Operating Paid Model Services
This page applies to operating deployments with external pricing, consumption records, and payment features enabled. Internal teams can start with Limits, Metering, and Costs without configuring top-ups or invoicing.
Distinguish service, consumption, and payment
| Object | Purpose |
|---|---|
| Model service | A callable model entry point |
| Price / SKU | The task's charging unit and rate |
| Consumption record | Charges resulting from a person or organization's service usage |
| Top-up record | Funds added to an account, distinct from consumption in that period |
| Invoice | The deployment's supported invoice request and processing workflow |
Provider cost and customer-facing price are accounted for separately. Dedicated resource charges, public model request charges, and top-ups also have different service rules.
Enable a paid service
- Connect the model and verify a request.
- Configure its SKU, metering unit, and price in Resource Pricing in the admin console.
- Check that customer-facing service information and pricing match.
- Make a controlled test call and reconcile usage with the correct personal or organization record.
- Explain available models, prices, credential setup, and charge-query entry points to customers.
See Asset Management to configure prices and Limits, Metering, and Costs to set spending allowances.
Reports and reconciliation
Billing Administration covers operating, consumption, top-up, voucher, and invoice information. Reconcile the following separately:
- Whether a request used a personal or organization key and reached the correct account.
- Units, applicable prices, and aligned time ranges.
- Reconcile account balance changes against top-ups and consumption records.
Compute voucher eligibility
Compute vouchers offset eligible compute-resource charges within the scope specified by the voucher. They cannot be used for token-based API calls. Token-call charges are deducted from the caller’s personal or organization account balance.
For billing questions, find consumption records by request time, model, and API key, then check request outcomes and retries against the bill. See Monitoring and Troubleshooting.