“OpenAI's O1 model, released in late 2024, was the first to exhibit the "reasoning trick" where models articulate their thought process.”