OpenAI publishes 372 AI-generated math results on GitHub, claims they solve or advance open problems
OpenAI has published 372 mathematical results generated by an unnamed internal frontier model, hosted on GitHub instead of in peer-reviewed journals. The company claims each result solves or substantially advances an open problem, at an average of about three hours of ChatGPT Pro Thinking compute per result. Many include Lean formalizations, but independent validation of significance is still pending.
OpenAI has published 372 mathematical results generated by an internal frontier model, hosting them in a GitHub repository rather than submitting them to academic journals. The company claims each result solves an open problem or makes substantial progress toward one. The model has not been named or released.
What OpenAI published
According to OpenAI, the collection includes improvements to major computer algorithms and advances related to the Riemann hypothesis. The repository includes revision logs and citations. OpenAI has not disclosed the model's name, parameter count, context window, or pricing.
OpenAI says nearly every result came from a single prompt to a single AI agent, though some required multiple attempts. The company says each result consumed roughly three hours of ChatGPT Pro Thinking compute on average. OpenAI shared only average compute costs, not per-problem figures, and did not publish the prompts.
The company also published methodology details: summaries of the model's reasoning process, statistics on how many problems the model attempted, and compute cost estimates.
The Navier-Stokes comparison
OpenAI says the same model previously produced a solution to a Navier-Stokes problem that has been under formal review for weeks. According to the report, that solution required a swarm of 10,000 agents and millions of dollars in compute, far more than the single-agent runs behind most of the 372 results. The Navier-Stokes result has not been confirmed by outside reviewers.
Formal verification as a review strategy
Many of the proofs ship with formalizations in Lean, a language for machine-checkable proofs, and OpenAI plans to add more. The rationale is review capacity: the volume of AI-generated output could overwhelm manual checking by mathematicians.
Lean can confirm that a proof is logically valid. It cannot judge whether a result is relevant, original, or interesting. Those judgments remain with human mathematicians.
Advisory group and limits
OpenAI consulted the Advisory Group on Mathematics and Artificial Intelligence at the Institute for Advanced Study, which includes Fields Medal winner Timothy Gowers, and says it loosely followed the group's public recommendations. It set one boundary in advance: the mathematicians can advise on how results are communicated, but not on whether or how fast they are produced.
OpenAI plans to fund workshops and conferences on understanding AI-produced results. It acknowledges its citations and presentation need improvement. It also says it is working on a responsible release of the model to "directly empower scientists with state-of-the-art capabilities." No release date has been given.
Community reaction
Reactions among mathematicians range from interest to frustration. In an open letter titled "A Severe Misalignment of AI in Mathematics," 25 Fields Medal winners argued that problem-solving is a proxy for the real goal of conceptual understanding. They warned that mass-producing true statements could damage the field rather than generate new ideas.
Gowers has warned that within one to two decades, the mathematical literature could grow enormously while no human community truly understands it. Terence Tao has argued that training young mathematicians should emphasize the human side and tightly limit AI tool use.
What this means
The significant claim here is not any single theorem but the cost structure. If a typical result really takes about three hours of ChatGPT Pro Thinking compute, the bottleneck in mathematics shifts from generating results to evaluating them. That is the logic behind OpenAI's emphasis on Lean.
Several things limit how much weight the numbers deserve today. The model is unnamed and unavailable. Prompts are unpublished and per-problem costs are undisclosed, so the "three hours" figure cannot be independently checked. The count of 372 does not show how many results are substantive. The attempt statistics OpenAI published should help with that, and the repository's contents can now be scrutinized.
Bypassing journals is a deliberate choice, and it sets up a conflict over who certifies mathematical knowledge. Formal verification can establish correctness, but significance still depends on the community, which is already publicly divided. How mathematicians receive these results, and whether the Navier-Stokes review concludes, will test OpenAI's claims more than the repository's size does.
Related Articles
Anthropic mathematician claims Riemann Hypothesis proof with Claude; Lean community certifies it correct
Anthropic mathematician Levent Alpöge said on November 20 that he and Claude found a proof of the Riemann Hypothesis, published as a formal proof in Lean. On November 21, the Lean and Mathlib communities issued a joint statement certifying the proof as correct. No human-readable paper exists yet.
Anthropic mathematician Levent Alpöge claims Claude-assisted proof of the Riemann hypothesis, posts it in Lean
Levent Alpöge, a mathematician employed by Anthropic, says he has proven the Riemann hypothesis, first proposed in 1859, working with Claude. He posted the proof coded in Lean, and experts have yet to confirm that it is correct.
OpenAI tells developers to rewrite system prompts for GPT-6.2 Astra and Sol as defaults shift
OpenAI says developers moving to GPT-6.2 Astra or Sol should review their system prompts. The company published copy-ready templates in the models' documentation, and the Responses API now discards earlier reasoning content if conversation history is edited.
OpenAI releases GPT-Live-1.1 in the API, with standby mode billing idle WebRTC time at $0.02 per minute
OpenAI has released GPT-Live-1.1 in the API. The full-duplex voice model improves multilingual accents and transcription, and a new standby mode bills idle WebRTC time at $0.02 per minute instead of $0.05.
Comments
Loading...