Docs / Reference
5xx errors
The difference between a model failure and a Cent AI failure, and when to retry.
Last updated
A 5xx response means the request was valid but could not be completed. Cent AI tries the other routes for the model before it returns an error, and only before the first byte of the answer.
503 upstream_unavailable
"The model is temporarily unavailable. Try again shortly."
Every route for the model failed. Nothing is charged for a request that failed before the first byte. Retry after a short wait, honouring Retry-After when it is present, or switch to another model. The status page shows per-model status.
503 upstream_rate_limited
The model is rate limited on every route at the moment. This is not your key's rate limit. Retry later.
503 upstream_timeout
The model did not start answering within the time budget for the first output, or, inside a stream, the stream reached its maximum duration. See Stream timeout.
upstream_error
The model received the request and rejected it, for example because of a parameter it does not support or its own safety system. The status is the model's own; a 502 or 504 from the model is sent as 503. The message is always "The model rejected the request."; check the parameters you sent, change the one the model does not accept, and retry.
Cent AI passes your parameters to the model as you send them and does not filter content itself.
500 or 503 internal_error
An error inside Cent AI. The request was not sent to a model and nothing was charged. Retry; if it continues, write to support@cent-ai.dev with the x-request-id.
No 502 and no 504
Cent AI never sends HTTP 502 or 504. If you see one, it did not come from Cent AI: check proxies between your client and api.cent-ai.dev.
Seeing the details
Open Requests in the dashboard and find the x-request-id. The request shows its status, error code and, if it ran, the tokens and cost.