Limits apply per API key and are enforced at the edge before the request reaches an upstream provider, so you get a fast 429 instead of a slow timeout.
| Tier | Requests / min | Concurrent streams | Burst |
|---|---|---|---|
| Starter (default) | 60 | 10 | 20 |
| Growth ($500+ lifetime top-up) | 600 | 60 | 120 |
| Scale ($5,000+ lifetime top-up) | 3,000 | 300 | 600 |
Response headers
X-RateLimit-Limit: 60
X-RateLimit-Remaining: 42
X-RateLimit-Reset: 1761523260
Retry-After: 12
Rate limits are separate from provider limits. If an upstream provider throttles us, the request is retried on another healthy channel for the same model before surfacing an error.
技术文档以英文维护。