Reasoning Effort

Some models think a question through before they answer. Reasoning effort sets how much thinking they may do: more effort usually gives better answers to hard questions, such as analyses, calculations or code, but the answer takes longer and uses more tokens.

Choosing the effort

1. Select a model that supports reasoning effort. The Reasoning effort selector appears in the message box only when every selected model supports it, so it disappears again when you add a model that does not.
2. Open it and pick Low, Medium or High.
3. Send your message. The effort applies to the messages you send until you change it.

Depending on the model, the list also offers a first option that leaves the effort to the model: Default, Auto, or Disabled for models that then answer without thinking first.

When to raise it

Start with the model’s default. Raise the effort when an answer to a demanding question misses steps or gets details wrong, and lower it for quick questions where speed matters more.

Reasoning effort does not apply to Deep Dive, which decides how long it researches by itself.