How billing works

Metered per request, to the token. No subscription, no minimum, nothing charged for an idle model.

What a request costs

Input, cached input and output tokens are priced separately. Cached input is much cheaper, and reasoning tokens count as output unless a model prices them separately.

cost = input_tokens        × input_price
     + cached_input_tokens × cached_price     # 便宜得多
     + output_tokens       × output_price

# 再乘以你所在等级的折扣倍率

Pre-authorization

When a request starts we place a temporary hold on your balance sized to the worst case, then settle against real usage when it finishes. If a request fails or you disconnect, the hold is released and you are charged nothing.

This is why a large max_tokens on a nearly empty balance can be rejected: the hold, not the actual reply, is what has to fit.

Reconciling

Every call produces one usage row and one ledger entry. Your balance is the sum of the ledger, so any charge can be traced back to the request that caused it.

Volume discounts

Tiers are based on cumulative top-ups and apply automatically to every model. They are permanent — we never downgrade you.