Fields Medalist: ChatGPT 5.5 Pro Solved Open Math Problems at ‘PhD Level’
LONDON — Fields Medal-winning mathematician Timothy Gowers said OpenAI’s ChatGPT 5.5 Pro solved open number theory problems at a “PhD level” in under two hours, according to The Decoder.
Gowers tested the AI model on open problems in number theory and found that it improved an exponential bound to a polynomial one in under an hour — a result that would represent a meaningful contribution to the field if produced by a human researcher.
An MIT researcher involved in evaluating the work called the model’s key idea “completely original,” according to The Decoder, suggesting the AI system was not merely recombining known techniques but generating novel mathematical reasoning.
AI systems have previously demonstrated proficiency in competition-level mathematics and formal proof verification. Producing original research-grade contributions to open problems, if confirmed, would represent an advance in capability beyond what has been previously documented at this level.
Gowers’ assessment carries particular weight given his stature in the field. He received the Fields Medal — often called the Nobel Prize of mathematics — in 1998 for his contributions to combinatorics and functional analysis, and holds positions at the Collège de France and the University of Cambridge.
The mathematician’s takeaway was pointed: the bar for what constitutes a meaningful human contribution to mathematics has now shifted. Gowers suggested that the standard for mathematical work has moved to proving things that large language models cannot, according to The Decoder.
ChatGPT 5.5 Pro is a product of San Francisco-based OpenAI. The involvement of MIT researchers in the evaluation underscores relevance to the U.S. academic establishment. The result adds to evidence that frontier AI models are approaching expert-level performance in scientific reasoning, according to The Decoder.
OpenAI had not publicly commented on Gowers’ assessment as of Friday.
