Skip to main content

​
Errors and rate limits

​
Rate limits

The API applies per-key rate limits to protect stability. Limits vary by endpoint (e.g. inference vs batch). When you exceed the limit:
  • Status code: 429 Too Many Requests
  • Response: JSON with an error message; some responses include retryAfter (seconds).
Recommended behavior:
  1. Respect Retry-After if present; otherwise use exponential backoff.
  2. Cache responses where it makes sense to reduce calls.
  3. For bulk work, use the batch inference endpoint where applicable.

​
Common inference errors

StatusMeaningWhat to do
400Bad request (e.g. missing model or messages)Fix the request body and retry.
401Invalid or missing API keyCheck Authorization: Bearer sk_... and key validity in the dashboard.
402Insufficient balanceAdd funds in the dashboard and retry.
404Model or resource not foundUse a valid model ID from the models list.
429Rate limitedBack off and retry after the indicated time.
500 / 502Server or upstream errorRetry with backoff; contact support if it persists.

​
Error response format

Errors return JSON with at least: 
{
  "error": "Short error code or description",
  "message": "Human-readable details (optional)"
}
For 429, you may see: 
{
  "error": "rate_limit_exceeded",
  "message": "Too many requests. Please slow down.",
  "retryAfter": 60
}
Use the error and message fields to show clear feedback (e.g. “Add funds” for 402, “Sign in or check API key” for 401).

​
Usage tracking

Every successful inference request is recorded for usage and billing. Failed requests (4xx/5xx) are not charged. See Billing & usage and Errors & status codes.