Account & Security
Rate Limits
Account-level request limits and how the API behaves when you exceed them.
Default limit: 100 requests per minute
Every account has a limit of 100 requests per 60 seconds, measured on a sliding window.
The limit applies to the account as a whole. Requests to different models draw on the same allowance.
Rate limit errors
When a request returns HTTP 429, check the Retry-After response header for the required wait time in seconds:
- Account rate limit: returns a
Retry-Aftervalue giving the seconds remaining in the current window. - Model capacity: returns
Retry-After: 5when the model is temporarily overloaded.
Actions to take:
- Wait for the duration specified in the
Retry-Afterheader before retrying - Slow down your request rate if encountering account limits
A model can occasionally reach capacity independently of your own rate limits. This returns HTTP 429 with a shorter Retry-After interval of 5 seconds. Clients should retry after the interval rather than treating it as a hard failure.