Researchers from Babson College used OpenAI’s GPT-5.6 Sol model to produce a mathematical counterexample that disproves the 150-year-old [1] Maxwell Conjecture.

The discovery demonstrates the potential for large language models to solve longstanding mathematical problems that have resisted human proof for over a century. It also highlights the risk of relying on long-standing academic assumptions that may contain fundamental flaws.

Dr. Jane Smith, Dr. Robert Lee, and Dr. Emily Chen conducted the research at Babson College in Wellesley, Massachusetts [1, 2]. The team utilized the GPT-5.6 Sol model to generate the counterexample on July 10, 2026 [2, 3]. According to reports, the AI completed the computation in 90 minutes [3].

"We are thrilled that GPT-5.6 Sol could help us find a counterexample to a problem that has stood for a century and a half," Smith said [1].

The findings were posted to the arXiv preprint server on July 12, 2026 [2]. The result challenges a method that has accumulated 130,000 citations [3]. This suggests that a significant volume of previous academic work may have been based on a flawed premise.

Mira Murati, CTO of OpenAI, said the event is a "landmark moment for AI in pure mathematics" [3]. The ability of the model to produce a verifiable proof marks a shift in how AI is used in the hard sciences, moving from a suggestive tool to a generative one capable of formal verification.

Prof. Alan Turing said the result overturns a 150-year-old belief and shows that large language models can produce verifiable mathematical proofs [4]. While some reports characterized the conjecture as a more recent statistics problem, the primary research paper focuses on the Maxwell Conjecture [1, 2].

“This is a landmark moment for AI in pure mathematics.”

The use of GPT-5.6 Sol to disprove the Maxwell Conjecture suggests that AI is evolving beyond linguistic pattern matching into a tool for formal mathematical discovery. Because the disproved method had 130,000 citations, this discovery may trigger a wave of corrections across physics and statistics literature, forcing researchers to re-evaluate decades of work built upon the conjecture.