Reasoning effort and advanced settings

Some models can think before they answer. Thinking often gives better answers to hard questions, but thinking is written text too, so it costs money.

Setting the effort

Open Advanced next to the message box. If the model you picked supports it, you see Reasoning effort with some of these levels: Off, Low, Medium and High.

• Each level shows a typical cost and an “up to” cost per reply, so you can see the difference before you send.
• The estimates start broad and firm up as the model is used (“Estimates firm up as this model is used”).
• Your choice is remembered for this chat. If you switch to a model that doesn’t offer the same level, its default is used.
• Some models always think; for them, effort sets how deeply (“{model} always thinks — effort sets how deeply.”).
• Claude models where thinking is optional start on Low to keep costs down. Raise it for harder problems.

Effort changes the price, never the model: the reply is still served by the model you picked.

Watching the model think

While a thinking model works, a Thinking section opens above the answer. When it’s done, it folds into “Thought for a moment” — tap to read it again. Thinking isn’t included when you copy or export a reply.

Creativity

The Creativity slider runs from Precise to Creative. Lower sticks to the likeliest wording; higher varies it. It never changes the price. Some models set their own creativity, and the slider says so.

Limit response length

Turn on Limit response length to cap how long the reply can be. A cap can only make a reply cheaper, never dearer. On thinking models at Medium or High effort, thinking counts towards the cap, so a low cap may cut the answer short.

Reset puts every advanced setting back to its default. A dot on Advanced means something is changed.