# Cost: price steps and cheaper hours

> If a model charges more past a certain conversation size, or less at certain hours, the app knows and acts by itself so the same work costs less.

Models and accounts · Updated: 2026-11-02 · version 2.0.117 · https://nobreiniers.com/en/help/price-steps/

Some models charge several times more for every request once the conversation passes a certain size. Others cost less at certain hours. NoBreiniers reads this from each model's price and acts by itself. Both are on from the start and are turned off in «Settings» › «Accounts» › «Cost».

## The price step

A step is a conversation size past which every request to that model costs more: not only what goes over, the whole request.

Before reaching a step, the app does this, in this order:

1. **It trims what does not hurt.** Old tool results become short previews. The full text stays saved in the chat.
2. **It compacts before the step.** The conversation is compacted a little before getting there, with a margin. See [Context, compaction and the task record](https://nobreiniers.com/en/help/context-and-compaction/).
3. **It watches what comes in at once.** A result so large that it alone would cross the step enters as a preview, with all its text in a file of the chat. A message of yours that long reaches the model as its start and its end, and whole in the chat's files. In the chat your message stays complete.
4. **It does the sum.** It stays under only when compacting costs less than paying the higher price. Otherwise it uses the whole window and tells you why.

A model with no steps, a plan that does not bill by the token, and a model running on your Mac stay exactly as they were.

## What you see

- On a model with a step, «Conversation shortened · …» shows more often in the chat. Nothing else changes.
- The balance's card, in the title bar, says so: «The conversation is kept short so each answer costs less».
- The «Context» window has one line that explains it, only on models that have a step:

| The line ends like this | What it means |
|---|---|
| «…so this chat is kept under that: it compacts by itself at …» | The usual case: the chat stays under the step. |
| «This chat compacts before it gets there.» | The point where the chat already compacts is below the step. |
| «Staying under that is turned off in Settings › Accounts, so this chat uses the whole window.» | You turned it off. |
| «This chat's instructions and tools leave too little room under that, so it uses the window instead of compacting every few steps.» | The step is too low to work well under it. |
| «Compacting to stay under that would cost more than it saves, so this chat uses the window.» | The step raises the price so little that it is not worth it. |

They all begin the same way: «Past … tokens every request to this model costs … times more».

## Does quality change?

In our tests, no. Six long coding chats (100,000 to 267,000 tokens of context, one model) were run three ways: compacting at the usual point, and staying under a step at 100,000 and at 200,000 tokens. Every task passed every way: 34 of 34.

What does change is how often. Under a step at 100,000 tokens, those chats compacted about every four minutes of continuous work. Under one at 200,000, once in a 72-step chat. A chat that does not reach the limit is not compacted.

> **Note:** At the prices of a model that has a step at 100,000 tokens, those same chats cost 65% less when kept under it. This is a calculation with the price table over recorded chats, not an invoice.

## Turn it off

1. Open «Settings» (⌘,) › «Accounts».
2. Open «Advanced», then in «Cost» turn off «Keep long chats from costing more».

Off, the chat uses the model's whole window.

## Cheaper hours

In «Advanced» › «Cost», «Cheaper hours for work that can wait» makes the work nobody is waiting for run when the model costs less:

- Automatic tasks with no fixed time that repeat every six hours or more wait, and so does what the app learns in the background. Four hours at most.
- Nothing you ask for waits: that always runs at once. Neither does a task with a fixed time, nor one you start with «Run now».

An automatic task that waits says so on its row: «Waits for cheaper hours · runs at …». To run it now, press its play button.
