AI build receipts: tokens, cost and share of your limits

Why a receipt per build helps

A usage meter that only moves tells you that something was spent, not what. When an allowance drops faster than expected, the useful question is which request did it, and whether it was a long brief, an app rather than a page, or a run of retries. Itemised AI build receipts with tokens and cost answer that row by row, and because the receipt is written by the server when the build finishes, it reflects what the model really used rather than the estimate you saw beforehand in the build quote.

What each receipt records

Model: The model that ran the build, shown by name, for example Claude Opus 5.5.
Tokens: Input, output and cached input tokens, added up into one total.
Cost: What those tokens cost in US dollars. Very small amounts are shown to four decimal places in the table rather than rounded away to zero.
Share of today: The cost as a percentage of your plan's daily limit.
Share of the week: The same against the weekly limit.
Result: Built, Refused, Failed or Running, plus Debugging pass when a one-off rescue attempt rewrote a build.

Shares are worked out against your account's current plan limits, so if you change plan the older rows are recalculated against the new allowance.

Where you see them

The Console's Recent builds table lists your model builds newest first, with columns for when it ran, the result, tokens, cost, today's share and the week's share. Hover the time for the full date and hover the result for the model's name. Before your first build the table simply says there are no model builds yet.

The same data is available to tools: the account's builds can be listed newest first, up to 100 at a time, optionally for one project. A project's very first build runs before the project exists on your account, so the builder links that receipt to the project once it has synced, which keeps the project's own history complete.

Why some rows say Not charged

Only builds that finish successfully are charged. A build can end badly in two ways, and neither takes anything from your limits:

Refused. The model declined to build the request. The builder suggests describing the business differently.
Failed. Something went wrong on the way, such as an error from the model service or output that couldn't be used.
Failed debugging pass. A one-off rescue attempt that also didn't work.

These rows still show the tokens the model used, because that's what happened, but the cost column reads Not charged and both shares are 0%. Seeing them is useful in itself: a run of failures on the same wording is a sign to rephrase, and Anticode pauses builds briefly after several in a row so that retries don't pile up.

One more kind of row costs nothing. Copying a design from the templates gallery involves no model, so its receipt says it came from a template with no model use.

Reading receipts against your allowance

Look at the Today and Week columns rather than the dollar figures: your usage allowance is measured in shares, and those columns add up towards it.
Compare page builds with app builds. A model-written app writes a whole program, so its token count is usually far higher.
If one change in your own words cost more than expected, consider whether a quick change would have done: direction, theme, colour, sections and the name are changed in your browser for free.
When the allowance runs out before the day does, credit packs pay for what goes past it.

What receipts don't cover

They aren't invoices. A receipt records model cost against your allowance. Plan payments and credit pack purchases are handled by billing, separately.
Model builds only. Quick changes made in your browser, visual edits and code edits don't use the model and don't produce receipts.
Your own builds. The list shows builds made by your account. Plan prices and allowances are on /pricing.

Questions people ask

Why does a failed build show tokens but no cost?

The model may have read and written something before the build failed. The receipt shows those tokens for honesty, but only successful builds are charged, so it says Not charged and takes 0% of your limits.

Is the receipt the same as the quote?

No. The quote is an estimate made before the build. The receipt is written afterwards from what the build actually used.

Do builds started by an AI assistant get receipts?

Yes. Builds started through the MCP server are model builds on your account like any other and appear in the same list.

Does copying a template cost anything?

No. A template is copied into your project with no model call, and its receipt says it came from a template with no model use.