AI solves open math problem independently with ChatGPT

Artificial intelligence has taken a step that until recently seemed like science fiction: the latest version of ChatGPT independently solved one of the complex mathematical problems classified as "open." This was highlighted by researcher and developer Neil Somani, who presented the model with a non-trivial problem and left it to work without human intervention. After approximately 15 minutes, the system produced a detailed solution, which Somani then verified and formalized using the mathematical tool Harmonic—confirming that the proof was correct.

As reported by BODA.SU, Somani's goal was to understand where the boundary of capabilities lies for modern large language models in mathematics. The experimental results turned out to be surprisingly strong: the latest ChatGPT version confidently handled complex concepts and demonstrated a level of reasoning previously considered inaccessible to AI. In the researcher's view, this significantly shifts our understanding of what tasks machines can solve on their own.

In its reasoning process, the model didn't just find an answer but sequentially applied fundamental mathematical ideas and formulas, drawing on works discussed in the professional community over a decade ago. The final solution wasn't a copy of known approaches but an independent and more complete derivation for a problem variant originally formulated by Paul Erdős. This level of generalization and adaptation was once characteristic exclusively of humans.

Somani notes that a noticeable leap occurred after the release of version GPT 5.2: the number of problems AI handles without direct assistance has sharply increased. While early autonomous successes in this field were recorded earlier, the progress has now become hard to ignore. Experts emphasize that AI isn't replacing mathematicians but is increasingly becoming a powerful tool capable of accelerating solution discovery and expanding the boundaries of human knowledge.