OpenAI drops another batch of mathematical breakthroughs

OpenAI has released a batch of 722 manuscripts, grouped into 372 result families, produced by an unreleased frontier model. The independent Advisory Group on Mathematics and Artificial Intelligence, or AGMAI, says the release includes solutions to "hundreds" of open questions. In September, OpenAI had said the model resolved more than 100 long-standing open problems.

The release also includes some summaries of the model's reasoning, compute estimates and statistics on how many problems were attempted. OpenAI says the average result used the equivalent of about three hours of ChatGPT Pro thinking. It is publishing on GitHub, with protocols for revisions and citations, and says it is still exploring community-hosted alternatives.

If even a fraction of these results hold, this is a remarkable moment for useful AI, and we don't want the controversy to bury that. But the number that stopped us is 722. Mathematics advances when people read proofs, check them, argue about them and fold them into what they already know. That work is done by a limited pool of experts, mostly unpaid, on their own schedule. A single drop of hundreds of papers is not a gift that arrives ready to use. It is a workload.

It's fair to grade this release against the advisory group's own guidance, since OpenAI says it drew on it. AGMAI's late-September recommendations asked labs to release promptly, use established academic channels where possible, disclose the model's name, prompts and compute, and stop treating math results as marketing. OpenAI moved on some of that: compute estimates, reasoning summaries and attempt statistics are genuine disclosures. But the model is still described only as unreleased, GitHub is not an established academic channel, and OpenAI itself frames other venues as something it is still exploring. That reads to us like partial compliance, not full.

The harm if this pattern scales is concrete. Human mathematicians whose work these systems build on may struggle to get credit. Early-career researchers could find their open problems closed overnight by a lab, with the paper landing in a repository instead of a journal. And the field's verification capacity, already stretched, becomes the bottleneck everyone else has to wait on.

Our view: release the results, absolutely. Hoarding them would be worse. But pace and format matter. A lab that can generate hundreds of papers should also help pay for the human checking they require.

This release adds to a fast-growing body of AI-produced math from OpenAI and rival labs including Anthropic, including results concerning a Millennium Prize problem. The way labs have announced these results this year has set off a sharp debate over research ethics and credit.