Skip to main content

​
Request rate

Authenticated API operations share an account-level token bucket: All API keys belonging to the account share the same allowance. This is a refill rate, not permission to send 120 simultaneous requests. The public presets endpoint does not use the authenticated request bucket.

​
Concurrent jobs

Each account has a separate active-job limit. Read concurrency_limit and active_jobs from Get usage. Queued jobs count toward this limit. A create request may return 429 with either:
  • RATE_LIMITED: too many API requests.
  • CONCURRENCY_LIMIT: too many active jobs on the account.

​
Back off and retry

Both return a Retry-After header, currently 5 seconds. Wait at least that long before retrying. For a concurrency limit, also wait for an active job to complete, fail, or be canceled. Use bounded retries with increasing delays and jitter for repeated transient failures. For create requests, preserve the same request body and idempotency key throughout all retry attempts.