Rate limits
All keys in an organization share its concurrency limit and credit balance. On per-minute plans, each key has its own core rate window. Management limits apply across the organization on every plan.
Headers and 429
A request that arrives while every concurrency slot is in use returns429 with these headers:
Retry-After seconds, before trying again.
On per-minute plans, a rate-limited request returns:
X-RateLimit-Reset is a Unix timestamp in seconds. Per-minute retry delays range from 1 to 60 seconds.
Size your worker pool to your concurrency limit, and add jitter to retries. Missing or invalid credentials can also trigger authentication protection; fix credentials instead of increasing retry traffic.
Concurrency caps for batches and the number of monitors are separate from these request limits.