Home

Switch language

Start Install GuideQuick StartDocs Overview
Docs ChannelsModels & APIGateway OpsTools & Skills
More ArticlesResourcesHelp Center
Get Started
Documentation navigation Tools & Skills
Tools & Skills 2 sections

Thinking Controls

Thinking Levels

/think, /verbose and /reasoning do not simply swap models. They control how much reasoning budget the model spends in the current session, whether tool details are echoed back and whether reasoning content is visible.

/think Levels

Supported values include off | minimal | low | medium | high | xhigh, with common aliases such as /t, /think:high and /thinking medium.

An inline instruction only affects the current message. Sending a message that only contains the command stores it as the new session default.

If the underlying model does not support fine-grained reasoning levels, OpenClaw degrades to the closest range the provider accepts.