Both input and output tokens count toward usage.
Input includes the serialized conversation and supported tool or response-format definitions sent with a request. Output is the completion generated by inference. BrokenGPT records provider-reported usage when available and otherwise uses the application metering path described in the live pricing page.
Included tokens and prepaid credits are separate balances.
When included tokens are available, usage is accounted against the plan allowance. Prepaid credit covers eligible usage outside that allowance according to the current pricing contract. Always use the live pricing page for the current amounts and rates.
| Balance | How it is obtained | Expiry behavior |
|---|---|---|
| Free allowance | Account plan | Resets under the current plan terms |
| Pro included tokens | Paid subscription period | Unused included tokens do not roll over |
| Prepaid credits | Credit purchase | Current pricing page states they do not expire |
Use scopes and limits to reduce surprise spend.
- Create separate keys for production, staging, and individual services.
- Grant only the scopes each key requires.
- Configure key-level token or spend limits where the dashboard offers them.
- Set workspace budgets and alerts before scaling traffic.
- Review chat and API usage attribution independently.
A quota 429 may need action rather than a retry.
Rate limits, key limits, workspace budgets, included allowances, and prepaid balances are distinct controls. Read the error code before deciding whether to wait, rotate traffic, change a limit, or add credit.
- A request-rate limit may clear after the reset time.
- A hard workspace budget requires an authorized configuration change.
- A per-key token or spend limit requires key administration.
- Insufficient included tokens and credits requires an eligible balance.