Understand credits and limits

Use Pricing for current published official model rates and your account for its available balance. Check both in the organization that owns the API key before sending a paid request. Pricing does not quote the rate of a possible Jailbreak execution; its final debit can differ from the displayed official rate. Do not estimate cost from a fixed example in documentation: available models, rates and entitlements can change through the catalog and plan. The free model catalog checks visibility without starting generation.

A logical request is charged once at the pinned rate of the access class actually used. Purchased top-up credits can pay for official access only. Jailbreak, when your subscription allows it, uses monthly subscription credits. The model catalog can list a selectable <canonical-model-id>-jailbreak selector; it enables the Jailbreak capability without creating a separate canonical model. The response, history and charge still use the canonical model ID. A paid retry with the same idempotency key and parameters replays the same terminal result within its retention window instead of charging twice.

An allowance visible for one access class does not guarantee another class has enough credit. Model availability, plan entitlement and credit eligibility are separate checks. A displayed rate is pricing context for the selected model, while the final debit depends on the completed logical request. A request that is still running must not be counted as a final charge.

Before a paid call

1. Confirm the selected organization, plan and balance. A balance in another organization does not authorize this key.
2. Read the published official model rate and plan in Pricing, then select an available canonical model ID. If Jailbreak is enabled, do not treat the official rate as a guaranteed final debit.
3. Start with a bounded max_output_tokens value and one intentional request. Save the public request ID and your Idempotency-Key if retries are possible. Reuse that key only with the same normalized parameters within its 24-hour window.
4. Check API Activity for the outcome and the finalized debit. Until the reservation closes, the row does not call the hold a charge. The balance and published rates add context; they do not replace that row's debit.

Recover from a limit

HTTP 402 insufficient_credits means an eligible credit source cannot cover the request. Purchased credits do not pay for a Jailbreak execution; check the subscription's available monthly credits if that capability is involved. HTTP 429 usage_limit_exceeded can reflect a subscription window or period. Read the public retry/reset guidance and x-should-retry; an immediate loop cannot create new allowance. A separate transient rate_limit_exceeded may be retried after its advised delay with the same logical request key. See errors and limits for the decision path.

Before topping up or changing plan, confirm whether the problem is credit eligibility, a subscription allowance or a temporary request rate. A top-up does not reset an exhausted subscription window, but eligible purchased credits can let official requests continue beyond that window; they cannot pay for Jailbreak. Retrying a 402 without changing available eligible credit will not help. For a temporary 429, respect Retry-After; for x-should-retry: false, change the underlying condition before trying again. Check the current Pricing page for available account actions rather than assuming a payment method from an API error.