Rate limits
How many requests you can make, how to read your current window, and what happens past it.
Limits protect the service for everyone. They count per API key, or per signed-in account in the console, and per IP address.
| Limit | Default |
|---|---|
| Chat completions | 60 a minute per key |
| Chat completions running at once | 10 per key |
| Model list and attestation | 300 a minute per key |
| Storage API | 3,000 a minute per key (50 a second) |
| S3 endpoint | 3,000 a minute per S3 access key (50 a second) |
| Any request, per IP address | 3,600 a minute |
Need more? Write to support@teepin.com with your account number and what you are building.
Reading your window
Answers carry the window they counted against:
| Header | Meaning |
|---|---|
X-RateLimit-Limit | Requests allowed in the window |
X-RateLimit-Remaining | Requests left in it |
X-RateLimit-Reset | When it resets, in Unix seconds |
Past the limit
The request is refused with 429 and a Retry-After header in seconds. For inference the body is OpenAI's error envelope with the code rate_limit_exceeded (too many in the window) or too_many_concurrent_requests (too many at once). The S3 endpoint answers SlowDown, as S3 does.
The SDKs wait and retry on 429 for you, up to their retry limit.