Reading a receipt

Every run that costs money produces a Receipt, and the receipt is the record — not the limit you set, not an estimate made beforehand. It shows what was charged, how long the work took, and which engine choice you submitted.

A run that consumed nothing produces no receipt at all, rather than a receipt reading $0.00. If there is no receipt, there was no charge.

Where receipts appear

Under each reply in a chat. A small chip sits beneath the answer showing three things: the amount charged, how long the turn took, and your engine choice. A real one reads:

$0.502 · 1m 46s · Auto

Hover it and it expands to the detail — the token counts, and a line per billed attempt naming the model that produced it.

On a handed-off task’s result page. A card headed RECEIPT shows the total for the task, the spend limit and how much of it has gone, and the line items that make up the total, with a running TOTAL SO FAR.

Those are the two places. The Wallet shows your balance, not a per-run breakdown — for what an individual run cost, go to that run’s receipt.

Every field, and what it means

Field	What it means
Amount charged	The authoritative figure — what actually came out of your balance for this turn. Everything else on the receipt explains it; nothing else overrides it.
Tokens	The total text measured for this turn, input and output together.
Input tokens	Your request, plus any material and earlier chat the model had to read. This is usually the larger number and the cheaper one.
Output tokens	What the model wrote back. Priced higher than input on nearly every model.
Duration	Wall-clock time from the start of the turn to its completion. It is shown for context; you are not billed by the second for thinking time.
Model	Which model produced the reply. On the chip this appears in the expanded detail rather than the visible line.
Your engine choice	What you submitted — Auto, a tier, or an exact model — captured before anything was decided. This is your instruction, not the outcome, which is why it can differ from Model.
Time	When the turn completed.
Why it stopped	Present only when a turn was cut short: because it reached its spend limit, its time limit, or a length limit. Absent on a turn that simply finished.
Attempts	One line per billed attempt, in the order they were billed, each with its own model and price. More than one line means an attempt was retried — see below.

Why there can be more than one attempt

If an attempt is paid for upstream, it appears as its own line, even if it did not produce the final answer. So a turn that needed a retry shows two lines, and the total is the sum. This is the honest version: itemizing a paid attempt you did not benefit from is better than hiding it inside a single rounded number.

Reading the task card’s spend limit line

On a task’s result page the receipt sits next to a line like Spend limit $1.37 — 88% used. That figure is the budget you set, not a price and not a quote. What the same budget buys depends entirely on which model runs — see Model prices.

And note what the percentage does not promise: the limit is checked between steps and compares money already spent, so a run can pass 100% before it pauses. Spending limits explains exactly how far that can go.

Two limits of the current receipt, stated plainly

The task card shows one aggregate line, not a per-model breakdown. When a task ran many steps across more than one model, the card gives you the total and the status rather than a row per model. The per-attempt detail in a chat’s receipt chip is finer grained than the task card’s.

The amount on a task card is not shown next to the model tier that produced it. You have to look at the run itself to see which engine choice was in force. We would rather name that gap than let you assume the number carries a tier with it.

What the receipt is for

Read the receipt rather than reasoning backwards from your balance. It is the only place that ties an amount to a specific piece of work, a specific model, and a specific moment — which is what makes a charge checkable rather than something you have to accept.

Where to look next

• How billing works — the five things that can cost money
• Model prices — the rate behind every amount you see
• When a run fails — what you are charged when nothing useful came back