OpenAI GPT-5.4 Pro: Breakthrough on an Erdős Problem

OpenAI GPT-5.4 Pro: Mathematical Breakthrough with the Resolution of an Erdős Problem

OpenAI's GPT-5.4 Pro model has recently made headlines by solving an open mathematical problem, Erdős problem number 1196. This challenge, which had stumped mathematicians for years, was overcome by the model in about 80 minutes. The result was then formalized in a LaTeX document in an additional 30 minutes. Currently, the solution is undergoing formal verification.

The renowned mathematician Terence Tao reacted to this advancement on the forum dedicated to Erdős problems. He emphasized that the model's work reveals an unprecedented connection between the anatomy of integers and Markov process theory. According to Tao, this discovery far exceeds the mere resolution of the Erdős problem, making a significant contribution to the understanding of the anatomy of integers.

Kevin Barreto, who recently announced his upcoming integration into OpenAI's AI for Science team, also commented on this achievement. He noted that the model's use of the Markov chain technique represents a creative step that human mathematicians had overlooked, despite years of research on this problem.

This discussion is part of a broader debate about the ability of language models to generate new knowledge in mathematics and other disciplines. The example of GPT-5.4 Pro demonstrates that novel discoveries can emerge from already known data, thereby expanding the potential of artificial intelligences in the scientific field.