AI systems aren’t just winning gold medals in Math Olympiads, but they’re also beginning to help professional mathematicians and physicists.
OpenAI’s Chief Research Officer Mark Chen shared insights about GPT-5 Pro’s unexpected capabilities in advancing hard sciences. His observations, drawn from real-world testing with professional researchers, highlight a significant leap in AI’s ability to tackle complex scientific problems that previously required extensive human expertise and time.
Chen emphasized what struck him most about the new model: “I think one big thing for me was just how much it moved the frontier in very hard sciences.” This advancement became apparent through direct collaboration with experts in the field. “We would try the models with some of our friends who are professional physicists or professional mathematicians,” Chen explained.
The results have been documented publicly, with researchers sharing their experiences on social media platforms. “You already saw some instances of this on Twitter where you can take a problem and have it discover maybe not very complicated new mathematics, but some non-trivial new mathematics,” Chen noted.
The pattern of discovery has been consistent across multiple interactions with researchers. “We see physicists, mathematicians repeating this experience over and over where they’re trying GPT-5 Pro and saying, ‘Wow, this is something that the previous version of the models couldn’t do,'” Chen observed. For these professionals, the capability represents a breakthrough moment. “It is a little bit of a light bulb moment for them.”
Perhaps most significantly, Chen highlighted the model’s potential to dramatically accelerate research timelines. “It’s able to automate maybe what could take one of their students months of time,” he said.
The implications of Chen’s observations extend far beyond incremental improvements in AI capabilities. GPT-5 Pro appears to represent a inflection point where AI systems can serve as genuine research partners rather than merely advanced calculators or search engines. This development could fundamentally reshape how scientific research is conducted, potentially democratizing access to high-level mathematical and physical analysis while accelerating the pace of discovery across multiple disciplines. As AI systems become capable of automating months of graduate-level research work, the scientific community may need to reconsider traditional research methodologies and supervision models, while also exploring new collaborative frameworks between human researchers and artificial intelligence.