OpenAI's New API Gets Powerful o1 Model

Developers leveraging OpenAI’s API will now gain access to the company’s latest supermodel, known as o1. This upgrade is part of a series of new features introduced for developer customers today. Instead of the older o1-preview, the new o1 model brings back essential features that missed the previous version. One key feature is the ability to guide chatbots with specific directions, such as “You are a helpful assistant for tax professionals. ” Besides that, developers can control how long the API thinks about certain queries, which saves both time and resources. Interestingly, even visual information like scanned documents can be inputted into the system. The new API also excels in internal function calling, where the model uses pre-written functions from developers to answer queries as necessary. Additionally, it improves structured outputs that format information specifically per the developer’s guidelines. Notably, the computational power of o1 is remarkably efficient, using 60% fewer tokens than o1-preview, making it faster and cheaper. However, the new model’s accuracy soars by 25 to 35 percentage points on benchmarks like LiveBench, according to OpenAI. Developers who are part of tier 5 will start seeing these changes today. But for those interested in the $200/month o1 Pro model, they’ll have to wait for more updates soon.