That doesn’t mean “use fewer thinking tokens”. It might use more as it mulls over how to make its response concise.