Errors and limits

Application errors return an error object:

{
  "error": {
    "message": "Human-readable description",
    "type": "invalid_request_error",
    "param": "input",
    "code": "unsupported_tags"
  }
}

Error types

Type	Condition
`invalid_request_error`	Authentication, invalid input, or a missing resource
`billing_error`	Insufficient credit or a spending limit
`rate_limit_error`	Request-per-minute or concurrency limit
`server_error`	Server response with status 500 or higher

Common error codes

Code	Meaning
`invalid_api_key`	Key is missing, invalid, expired, or revoked
`invalid_language`	Language is not supported
`model_not_found`	Model ID is unknown
`unsupported_tags`	Input contains an inline performance tag
`transliteration_unavailable`	HTTP 503: enabled transliteration worker is unavailable
`rate_limit_exceeded`	Requests-per-minute limit was reached
`concurrency_limit_exceeded`	Concurrent request limit was reached
`insufficient_credits`	Account balance cannot cover the request
`spending_limit_exceeded`	Project or key spending cap was reached

Pydantic field validation returns a detail array instead of the application envelope. Each item identifies the invalid location, value, and validation message.

Project limits

New projects default to 60 requests per minute and 4 concurrent requests. Project settings may override both values. Limit failures return HTTP 429 with rate_limit_error and the matching code.

Use x-request-id from a speech response when tracing a completed or partially completed request.

Return to the API overview