---
title: How billing works
description: Else is prepaid and metered. You top up a Wallet, and each run draws from it at a published price — no subscription, no credits, no seats.
---

Else is **prepaid and metered**. You put money in a Wallet, and each piece of work
draws from it at a published price. There is no subscription, no monthly plan, no
per-seat charge, and no credits or tokens to buy and interpret. Your balance is
ordinary money in US dollars, and **it never expires**.

Nothing is charged for having an account, opening the app, or leaving a run paused
and waiting for you. Money moves when work actually runs.

## The five things that can cost money

Everything on a bill is one of these. Each is metered by real use, not by a flat
per-run fee:

| What it is | How it's metered | Where the rate is |
| --- | --- | --- |
| **Thinking and writing** — the model reading your request and producing an answer | Per million tokens, priced separately for input and output | [Model prices](/docs/model-prices) |
| **Generating an image** | Per image, by size | [Model prices](/docs/model-prices) |
| **Reading a web page** | Per page rendered | [Model prices](/docs/model-prices) |
| **Running code in the secure workspace** | Per minute of run time | [Model prices](/docs/model-prices) |
| **Transcribing audio you send** | Per minute of audio | On the receipt |

The first one dominates almost every bill. Tokens are how models measure text —
roughly 4 characters, or about ¾ of a word. Output tokens cost more than input
tokens on nearly every model, which is why a long answer costs more than a long
question.

The last three are metered per unit of real use. Two of them carry a published rate on
[Model prices](/docs/model-prices); audio transcription is charged at the metered rate we
are charged for it with nothing added, so its amount appears on the receipt rather than
as a rate here. Either way the charge is itemized when it happens, which is the number
that actually matters — see [Reading a receipt](/docs/reading-a-receipt).

## Price depends on which model ran

There is no single "price per task", and we are not going to invent one. What a run
costs depends on which model did the work, and models differ by more than a hundredfold
from cheapest to most expensive.

You pick that through the engine menu in the composer, which offers four options:

- **Auto** — picks for each request
- **Fast** — cheapest acceptable answer
- **Smart** — balanced quality, long context
- **Max** — deepest reasoning

Each of the three named tiers runs one model by default, and
[Model prices](/docs/model-prices) lists which model that is together with its per-million-token
price. You can also point a tier at a different model in Settings → *How it works* →
**Set an engine per speed tier**, or pick a model for a single request from **Pick an
exact engine**. Whatever you pick is recorded on the receipt.

Because the tiers are named for what they are *for* rather than what they cost, a tier
name on its own is never a price. Always read the two together.

## One more thing that can appear

A very long chat eventually exceeds what a model can hold at once. When that happens
Else summarises the earlier part of the chat so the work can continue — and that
summarising step is itself a model call, so it is metered like any other. It is
unusual, it only affects long-running chats, and it shows up on the receipt like
everything else. We mention it because it is the one charge you would not otherwise
predict.

## What you are charged when things go wrong

Briefly, because it deserves its own page: **you are charged for work that actually
ran, including when the run failed and produced nothing useful, and there is no
refund.** If a run consumed nothing at all, there is no charge and no receipt. See
[When a run fails](/docs/when-a-run-fails) for the full picture.

## Your spend limit is not a price

The number you set on a run, and the Wallet's top-up amounts, are budgets you choose
rather than prices. A given limit buys very different amounts of work depending on
which model runs.

What that number *does* depends on where you set it. **On a chat it is not applied
today** — it is recorded and shown back to you, and nothing acts on it; what stops a
chat is a fixed per-turn ceiling you cannot see or change. On work you hand off the
limit you set is genuinely read, but only between steps and only against money already
spent. [Spending limits](/docs/spending-limits) is the page on that, and it does not soften it.

## Where to look next

- [Model prices](/docs/model-prices) — the per-model rate for every model you can pick, and what each tier runs
- [Reading a receipt](/docs/reading-a-receipt) — every field on a receipt, and what it means
- [Topping up](/docs/topping-up) — how to add money, and what happens the first time
- [Spending limits](/docs/spending-limits) — what the limit does, and what it does not
- [Glossary](/docs/glossary) — what *run*, *turn*, *step*, *receipt* and *Wallet* mean exactly; this page uses all five