Usage Analytics

Where the spend went: by model, by key, over time. Useful for finding the one prompt that quietly became the largest line item.

In the console sidebar (it appears once the feature applies to your account): Activity Open it ↗

Steps

1. Open Activity. The default view is recent usage across the workspace.
2. Switch tabs: Overview for the summary, Usage for request and token volume, Spend for cost. Performance — latency and errors — appears where it is enabled for your workspace.
3. Filter by model to compare what each one actually costs you, rather than what its list price suggests.
4. Narrow the date range to isolate a deploy or a traffic spike.

Limits and gotchas

• Most panels read an hourly rollup, so the current hour looks incomplete until it closes. Per-key drilldowns query raw records instead and are not hour-quantised.
• The model filter groups a model with its variants, so a family reads as one line rather than several.