Usage and billing details

Usage records show actual requests and charges. One coding task may trigger multiple requests; the tool’s final answer alone does not establish the task’s total cost.

Find requests for a task

1. Note the task’s start and end time.
2. Open Usage records and select the time range.
3. Filter by model, client, or request status. You can also open Usage records from API keys to filter to one key.
4. Total the charges for that range. Separate keys make it easier to distinguish tools.

View billing details

Hover over or click a record’s charge to see its billing breakdown. Open the record to inspect additional request information.

Field	Meaning
Model	Model ID used for the request
Input and output	Request input and generated output token counts
Cache read and write	Separately priced cache usage
Settlement price	Actual unit price used for the request
Charge	Confirmed fee for the record
Status	Review both request status and billing status

Use the settlement unit price in the record when checking individual charges.

Why a charge may differ from the catalog

Model pricing lists standard prices. An account or organization may have special settings, and historical requests retain their prices. Use each record’s billing details and settlement entries when checking existing requests.

When cache read and write have separate prices, their tokens are excluded from ordinary input to avoid double counting. Image models may be charged per image or by text and image tokens. Refer to the billing method shown in the record.

Failed or interrupted requests

Check request status and fees separately. A request with partial output may still be billed for confirmed usage. Unknown usage must not be treated as zero. Check records before resending; see API troubleshooting.