On the early morning of the 18th, OpenAI started the 9th day of technical sharing live broadcast, officially released the API of the o1 model, and made a major upgrade to the real-time API to support WebRTC. Among them, the API of the o1 model has reduced the cost of thinking by 60% compared with the previous preview version, and comes with advanced visual functions; the audio cost of GPT-4o is reduced by 60%, and the price of the mini version has dropped by 10 times. At the same time, OpenAI also released a new preference fine-tuning method, which allows the large model to better grasp the user's preference style through the direct preference optimization algorithm.