A newly trained OpenAI AI model has cracked more than 100 world-class mathematical problems in just 24 days since launch, even touching the Navier–Stokes Millennium Prize problem; OpenAI’s own mathematicians are surprised by its pace. The model, which began training in late August, can already produce original discoveries in new fields of mathematics, and deployment of related capabilities is moving forward steadily. In actual use since launch, it has broken through the limits of traditional AI math tools that could only handle routine problem-solving and calculation assistance. Instead of simply matching outputs from existing mathematical knowledge bases, it can independently derive new mathematical conclusions not previously published in academia. Many niche problems that have troubled the mathematical community for years have found new breakthrough paths through its derivations. Even on the Navier–Stokes Millennium Prize problem—a top-tier challenge at the intersection of fundamental physics and mathematics—it has produced new derivation ideas with reference value. Its speed of iteration and breakthroughs far exceeds OpenAI’s earlier internal expectations. Deployment is proceeding steadily according to a preset pace, and it is expected to provide new research support for mathematicians worldwide.

Behind this is an independent Mathematics and AI advisory group jointly established by OpenAI and the Institute for Advanced Study in Princeton. Its nine members are all top mathematicians, including three Fields Medalists. Members receive no compensation from OpenAI and may independently issue public comments and adjust the group's composition. The fully independent advisory group will oversee academic standards, review, dissemination, and deployment direction for AI-generated mathematical results, directly responding to earlier concerns about AI mathematics ethics jointly raised by mathematicians such as Terence Tao. This advisory structure, completely outside OpenAI’s conventional R&D system, prevents commercially driven interference in AI mathematical research from the root. All members maintain full academic independence; they are not bound by internal business metrics, and may independently publish academic views and adjust the group’s membership according to developments in the field. Its core responsibility is to ensure academic rigor for AI-generated mathematical results, regulate the dissemination pathways and application boundaries of such results, and directly address common concerns raised earlier by Terence Tao and other leading mathematicians about possible academic inaccuracy and ethical loss of control after AI enters mathematical research. It explores a more credible collaboration model for deep integration between AI and fundamental mathematics.