Skip to main content
Some models reason through a problem before they write an answer. The Thinking control lets you choose how much effort a model puts into that reasoning, one message at a time. Use more for a hard analysis or a tricky draft, and less when you want a quick answer.

​
Choose a level

1

Pick a model that supports it

In the chat toolbar, choose a named model in the model selector. If the model supports a thinking level, Thinking appears beside the model picker.
2

Choose a level

Pick Default, Low, Medium or High, then send your message.
The level applies to the message you send. The conversation remembers it, so when you come back, Thinking is set to the level of the latest reply.

​
When the control is hidden

  • The model doesn’t support it. ParléHub shows Thinking only for models confirmed to accept a thinking level. More models are added as they are tested.
  • You chose Auto. With Auto, ParléHub picks the model for you, so it also leaves thinking at the default.

​
See which level wrote a reply

When a level other than Default is used, the reply’s label shows it next to the model name, for example GPT-6 Astra · High thinking. Every project member sees the same label, both live and after a reload.

​
Cost and limits

A higher level doesn’t raise the reply’s token allowance. Thinking uses part of that allowance, so a higher level can mean longer waits and more output tokens for the same question, but your budgets stay in control. Providers that report thinking separately show it in the Reasoning Tokens column of your usage reports.
Start at Default. Move to High for the turns that need careful reasoning, such as reviewing a contract clause or reconciling figures, then drop back.