Skip to main content
The main gateway limit is 600 requests per minute per API key for /v1/*. Requests from all models using the same key share the 60-second window. When the limit is exceeded, the gateway returns HTTP 429 with:
  • Retry-After: suggested wait time in seconds.
  • RateLimit-Limit, RateLimit-Remaining, RateLimit-Reset: quota and reset window.
Rate-limited requests do not reserve or consume balance.
  1. Use Retry-After when present.
  2. Otherwise start at 1 second and double each attempt (1, 2, 4, 8 seconds) with jitter.
  3. Cap retries, for example at 5 attempts.
  4. Queue work in your app instead of sending bursts.