GPT-5.6 Pro Reportedly Refutes 30-Year Graph Theory Conjecture

Rumors claim that a user leveraged GPT-5.6 Pro to successfully find a counterexample to the Dinitz-Garg-Goemans conjecture in graph theory, overturning a mathematical problem that had been open for about 30 years. The news sparked heated discussion in the AI community due to the dramatic contrast of the user going to sleep after prompting the model and waking up to find the problem solved.

Confirmed

Based on the provided posts, the rumored counterexample includes specific mathematical parameters: the fractional flow cost of the graph is 58, while any unsplittable flow solution has a cost of at least 60 with a capacity violation of no more than 15. Additionally, according to a report by @新智元, Sol+Codex successfully solved 6 difficult problems within five days.

Unconfirmed

Currently, this is primarily circulating as an anecdote and meme in the AI community. The posts lack rigorous technical details and formal peer review. Furthermore, there are discrepancies among reposts regarding how long the conjecture has been open (some claim 22 years, others 30 years), and it is referred to by multiple names, such as the Dinitz-Garg-Goemans conjecture and WOWII Conjecture 91. Its academic validity remains to be verified.

2026-07-24 ~ 2026-07-24 · 5 related posts

Primary sources