OpenAI has unveiled its first "thinking" model called o1

Unlike GPT-4o, the new model is able to build logical chains, analyze tasks sequentially and draw conclusions. This has significantly improved the accuracy and relevance of her answers. For example, it was among the top 500 students in the qualifying round of the U.S. Math Olympiad, and it also outperforms the average PhD in solving problems in physics, biology, and chemistry.

The model also performed well in areas such as data analysis, text editing, writing, and programming. In these areas, the o1 was 40% more efficient than the GPT-4o.

The ability to reason helped to teach the model moral principles and values. It is now easier to integrate these aspects into her work so that she always takes them into account when forming her responses. An additional benefit was the ability to "read the mind" of the model, tracking how it arrived at its conclusions.