Skip to main content
The console usage log shows the time, model, tokens, latency, result, and charge for each request. When reconciling usage, confirm the selected time range and time zone first.

​
Common fields

​
Input, output, and cache

  • Input tokens include prompts, conversation context, and tool information.
  • Output tokens are generated by the model.
  • Cached reads are tokens read from a context cache; they can be priced differently from normal input.
Cache support and pricing differ by model. Use the live information in the Model Marketplace.

​
Why a log can be delayed

There can be a short delay between request completion, accounting, and the console query. If a finished request is not visible immediately, refresh later instead of sending the business request again just to create another log.

​
Errors and charges

A failed request is not automatically free. Errors that produce no billable result are commonly recorded as zero, but an upstream request may already have run, a client may have disconnected, or the client may have submitted a new request. Reconcile using the Request ID, time, and final log record—not only whether the client displayed content. For a suspected duplicate or unexpected charge, provide the Request ID, time and time zone, model, and whether the client received a complete response. Never provide a full API key.

​
View or export usage

  • Use the console overview and usage log for daily review.
  • Filter detailed records by time, model, and group.
  • Use the usage-log API for programmatic account-level reconciliation.
Keep the Request ID after exports or summaries so that an unusual record can be traced later.