Usage and pricing
Understand what each request consumes and how its cost is calculated.
Models may use different prices and metering methods. Before calling a model, review its billable items and units. After the call, use Requests to review the actual usage and cost of that request.
Common pricing units
| Unit | Description |
|---|---|
| USD per one million input tokens | Calculated from input token usage |
| USD per one million output tokens | Calculated from output token usage |
| Cache-related units | Available only for models that explicitly meter cache usage |
| Image or other units | Calculated by image count, dimensions, quality, or another model-specific unit |
Do not assume that every model has cache pricing, and do not convert image prices directly into text token prices.
Estimate text request cost
When a model bills only ordinary input and output tokens, use:
Input cost = input tokens ÷ 1,000,000 × input price
Output cost = output tokens ÷ 1,000,000 × output price
Estimated cost = input cost + output costDisplayed precision may differ from settlement precision. If cache, media, or other billable items apply, review those details as well.
Balance and top-up
- Open billing or top-up in the console.
- Review the amount, currency, credited balance, and any fees shown on the page.
- Complete payment with a method offered on the checkout page.
- Return to the console and review the order status and balance change.
Do not create another order while a payment is processing. Wallet balance does not allow every API key to call every model; key budget, expiration, and model access still apply.
When payment status is unclear
| Status | Recommended action |
|---|---|
| Processing | Keep the order number and wait for an update; do not pay again |
| Failed | Follow the checkout message to review verification or payment method details |
| Payment provider shows success but balance is unchanged | Confirm the account, order number, amount, and time before contacting support |
| Refund processing | Distinguish request, review, refund initiation, and bank settlement states |
Never provide a complete card number, password, verification code, or API key to support.
Charges for failed requests
A client timeout does not prove that the server did not process a request. Check Requests to see whether the request was accepted, generated usage, or reached settlement. Use the request record and applicable rules for failed, cancelled, and retried request charges.