OpenAI's GPT-5.4 Pro reportedly solves a longstanding open Erdős math problem in under two hours

A viral, unverified report claims OpenAI's next-generation model, GPT-5.4 Pro, has solved a specific, unsolved problem from the field of Erdős discrepancy theory. The problem, concerning the limits of sequences under signed sums, is a well-known combinatorial puzzle that has resisted formal proof for decades. According to the claim, the model produced a correct, novel proof in under two hours of reasoning, a feat that would represent a significant milestone in AI's ability to perform deep, structured mathematical deduction rather than just statistical pattern matching.

While OpenAI has not officially confirmed the model's existence or the result, the report has ignited intense discussion in AI and mathematics circles. The achievement would imply major advances in the model's chain-of-thought reasoning, handling of extremely long and complex logical chains, and its ability to navigate formal constraint systems. Experts caution that independent verification by human mathematicians is essential, but acknowledge that such a demonstration would mark a clear shift from AI as a research assistant to a potential co-author in theoretical discovery.

The implications extend beyond pure mathematics. Success in this formal domain suggests the underlying architecture could significantly improve reasoning reliability in other fields requiring rigorous logic, such as code verification, cryptographic analysis, and complex system design. It also raises immediate questions about the model's release timeline, its accessibility to researchers, and how such problem-solving 'agents' might be integrated into the scientific workflow.