Skip to content

Billing

How pay as you go works

Your balance is held in US dollars as whole micro-dollars. Every movement is a line in a double-entry ledger that you can read under Billing.
  1. 1

    Estimate

    Before you send, the composer shows an estimate of the input tokens and the most that could be held. It is an estimate: the tokenizer that counts for real belongs to the model provider.

  2. 2

    Hold

    When you send, the maximum possible cost is moved from your available balance to a reserved balance in one database transaction. If the funds are not there, the request is refused before anything is sent. Parallel requests cannot hold the same money.

  3. 3

    Run

    The request goes to the provider with the model's price as a ceiling. In chat, the reply streams to you as it is generated.

  4. 4

    Settle

    The provider reports how many tokens were used. The charge is those tokens at the price frozen when you sent the message. Settlement happens exactly once, even if a connection is retried.

  5. 5

    Release

    The unused part of the hold returns to your available balance in the same transaction as the charge.

What happens when things go wrong

The request fails before the model starts
Nothing is charged. The full hold is released.
You press Stop, or your connection drops
The provider has already done work, so it is not free. You are charged for what was generated. Without a final count from the provider, billing stays pending and funds remain held until verified usage is available or an administrator releases the reservation.
You refresh or the page reconnects
The same message is never sent twice: each send carries a unique id that the server accepts once.
The server restarts mid-reply
The request is not treated as free and is not guessed at full price. It is marked as awaiting final usage, the hold stays in place, and it is resolved automatically shortly after, or by an administrator.
Chat history
Each request in a chat sends the earlier messages again so the model has context. Those tokens count as input, so long chats cost more per message.
Regenerate
Regenerating a reply is a new request and is billed again. The app tells you before it sends.

Adding funds

Deposits use USDG on Robinhood Chain from a wallet you have linked by signature. A deposit is credited after the required block confirmations and is checked again until the block is final. Where enabled, supported Stock Tokens can be deposited at a quoted value. See pricing and the asset and risk disclosures.