Credits & usage
Understand package balances, model metering, and per-call costs.
On this page
Package creditsFree-model subscriptionsHow usage is meteredReservations & insufficient creditsInspect usagePackage credits
Paid requests draw from active package credits. Open Credits to view packages, remaining balances, and payment options. Packages can have expiry dates and account-specific conditions. The displayed console balance and current package details are the source of truth for your account.
Free-model subscriptions
Subscriptions provide fixed-duration access to a free-model rate-limit group. Choose a plan and a supported payment method in Subscriptions. A confirmed purchase verifies your account and activates access; buying the same plan again extends its expiry. Other plans are available after your current subscription expires. There is no automatic renewal. Subscriptions do not add credit tokens or change RPM limits. Daily usage carries over on purchase and renewal, and group rules may change. After expiry, your existing rate-limit assignment resumes.
How usage is metered
| Request type | Credit basis |
|---|---|
| Paid text generation | Input plus output tokens multiplied by the configured pool credit multiplier, rounded up |
| Image generation | Generated image count multiplied by credits per image for the resolution tier |
| Tools API | Configured price for each successful operation |
| Free models | Zero credits within applicable free policies; paid overage can apply when configured |
Token counts and credits are different measures. A credit multiplier can make billed credits differ from raw tokens. Current model multipliers, image prices, and tool operation prices are shown in the console.
Reservations & insufficient credits
The service reserves credits before paid execution to prevent overspending. Text reservations depend on estimated input and the requested output limit. A large max_tokens value can require more available credits than the eventual generation consumes. Reduce the limit or add credits if admission returns 402. Successful requests settle against recorded usage; failed execution releases reservations.
Inspect usage
Use Usage to review model requests, token counts, status, and consumption. Use Tools Studio for tool-call history, latency, success, and charged credits. Keep the response request ID when reporting a discrepancy. Dashboard totals may reflect usage after buffered accounting is processed.