Rate limits and technical limits
Limits protect your balance and shared capacity. The values below are service defaults; you can lower per-key limits in the dashboard.
Default values
| Limit | Default |
|---|---|
| Requests per minute per key | 600 (can be lowered per key) |
| Requests per minute per account | 3,000 |
| Concurrent requests per account | 8 active; the rest queue, and a full queue returns 429 too_many_concurrent_requests |
| Body size | 8 MiB (images and video: 64 KiB) |
| Maximum duration of one stream | 30 minutes |
| Wrong-key attempts per IP | 30 per minute, then 429 |
The service operator can change these defaults.
Retry-After
429 and 503 responses carry a Retry-After header in seconds. Wait that long before trying again. The gateway does not send x-ratelimit-* headers.
Tips
- Agent tools often fire many parallel requests. Lower the parallelism if you see
429 too_many_concurrent_requestsoften. - Use separate keys per app so one app's per-minute limit does not slow down another.
- Video generation is slow and capacity-limited. Run videos one after another, not all at once.