Usage and request logs
Aggregated usage statistics and individual request records.
The console provides Usage for aggregate statistics and Request Logs for individual records. Personal and organization spaces are separate. Organization members can see only their own data; administrators can view the whole organization and filter by member.
Usage
The Usage page has four views for the selected date range:
| View | Description |
|---|---|
| Overview | Input / output / cache-read / cache-write tokens, requests, successes and failures, and spending |
| By day | Daily trends for changes in usage and peak/off-peak distribution |
| By model | Each model’s share of usage and spending |
| By key | Usage attribution by API Key for cost allocation between projects |
Usage displays “Input” as input tokens − cache-read tokens, matching the billing convention.
Request logs
Request Logs records each API request:
| Field | Description |
|---|---|
request_id | Unique request identifier; include it when reporting a problem |
| Time / model / channel | Request start time (UTC), model ID, and supply channel type |
| Status / duration | Success or failure, and end-to-end duration |
| Token breakdown | Input, output, cache-read tokens, and cache-hit status |
| Charge | Actual amount charged for this request |
| Key | Name of the API Key used |
Platform request logs do not display request or response bodies. They retain metering and status information only.
Common uses
- Reconciliation: individual charges in request logs sum to the consumption records in the ledger.
- Troubleshooting: filter failed requests by status and provide their
request_idto support. - Cost optimization: find expensive models in the model view; improve low cache-hit rates with prompt caching; move batch work on time-priced models to off-peak windows.