OpenAI has said a newly trained internal AI model has resolved more than 100 long-standing open problems across multiple areas of mathematics, signalling what the company describes as a rapid advance in AI-driven mathematical research. The disclosure follows OpenAI’s recent announcement of an AI-generated solution to the Navier–Stokes existence and smoothness problem, one of the seven Millennium Prize Problems.
According to OpenAI, the internal model used for the Navier–Stokes work is significantly more capable in mathematics than GPT-6 Astra. The company said the model’s progress has surprised its own mathematicians and has prompted internal discussions about how the mathematical community should be informed about the pace of AI development and how the field may need to adapt.
OpenAI's Navier–Stokes work involved a multi-agent system in which agents explored different approaches to the problem and exchanged intermediate results. The company said the agents produced a resolution approximately 88 hours after the effort began, followed by a further 17 hours for Lean formalisation and verification using GPT-6 Astra.
The development comes amid wider debate within the mathematics community about the use of open mathematical problems as benchmarks for AI systems. In an open letter published on 11 September, a group of mathematicians argued that the growing emphasis on AI systems solving major open problems could create negative externalities for mathematical research and contribute to a misalignment between AI companies and the mathematical community.
In response, OpenAI said it is working with mathematicians to establish an independent mathematics advisory group. The group will advise on the review and communication of emerging mathematical results, dissemination practices and academic and professional standards. It will also advise on how AI tools can support mathematical research and learning.
OpenAI said the advisory group will operate independently, with the freedom to provide unsolicited advice, comment publicly on the company's impact on mathematics and challenge OpenAI's decisions. Its members will not be paid by OpenAI, according to the company.
The company has framed the development as part of a broader effort to build AI systems capable of accelerating scientific discovery. In August, OpenAI also published a set of 10 AI-generated mathematical and theoretical computer science results covering areas including geometry, coding theory, group theory, quantum complexity and combinatorics.
The latest disclosure therefore extends OpenAI's focus from AI-assisted problem solving towards the potential use of increasingly capable models as research tools for mathematicians, while also highlighting the need for independent review, appropriate attribution and responsible dissemination of AI-generated mathematical discoveries.
