“Implementing test-time compute, or reasoning, makes AI model inference approximately three to ten times more expensive.”