OpenAI Claims GPT-5 Solves Century-Old Math Problem

OpenAI announced on Tuesday that its AI system has produced a proof of the conjecture known as the “XYZ problem,” a long‑standing puzzle in number theory that has stumped mathematicians for over a century. The company released a short statement on its website, claiming the proof was generated by its latest language model, GPT‑5, and that the model had independently derived the result using a novel algorithmic approach. The announcement came during a week of heightened interest in AI‑assisted research, following DeepMind’s success with AlphaFold and a recent paper from researchers at MIT showing that neural networks could find new patterns in algebraic geometry.

The XYZ problem, formally stated in 1953, has been a touchstone for the field because it links prime numbers to complex analysis. Solving it would close a chapter on a central question about the distribution of primes and could unlock new cryptographic techniques. The timing of OpenAI’s claim is no accident: the company has been investing heavily in mathematics research, and the release of GPT‑5 coincides with a broader push to demonstrate AI’s capacity for “high‑level reasoning.” Experts note that AI systems are now capable of manipulating symbolic expressions and exploring vast search spaces far beyond human reach, making them promising tools for tackling deep conjectures. However, the mathematics community has long been cautious, insisting that any claimed proof must undergo rigorous peer review before being accepted.

Mathematicians have reacted with a mix of intrigue and skepticism. Dr. Elena Kovács of the University of Cambridge said the proof “contains many novel steps, but the logic is not yet transparent enough for us to confirm its validity.” Others point out that the proof’s source code was not made public, raising questions about reproducibility. A group of researchers on Twitter has begun to trace the proof’s lineage, finding similarities to a preprint published last year by a small team of independent researchers. OpenAI has denied any plagiarism, stating that the model was trained on publicly available data and that the proof was generated in real time during the experiment. The company has offered to release the model’s internal logs to a panel of independent reviewers, but it has not yet provided the full proof for external scrutiny.

The next step will be a formal vetting process. If independent mathematicians can verify the proof, it could be submitted to a leading journal such as the Annals of Mathematics. Even if the proof is flawed, the episode will likely accelerate discussions about AI’s role in scientific discovery and the standards required for machine‑generated claims. The controversy also highlights the tension between open‑source AI research and proprietary development, as OpenAI’s approach to sharing its findings remains cautious. Whether the XYZ problem is finally solved will hinge on the community’s willingness to engage with the evidence and on the transparency of the tools that produced it.