DocsAPI reference

Errors

Errors use OpenAI's format and HTTP status codes, so SDK error handling works unchanged.

402 Payment Required

{
  "error": {
    "message": "Insufficient credits: this request may cost up to $0.0563 (32768 output tokens reserved), and your available balance is $0.02. …",
    "type": "insufficient_credits",
    "param": null,
    "code": "insufficient_credits"
  }
}

Status codes

Status	code	Meaning	What to do
400	(none) / context_length_exceeded	The request is invalid, or the model rejected it.	Fix the request. Not charged.
401	missing_api_key, invalid_api_key, api_key_revoked	No usable API key.	Send a valid key.
402	insufficient_credits	The available balance cannot cover the request's reservation.	Add credits or lower max_completion_tokens.
403	key_spend_limit	The key would pass its spending limit.	Raise or remove the limit in the dashboard.
403	account_disabled	The account is suspended.	Contact support.
404	model_not_found, unknown_endpoint	Unknown model id or path.	Check GET /v1/models.
413	request_too_large	The body is larger than 4 MB.	Send a smaller request.
429	rate_limit_exceeded	The key's per-minute request limit is reached.	Wait for retry-after seconds.
503	upstream_unavailable	Every deployment of the model failed.	Retry with backoff. Not charged.
503	model_unavailable	The model is temporarily out of service.	Retry later or use another model.
503	database_unavailable, catalog_unavailable	A temporary internal problem; the request was not sent.	Retry with backoff. Not charged.
500	internal_error	Unexpected failure.	Retry; contact support with the x-request-id.

Retry 429, 500 and 503 with exponential backoff; the OpenAI SDKs do this by default. Avenro already retries failing deployments for you before answering, so a 503 means every deployment was tried. Don't retry 400, 401, 402, 403 or 404 without changing something.

Errors in the middle of a stream arrive as a final chunk with an error object; see streaming.

Status codes

Retries