Request evidence
Use request-level records to explain performance and spend.
Requests stores status, model, usage, latency, and cost information for each API call. Use a Request ID to locate a problem and connect an API response with its console record.
Find a request
- Open Requests in the console.
- Select a time range that includes the API call.
- Use available filters such as model, status, or API key.
- Open the matching record to review its details.
When reviewing activity across time zones, first confirm the time and time zone displayed by the page.
Record fields
| Field | Purpose |
|---|---|
| Request ID | Locates one request on the platform |
| Time | Helps find the request and review call order |
| Model ID | Identifies the model used for the call |
| Status | Distinguishes completed, failed, and processing requests |
| Input and output usage | Shows token or other metering information |
| Latency | Shows request processing time |
| Cost | Shows the amount and settlement state for the request |
An HTTP status is not always the final settlement state. Missing cost, zero cost, and unsettled cost may also represent different states.
When cost is higher than expected
- Confirm the actual Model ID and pricing unit.
- Check for long conversation history, images, or a large output.
- Review cache, media, and other separately metered items.
- Check whether the application or SDK retried and created additional requests.
- Keep the Request ID and redacted error details before contacting support.
When a record is missing
Check the account, workspace, time range, time zone, and filters, then refresh the page. A record that is not visible yet does not prove that no usage occurred. Do not resend the same request only to create another record.
When contacting support, provide the Request ID, time and time zone, Model ID, status code, and redacted error. Do not provide a complete API key or sensitive request content.