Evaluating open ai tools o1 and o3-mini in solving high school problems from Serbian national mathematics competition

Abstract

Artificial intelligence (AI) tools are already being widely used in mathematics education in various ways and for different purposes. One important direction of their application lies in using AI tools for solving complex, competition level mathematical problems. In this study, we contribute to this field by analyzing and evaluating the solutions to problems that high school students solved at the National Mathematics Competition in the Republic of Serbia, which were generated by the OpenAI tools o1 and o3-mini. The results indicate that not all solutions produced by these two tools are accurate and fully correct. However, based on the evaluated solutions, it can be concluded that, in most cases, the performance of these two tools would qualify them for one of the top three prizes in competition with the students. Experienced students, mathematics competitors who qualified for the national competition– could benefit from reviewing the solutions to problems they were unable to solve independently, provided by o1 and o3-mini, as long as they approach these solutions critically and carefully analyze each aspect of the reasoning process.