OpenAI Releases New Batch of Mathematical Results From an Unreleased Model
OpenAI has disclosed 722 manuscripts with solutions to long-standing math problems, produced by a frontier model that has not yet been released. The independent advisory group AGMAI says they include solutions to hundreds of open questions.
OpenAI has made public 722 manuscripts containing solutions to a range of long-standing mathematics problems. The work was produced by an unreleased frontier model and is organized into 372 result families, which group related papers together.
According to AGMAI, a newly formed independent advisory group of leading mathematicians set up to help communicate such results responsibly, the release includes solutions to "hundreds" of open questions.
What the release contains
The results had been anticipated for weeks, but until now OpenAI had not said which problems the model solved or when the material would appear. In September, the company stated that the model had "resolved more than 100 long-standing open problems across most areas of mathematics." The newly published materials also include some summaries of the model's reasoning, estimates of compute used, and statistics on how many problems were attempted. OpenAI says the average result required the equivalent of three hours of ChatGPT Pro thinking.
The papers are hosted in a GitHub repository, along with protocols for revisions and citations. The company says it is still exploring community-hosted alternatives that would meet the committee's guidelines.
Ethics debate
AGMAI, the Advisory Group on Mathematics and Artificial Intelligence, issued its first recommendations in late September. It urged AI labs to publish mathematical results promptly and, where possible, through established academic channels, while disclosing the model name, prompts and compute costs. The group also asked companies not to treat such releases as marketing vehicles for their models, saying the practice significantly harms the mathematical community.
The full impact of the results will take time to become clear as mathematicians evaluate them. Rival labs such as Anthropic have also produced results this year, some concerning a Millennium Prize problem. The speed at which AI companies have entered the field has sparked fierce debate over research practices, ethics and how the human mathematicians whose work these systems build on are credited.
