Docs / For renters
Rate limits and caps
The fixed limits on keys, requests, length and rentals, and how to stay under them.
These limits are enforced on the server and are the same for every account.
| Limit | Value | What happens past it |
|---|---|---|
| Requests per key | 60 a minute | 429 rate_limited, retry after 5 seconds |
| Requests in flight per key | 8 at once | 429 too_many_concurrent, retry after 2 seconds |
| API keys per account | 10 | Creating an eleventh is refused. Revoke one first. |
max_tokens | 4,096 (1,024 if you leave it out) | Larger values are cut to 4,096, not refused |
| Time for one request | 120 seconds | The reply stops and ends with finish_reason: "length" |
| Request body | 2 MB, 256 messages | 413 request_too_large or 400 invalid_request |
| Open dedicated rentals | 3 | Starting a fourth is refused |
| Smallest credit top-up | $5 | The Billing page will not start a smaller checkout |
Staying under them
- Back off when you see a 429 or 503 and wait for
Retry-After. Add a little random delay so parallel workers do not retry together. - Limits are per key. If one service needs more than 60 a minute, that is a sign to talk to us through Support before spreading load across keys.
- Revoking a key takes effect within 5 seconds, so a leaked key can be shut off quickly from API keys.
Something here wrong or missing? Tell us on Support.