Thinking Effort

A Profile stores the default reasoning settings for its model. While a task is open, you can override them for that task alone from the input toolbar. The Profile itself does not change, and other tasks keep their own settings.

Where the control appears

When a task is open and the selected model supports thinking, a thinking selector appears in the input toolbar next to the model name. It shows the current value, for example Medium or Budget; hover it to see Thinking: <value>.

The options come from the model metadata:

Option	Meaning
Effort levels (for example Low, Medium, High)	Ask the model for a fixed reasoning effort. Only the levels the model supports are listed
Budget	Give the model a thinking token budget. A number field appears; type a value between 0 and the model’s maximum and press `Enter` or leave the field

If the model does not support thinking, or the Profile turns thinking off, the selector is hidden.

How the override behaves

Task only. The value is saved in the task’s own settings. It never becomes a global default and does not modify the Profile.
Per mode. Plan mode and Act mode keep separate overrides, because they can use different Profiles.
Validated against the model. An effort level the model does not list, a negative budget, or a budget above the model’s maximum is rejected, and the error appears next to the control.
Inherited default. Until you pick a value, the task uses the Profile’s reasoning settings.
Subagents are separate. Subagents use their own Profile settings; the main task’s thinking override does not apply to them. See Subagents.

OpenAI service tier

For OpenAI and OpenAI Codex Profiles that enable service tiers, an icon-only Task service tier button appears beside the thinking selector. Choose Auto, Default, Flex, Scale, Priority, or Ultrafast for the current task. Like the thinking override, the choice is saved per task and per mode.