Include enough information to investigate
Send one report per incident and include the fields that apply:
If neither
responseId nor task_id exists, include the request time, operation, model ID, and the HTTP response that was available.
Check the documented action first
- For HTTP status and retry decisions, use API errors and safe retries.
- For a task that remains pending, use bounded polling and include the saved
task_id. - For reserved or returned API quota, use image generation billing.
- For request capacity, upstream
429, and the availability objective, use traffic capacity and availability.
Service status and response expectations
There is currently no public historical status dashboard or incident-subscription feed. The availability figure in this documentation is an operational objective, not measured public uptime or a contractual SLA. No public first-response time is promised. A complete diagnostic package reduces the clarification needed before an incident can be investigated. Material model or API documentation changes are recorded in the documentation changelog; the current model catalog and prices remain available in the model marketplace.Email API support
Send a redacted report with the request or task identifiers.
Protect submitted data
Review upstream transfer, diagnostics, and retention boundaries.