Errors

Crustoff returns errors in the OpenAI error envelope, so the official SDKs’ error handling works unchanged:

{
  "error": {
    "message": "Insufficient credits. Top up your balance to continue.",
    "type": "insufficient_quota",
    "code": "insufficient_quota",
    "param": null
  }
}

Status codes

HTTP	`type`	`code`	When
`400`	`invalid_request_error`	—	Malformed body, bad `size`/`n`/`response_format`
`401`	`authentication_error`	`missing_api_key`	No `Authorization` header
`401`	`authentication_error`	`invalid_api_key`	Unknown key
`401`	`authentication_error`	`inactive_api_key`	Key revoked
`402`	`insufficient_quota`	`insufficient_quota`	Balance too low for the request
`403`	`permission_error`	`inactive_account`	Account suspended
`404`	`invalid_request_error`	`model_not_found`	Unknown model, or wrong modality for the endpoint
`429`	`rate_limit_error`	`rate_limit_exceeded`	Over your RPM limit
`502`	`server_error`	`worker_error`	The model worker failed (you are not charged)
`503`	`server_error`	`model_cold`	No GPU is warm yet — retry shortly

Notes

model_not_found on the wrong endpoint. Calling /v1/chat/completions with an image model (or /v1/images/generations with a text model) returns 404 model_not_found. Check the model’s modality in /v1/models.
You are never charged for our failures. worker_error and model_cold release the credit hold in full.
x-prox-request-id. Every response carries this header — include it if you contact support about a specific request.

Rate limits