Errors

Errors use the OpenAI shape. OpenAI SDKs raise their usual exceptions.

{
  "error": {
    "message": "Rate limit reached: 60 requests per minute. Retry in 12s.",
    "type": "rate_limit_error",
    "param": null,
    "code": "rate_limit_requests"
  }
}

Codes

Status	Code	Meaning	Retry
400	`invalid_json`	The body is not valid JSON.	Don't retry
400	`invalid_request`	A parameter is missing or has the wrong type.	Don't retry
400	`context_length_exceeded`	Prompt plus reply exceed the 16,384-token context window.	Don't retry
400	`unsupported_parameter`	We don't support this parameter, e.g. `n` > 1 or `logit_bias`.	Don't retry
400	`unsupported_content`	A message has non-text content, e.g. an image.	Don't retry
401	`missing_api_key`	No `Authorization: Bearer` header.	Don't retry
401	`invalid_api_key`	The key does not exist.	Don't retry
403	`key_revoked`	The key was revoked. Create a new one.	Don't retry
403	`account_disabled`	The account is disabled. Contact us.	Don't retry
404	`model_not_found`	Unknown `model`. See `GET /v1/models`.	Don't retry
404	`unknown_endpoint`	We serve chat completions and models only.	Don't retry
429	`rate_limit_requests`	Over 60 requests in a minute.	Retry after Retry-After
429	`rate_limit_tokens`	Over 500,000 tokens today.	Retry after Retry-After
429	`concurrency_limit`	Over 2 requests at once on this key.	Retry after Retry-After
500	`internal_error`	Something failed on our side.	Retry with backoff
503	`model_unavailable`	The model is offline or loading.	Retry with backoff
503	`engine_busy`	The model is at capacity.	Retry with backoff
503	`model_timeout`	The model did not answer in time. Try a shorter prompt.	Retry with backoff
503	`stream_interrupted`	The model stopped mid-reply.	Retry with backoff

When to retry

429: wait for the Retry-After header, in seconds, then retry.
500 and 503: retry with exponential backoff, for example 1 s, 2 s, 4 s.
Other 4xx: don't retry. Fix the request.

Rate-limit headers

Every accepted request returns these headers. A 429 returns Retry-After instead.

Header	Meaning
`x-ratelimit-limit-requests`	Requests allowed per minute.
`x-ratelimit-remaining-requests`	Requests left in the current minute.
`x-ratelimit-reset-requests`	Time until the minute resets, e.g. `12s`.
`x-ratelimit-limit-tokens`	Tokens allowed per day.
`x-ratelimit-remaining-tokens`	Tokens left today.
`x-request-id`	This request's id. Include it when you contact us.

Errors in a stream

If the model stops unexpectedly during a stream, the last event is an error, not [DONE]:

data: {"error":{"message":"The model stopped unexpectedly. Retry the request.","type":"server_error","param":null,"code":"stream_interrupted"}}

Check each event for an error field before you read choices.