Skip to main content

GET /v1/credits

Returns the credit balance of the account that owns the API key, and the credits spent by the calling key. The call is free: it charges nothing and places no hold.

Response

A 200 is returned even when available is 0, so you can alert and pause before your next request fails with 402 insufficient_credits.
balance counts paid credits only. Free promo credits and subscription plan allowances are not included, so a request can still succeed on a plan or promo model while available is 0. A promo model is billed to promo credits only: it never spends balance or a plan allowance, and returns 402 once the account has no promo credits left for it.

Per-key spend

balance is shared by every key on the account. To reconcile one key, read key_spent. Pass since (Unix seconds, UTC) to count only requests started at or after that time:
A request counts once it finishes, against the time it started. Spend from a deleted key cannot be read, since its key no longer authenticates.

Avoiding cut-off streams

Every request holds its worst-case cost (prompt plus max_tokens) before it starts. When available is below that, the hold shrinks max_tokens to fit, and a stream can end early with an insufficient_credits error event. Check available before starting long generations, and top up or lower max_tokens when it runs low.

Errors

Errors use the standard error shape.