own rate management so you don’t flood the service with requests the API has to deny, that’s not a concern; that’s what’s being called a good customer.
4 Likes
Thanks for your response
If the number of tokens per minute (TPM) exceeds, is it correct that the limitation occurs only for the specific minute corresponding to TPM, and after that, there won’t be any issues? I’m worried because it’s an initial product, and proactive management might not be fully implemented.
1 Like
You don’t need to be worried,
The rate limits are based on a sliding time window, when you make too many requests within that timeframe, you’ll be unable to make new request’s until that usage has slit it’s way out of the time window 
1 Like
Please also refer to exponential backoff. 
https://platform.openai.com/docs/guides/rate-limits/retrying-with-exponential-backoff
If you exceed the established usage limits, your access to the service features may be temporarily restricted until the end of the current limiting period (day, month). This means that you will not be able to make requests until a new period begins.