Rate Limits
Applicable limits
Section titled “Applicable limits”Limits depend on account, API key, model, and site configuration. This guide does not promise a fixed requests-per-minute, token-per-minute, or daily allowance. Ask support about the limits applicable to your workload.
HTTP 429
Section titled “HTTP 429”When a request is rate-limited, reduce request frequency and concurrency. Use bounded exponential backoff with jitter. Honor a valid Retry-After header when present, but do not assume it or any X-RateLimit-* header is always returned.
For an existing video task, retry the status GET using the same task ID. Do not repeat the submission POST as part of a polling retry.
Polling practice
Section titled “Polling practice”- Start with a moderate interval, such as five seconds, and increase it for long-running tasks.
- Limit the number of simultaneous polling requests.
- Stop when a task reaches
completedorfailed. - For
unknown, retain the task ID and contact support instead of resubmitting. - A local deadline stops your polling loop; it does not cancel the task.
Requesting higher limits
Section titled “Requesting higher limits”Contact support with expected request frequency, concurrency, model IDs, and typical task duration. Never include your secret API key.