OpenAI has released two new AI models designed to improve the reasoning capabilities of artificial intelligence-powered systems. These models indicate a step up in sophistication compared to the company’s previous AI models, such as GPT-4o.
The Microsoft-backed company explained in its blog post that it will be launching two new AI models, o1 and o1-mini. The o1 will be available in ChatGPT and its API beginning on Thursday, and the o1-mini will come to ChatGPT Free users. However, it will be a much smaller model, meaning it will be less powerful and, therefore, more affordable than the o1 model. o1-mini is expected to go to ChatGPT Free users at an unknown date.
OpenAI explained in its blog post the o1 model scored 83% on the qualifying exam for the International Mathematics Olympiad, which is a stark jump compared to the 13% scored by its previous model, GPT-4o. This new-found “reasoning” capability was made possible by implementing a technique called “chain of thought,” which involves the AI being taught to break down a user’s prompt into more logical steps that are then carried out individually rather than addressing the prompt in its entirety.
This process makes the AI’s response much slower but produces better and more well-thought-out results. An example of this is how OpenAI demonstrated that o1 is capable of creating a basic video game from a prompt. The AI tool was shown coding a video game from a prompt, which the user then simply copied and pasted into a window to play.