Rate limits and admission

Requests, pending uploads and review work are limited per organization. A 429 response explains the reason and includes Retry-After. For temporary limits, wait that long and add jitter rather than immediately retrying.

quota_exhaustedinsufficient allowance

The estimated review cost exceeds currently available capacity. Pending jobs reserve units too. Wait for capacity to free up or contact us about your allowance; repeated immediate retries will not help.

queue_capacitypending-work limit

Too many files or bytes are pending. Retry after existing work completes, preserving the original idempotency key.

Rejected submissions do not start model work or charge review units. For large catalogs, use a bounded worker pool and persist every accepted review ID before submitting more files.