Rate Limits
Rate limits on CloudService are applied per API key. Your API key may include request or token limits. If you need higher limits, contact CloudService support.
What gets limited
- Requests per minute (RPM). Total number of requests, streaming or not.
Headers
Every response carries rate-limit info:
| Header | Meaning |
|---|---|
X-RateLimit-Limit-Requests | Configured RPM cap. |
X-RateLimit-Remaining-Requests | Remaining requests in the current window. |
X-RateLimit-Reset-Requests | Seconds until the current request window resets, written with an s suffix (for example, 27s). |
X-RateLimit-Remaining-Tokens | Remaining prepaid token allowance (token-mode keys only). This is a balance, not a per-minute window. |
The same remaining allowance is readable outside a request through the Usage Checker.
When you hit a limit
The API returns 429 with one of two codes (see Error Codes):
rate_limit_exceeded— requests-per-minute limit. Retry after the window resets.
Recommended client behavior
- Read the rate-limit headers on every response.
- When
Remainingdrops below 10%, slow down proactively — don't wait for a 429. - Retry 429s with exponential backoff + jitter. Honor
Retry-Afterif present. - If you need consistent high throughput, contact CloudService support for a higher quota. Bulk workloads can be split across multiple keys.
Burst behavior
Short bursts above the per-minute average are tolerated up to a configurable cap, then the gateway starts shedding load with 429s. Plan for your steady-state rate, not your peak.