The Analytics page provides observability for your pay-as-you-go usage, including token consumption, request counts, latency, success rates, and per-model performance metrics.
Below the usage charts, a per-model table shows model code, average TPM and RPM, total and successful requests, average success rate, average time to first token, and average latency for each model. Use this to identify underperforming models or unexpected error spikes.
Go to the Logs page, or select the Logs tab on the Analytics page, to inspect individual API requests. Only pay-as-you-go call logs are shown, and they are retained for the last 14 days. Logs are scoped to the current workspace and region: switch the workspace in the console sidebar or the region in the console to view other data.
Request logs contain request-level metadata only. They do not record or show HTTP headers, the request body (prompts, images, audio, and other inputs), or the response body (model output). Request details also do not include an error code or error message. To troubleshoot parameter or output issues, use the error code and message returned by the client together with your application logs.
To reproduce a failing request, capture the following on your side and retry:
Request parameters, request body, and response content (never log API keys or other secrets).
SDK debug logs: set DASHSCOPE_LOGGING_LEVEL=debug for Python; for HTTP calls, use curl -v or a local proxy to capture the request.
The Request ID returned in the response, so you can locate the call in Request logs.
Set up custom alert rules to monitor API call metrics and receive real-time notifications when anomalies occur. Alerts are available only in Singapore. For details, see Monitoring alerts.
