# Reasoning effort

Tell a model to think longer for hard problems, or less for quick ones. Where the effort control is, what the levels mean, and when to change it.

Canonical page: https://themachine.myapps.ai/docs/desktop/models/reasoning-effort

Some models can **think longer** before they answer. This is called
**reasoning effort**.

| | Higher effort | Lower effort |
|---|---|---|
| **Answers** | Often better on hard problems | Fine for simple tasks |
| **Speed** | Slower | Faster |
| **Cost** | More | Less |

## Where to find it

When the current model supports it, a button with **four bars** appears in
the message box. Hover it to see the current level.

If you don't see it, the current model doesn't support effort, or you can
view this bot but not change its settings.

## Change the effort

![The reasoning effort slider](/images/docs/desktop/effort.webp)

1. Click the **four bars**.
2. Drag the slider, or use the arrow keys, to pick a level.
3. Click outside to close.

To go back to normal, click **Reset to the model's default**.

At the highest level, the bars animate in color.

### The levels

Each model has its own set of levels. You might see some of these, from
lowest to highest:

**None**, **Minimal**, **Low**, **Medium**, **High**, **Extra high**, **Max**

Some models only have **Low**, **Medium**, and **High**. **Model default**
means you haven't set a level, and the model decides.

## What it applies to

The effort applies to **every thread** of that bot, for that model, starting
with the next reply. It isn't per thread. It's shared with the Machine web
app, so changing it here changes it there too.

Each model keeps its own level. If Scout uses one model in one thread and a
different model in another, each can have a different effort.

The four bars in the message box are the only place to set it in Machine
Desktop. **Bot settings → Model** sets the default model, not the effort.

## When to change it

**Raise it when:**

- the task needs careful reasoning, like planning, math, or tricky logic,
- the bot is writing or fixing complex code,
- the bot keeps missing something.

**Lower it when:**

- the task is simple and you want speed,
- you're doing lots of small requests and want to save money.

**Leave it alone when:** you're not sure. The model's default is a sensible
middle.