Rate limits

Limits apply to the authenticated project/key and are returned in response headers when available.

x-ratelimit-limit-requests: …
x-ratelimit-remaining-requests: …
x-ratelimit-reset-requests: …
x-ratelimit-limit-tokens: …
x-ratelimit-remaining-tokens: …

The exact policy is server-defined and can vary by environment, project, and operational safety state. Treat headers as the current authority; do not hard-code a public quota in your client. 429 rate_limit_exceeded includes Retry-After when an authoritative reset is known.

Use a bounded queue, token-aware concurrency, and jittered backoff. Requests rejected before reservation are not billed. If a stream has already emitted customer-visible output, a retry can create a second billable attempt and should require explicit application policy.