Usage shows what your account has called and what it cost. What you can see depends on your role:
- A model owner can view their own usage or the usage of all their models.
- An admin sees their own usage and that of the members they manage.
- A member sees only their own usage.
The server enforces this, so filters can only narrow what you already have access to.
Time range
Choose Today, 7 days, or 30 days, and optionally narrow to a model. The range changes what is counted; it does not reset any budget. Budgets reset on their own periods, shown on the model.
Key figures
The top of the page summarizes the selected range:
- Cost
- Requests
- Tokens
- Success rate
- Cache hit rate
- Average latency
- End-to-end throughput
- Cost per request
Below it, the page charts daily activity, token trends stacked by input, cache, and output, the composition of tokens and of cost, and the outcome of requests.
Per-model table
The table at the bottom has one row per model: requests, token breakdown, cost, cache hits, average latency, data status, and a trend.
- Average latency counts only requests that reported a latency, divided by the number of valid requests.
- Unknown and partial data are labelled. A figure the gateway has not reported yet is marked provisional rather than shown as zero.
How the numbers are counted
- Days follow the Asia/Singapore time zone, so a day boundary is not your local midnight.
- Cost comes from the per-request cost reported by the gateway and is used for budget control. It is not necessarily what a provider will invoice you, and it is not an amount anyone owes.
- All amounts are in USD.
- Usage is tracked per model. Two models backed by the same upstream account still have separate totals and separate limits.
- Filtering by time does not change budget counters. A budget period resets on its own schedule.