OpenAI posts a batch of math results from a model it has not released

This is a follow-up to two earlier stories. On Sept. 9, The File reported OpenAI’s claim that an internal model had resolved the Navier–Stokes Millennium Prize problem. On Sept. 22, The File reported the company’s work with an independent Advisory Group on Mathematics and Artificial Intelligence at the Institute for Advanced Study. Tuesday’s post is the release step that work pointed toward. It names that same advisory group. It does not repeat or update the Navier–Stokes claim, so this story does not tie the new papers to that claim beyond the shared group.

OpenAI’s page, opened by this desk Tuesday morning, says the company is “releasing a broad range of new mathematical results produced by an internal frontier model,” and that it is “working to responsibly release the model that produced these results.” It says it will pay for workshops and conferences so mathematicians can study the work. The repository, the company says, includes 10 summaries of the model’s reasoning, compute estimates, and statistics on how many problems were tried. “The average result used the equivalent compute of roughly three hours of ChatGPT Pro thinking.” OpenAI says it will add more Lean formalizations as it gets them.

A later check of the article page from this publish environment returned 403. The news feed still listed the item, with a publication time of 12:00 UTC on Oct. 6. The feed’s description matches the desk’s reading: new results on open problems from an internal frontier model, plus Lean proofs on GitHub. There is no model name and no public launch. No AI lab released a new model in this window.