Rate limits and TPM caps

Requests can be rate-limited at two levels: platform limits that protect the gateway, and per-key limits you set yourself on API keys (including tokens-per-minute caps). A limited request returns HTTP 429 — back off and retry.

If you’re hitting limits in legitimate steady-state traffic, open a ticket and tell us your throughput profile — include a request id of a 429 from Usage. The docs cover limit headers and retry guidance.