According to PANews, OpenAI has announced the official release of its o1 model API during the ninth day of its technology sharing live stream. This release includes significant upgrades to the real-time API, incorporating WebRTC support and reducing costs by 60%. The o1 model is equipped with advanced reasoning capabilities, supporting function calls, structured output, and visual features. Compared to its preview version, the thinking cost has been reduced by 60%.
In addition to these enhancements, the cost of GPT-4o audio has decreased by 60%, and the price of the mini version has been reduced to one-tenth of its original cost. OpenAI has also introduced a new preference fine-tuning method, utilizing a direct preference optimization algorithm to better align the model with user and developer preferences. Furthermore, developers can now use newly added Go and Java SDKs, making it easier to integrate OpenAI's functionalities into their applications.