# OpenAI's one-prompt maths haul claims the Unique Games Conjecture

> Scott Aaronson says OpenAI's 372 maths results include a Lean-verified proof of the Unique Games Conjecture.

*A claimed Lean-verified proof inside OpenAI's 372-result release would be the strongest credibility signal yet for its research push, if outside experts confirm it.*

By Bez · InsidersFeed
Canonical: https://insidersfeed.com/news/openai-s-one-prompt-maths-haul-claims-the-unique-games-conjecture

OpenAI's bid to be taken seriously as a research engine now rests on one claim: that its latest maths release contains a machine-checked proof of the Unique Games Conjecture. The claim is unconfirmed by outside experts, and the stakes for the company's credibility are high.

The claim comes from computer scientist Scott Aaronson, who wrote about the release in a post titled [The Mathocalypse](https://scottaaronson.blog/?p=10169) on his Shtetl-Optimized blog on 8 October. He reports that OpenAI's release of 372 maths results includes a proof of the conjecture that has been verified in Lean, the proof-checking language.

Aaronson also gave the result a personal frame. His wife, the complexity theorist Dana Moshkovitz, worked toward it for her career.

The volume and method matter as much as the headline result. According to [AI Weekly](https://aiweekly.co/alerts/openai-says-372-math-results-came-mostly-from-one-prompt), OpenAI says most of the 372 results came from one prompt. That points to a cheap, repeatable process rather than a long effort by a team of researchers, though the notes give no cost or compute figures.

The post travelled fast. Four tracked experts shared it within hours of publication.

## The Insider view

If the proof holds, OpenAI gains a proof point that no benchmark score can match: a famous open problem, closed and checked by software. The one-prompt origin sharpens that. It suggests the value lies in a method that could be repeated, which is what customers and investors in research tooling would want to see.

The exposure is just as clear. Our desk has not independently verified any of this, and the account rests on a blog post and OpenAI's own description of how the results were produced. Outside mathematicians still have to check both the proof and the Lean formalisation. Until they report, treat the claim as a strong lead rather than a settled result, and watch whether the 372 results other than the headline one survive the same scrutiny.

## Key takeaways

- OpenAI's 372-result maths release reportedly includes a Lean-verified Unique Games Conjecture proof.
- OpenAI says most of the 372 results came from a single prompt.
- Nothing is independently verified yet; outside mathematicians must check the proof and formalisation.

## Sources

- [The Mathocalypse](https://scottaaronson.blog/?p=10169) — Shtetl-Optimized (Scott Aaronson), 2026-10-08
- [OpenAI says 372 math results came mostly from one prompt](https://aiweekly.co/alerts/openai-says-372-math-results-came-mostly-from-one-prompt) — AI Weekly, 2026-10-07