“OpenAI's O1 model was famous for introducing inference-time scaling, where a model generates hidden thoughts or intermediate steps before providing a final answer.”