Skip to main content
Every request errorbar serves writes a ledger entry with its model, token counts, and cost. Nothing is estimated; the ledger is the exact record of what you spent, down to the micro-cent (1,000,000 micros = $1.00).

​
Viewing usage

In the dashboard, the Usage page brings your billing picture together:

Balance

Your current wallet balance.

Recent top-ups

The payments that funded the wallet.

Recent usage

A paginated table of individual requests.
The recent-usage table is paginated so long histories stay navigable, and each row is one request.

​
What a ledger row contains

Every entry records the following:
Per-request costs are often a small fraction of a cent. The ledger stores full micro-cent precision, so you see the real charge rather than a rounded $0.00.

​
Exporting

Export your full usage ledger as CSV from the Usage page: one row per request, with time, request id, model, prompt/completion/total tokens, cost, and status. Use it for your own accounting, reconciliation, or per-model cost analysis.

​
Reading a charge

Costs on the ledger are computed from the tokens used and the model’s rates. If a model is priced at, say, $0.50 per million input tokens** and **$1.50 per million output tokens (illustrative), a request with 1,200 prompt tokens and 300 completion tokens is charged:
That $0.00105 (1,050 micro-cents) is what appears in the Cost column.
Because settlement is idempotent (keyed on the request id shown in each row), a request never produces two ledger entries or a double charge.

​
Observability vs. billing

The Usage page is your billing record: money spent, to the micro-cent. For performance metrics such as latency, throughput, error rates, and per-model breakdowns, see Observability. The two are separate: observability does not affect what you’re billed.