Platform

Errors and limits

Common API failures, retry behavior, and verified input limits.

HTTP errors

StatusMeaningAction
400Invalid request or unsupported recorded file.Correct the request before retrying.
401Missing, invalid, or revoked credential.Check the bearer credential.
402Insufficient credit.Add credit before resubmitting paid work.
403The credential cannot perform an account operation.Use a dashboard login session for account routes.
404Resource not found or not owned by the account.Verify the ID and credential.
413Input exceeds a request limit.Reduce the input or use an async flow where supported.
422Invalid field, language, format, or option.Correct the request using the capability guides.
429Rate limit exceeded.Wait for the Retry-After interval.
502The model returned no usable output.Retry only when your workflow can tolerate it.
503A worker, model, or queue is temporarily unavailable.Retry with backoff.
504Processing exceeded the request timeout.Use an async job or retry when appropriate.

Verified limits

InputLimit
Recorded audio upload200 MB
Text-to-speech text10,000 Unicode characters
Voice-cloning reference audio25 MB
Chat messages40 messages per request
Chat message or instructions12,000 characters each
Chat outputmax_tokens from 1 to 2,048
Active API keys5 per account

Usage limits can vary by deployment and account. The API returns 429 with Retry-After when a request should be delayed; do not hardcode a request rate from this page.

Job and WebSocket failures

An accepted async job may later enter failed; inspect its error. Reuse the same ID after a polling network failure rather than resubmitting the work.

Realtime sessions send an error event before closing when possible. Authentication, unavailable transcription, insufficient credit, invalid configuration, and malformed JSON are terminal until the underlying request is corrected.

Next steps

On this page