OpenAI releases 722 math manuscripts from unreleased frontier model
11 sources · The Neuron · The Decoder · Gizmodo- Doom: 25 Fields Medal winners warn mass-produced math results could damage research fields
- Boom: OpenAI published 722 manuscripts and 372 result families from an unreleased frontier model
- Boom: Lean formalizations included for machine verification; each result used ~3 hours of ChatGPT Pro compute
- Doom: Mathematicians cite concerns over research ethics and describe AI company conduct as "mobster behavior"
- Neutral: Independent advisory group AGMAI, formed from elite mathematicians, created to engage with OpenAI output
The story in full
On October 6, 2026, OpenAI published 722 mathematical manuscripts on GitHub covering 372 result families, produced by an internal frontier model not yet publicly released. The company also shared Lean proof formalizations for machine verification, with each result consuming roughly three hours of ChatGPT Pro compute on average. The release extends an earlier batch of 10 open problem solutions OpenAI had previously posted.
The drop has drawn sharp criticism from parts of the mathematical community. Twenty-five Fields Medal winners have warned that mass-producing mathematical results could damage the research environment rather than advance understanding. One mathematician described a "perception of mobster behavior" from leading AI companies, and questions have been raised about research ethics and academic conduct. An independent advisory group called AGMAI, composed of elite mathematicians, has been formed to engage with the work.
Analysis
383 wordsOn October 6, 2026, OpenAI published 722 mathematical manuscripts on GitHub covering 372 result families, produced by an internal frontier model that has not been publicly released. The release included Lean proof formalizations to allow machine verification, and OpenAI disclosed that each result consumed roughly three hours of ChatGPT Pro compute on average. The drop followed an earlier release of solutions to 10 open problems and prompted the formation of AGMAI, an independent advisory group drawn from elite mathematicians, to engage with the output.
The scale of the release is what makes it contested. Mathematics traditionally advances through peer collaboration, with open problems serving as long-term motivational anchors for careers and communities. Twenty-five Fields Medal winners have warned that mass-producing results could damage the research environment rather than advance it, and one mathematician told WIRED there is a "perception of mobster behavior" from leading AI companies. The core dispute is not simply whether the results are correct, but who benefits, who was consulted, and what happens to a field when its open problems are bulk-resolved by a corporate model without meaningful community involvement.
The Pro-AI camp treats the release as a historic milestone. Jarno N. Alanko highlighted specific results, including a subset-sum algorithm faster than the textbook approach and a 2-approximation for the shortest common superstring, calling it "what a morning." DevNews argued OpenAI "may have just changed math forever." Critics in the Anti-AI camp are less impressed by the results than alarmed by the method. Alexander Kasprzyk argued that OpenAI's concern is not the mathematics community but its own influence, describing the release as "fire-and-forget." Sos Sosowski drew a sharper analogy, comparing the dynamic to hiring a thief. The Middle Ground camp is asking procedural questions that neither side has fully answered. Vladimir Salnikov objected to GitHub as a venue for mathematical research, Rolf van der Hulst flagged the absence of reported inference costs, and Jarno N. Alanko noted that at least one claim, a non-constructive proof of matrix multiplication in O(n^(9/4)), raises immediate questions about practical value.
What to watch is whether independent mathematicians working through AGMAI can verify or challenge the manuscripts at scale, and whether OpenAI releases details on the frontier model itself, which would allow the community to assess reproducibility and decide how much weight the results deserve.
What Pro-AI voices are sayingThis release represents a historic leap for mathematics, equivalent to decades of human progress, and sharing it openly on GitHub rather than behind paywalls is exactly the right move. Some also want OpenAI to now direct similar compute toward medical and scientific breakthroughs.
Quote 1 of 13What Anti-AI voices are sayingOpenAI is depleting mathematics of open problems without adequately involving the community, and the bulk release blindsided researchers despite prior assurances otherwise. Some doubt the proofs are reliable at all, while others accept the results but object to how the field is being treated.
Quote 1 of 13What Middle Ground voices are sayingMany are cautiously impressed but want independent verification and clearer community involvement before drawing conclusions about significance. A notable skeptical thread questions whether results with impractical parameters or unverified claims represent genuine scientific value.
Quote 1 of 13Add your take
0 reader votesSign in with Google to pick a side and post. Your vote moves the story's Doom / Boom score.
More Pro-AI reactions (12)
“this is a take against progress so that mathematicians can keep their jobs. What would we do if, instead of math, we were talking about diseases? Are we going to keep diseases around so that doctors can keep their jobs”
aaraujo002, Hacker News · 22:44 UTC“In a way this is probably 50-100 years of math progress by humans”
gizmodo59, Hacker News · 22:34 UTC“That copium didn't last for what, three months?”
yewenjie, Hacker News · 22:57 UTC“Extremely unfortunate that gate keeping got to the point where they felt the need to ask for permission to share math.”
karahime, Hacker News · 22:27 UTC“A valid proof is a big deal!”
enoether, Hacker News · 22:34 UTC“Six years later we are beginning to understand the answer to this question.”
xanderlewis, Hacker News · 23:36 UTC“This includes a proof of Barnette's Conjecture, which is one of the graph theory conjectures that I tried attacking with SOTA models a few months ago. I like it because it is easy to understand with a basic knowledge of”
prideout, Hacker News · 22:54 UTC“Morning coffee proven turing complete”
kingstnap, Hacker News · 22:54 UTC“"Here you go, have fun with this, fuck all your demands, by the way we're gonna be releasing the model stay tuned!"”
Catloafdev, Hacker News · 22:38 UTC“these OpenAI proofs are probably correct and not plagiarized in any normal sense of the word and highly consequential”
Colin, Bluesky · 00:44 UTC“OpenAI just made the single largest advancement in mathematics and number theory in modern history.”
Alejandra Caraballo, Bluesky · 23:44 UTC“if it holds, it's the biggest math news ever basically”
cee, Bluesky · 23:11 UTC
More Anti-AI reactions (12)
“Prediction: one of these is wrong and this (publicity stunt) will backfire.”
mathisfun123, Hacker News, skeptic · 22:46 UTC“Seems like a lot of PHD students are doing to have to pivot the entire structure of their PHD studies? Or just produce something which is already written by OpenAI?”
open592, Hacker News · 22:41 UTC“As an AI "doomer" can I ask the non-doomer people here how you interpret the significance of results like these, and what kind of progress you expect to see in the next 1-5 years?”
againstapples, Hacker News · 23:31 UTC“Why didn't they just give mathematicians access, so they could at least understand and write up the results in publishable form?”
applicative, Hacker News · 23:21 UTC“this does not feel like the message hit OpenAI where it needed to hit, if this is their primary response.”
rafterydj, Hacker News · 22:31 UTC“Hey how about 722 advanced math papers dumped on everyone’s head. Don’t worry, there’s a forty page catalog summarizing all the claimed results. github.com/openai/math”
Kieran Healy, Bluesky · 23:54 UTC“Indiscriminate brute-force application of AI to mathematics is depleting the field of good questions without renewing this resource or providing commensurate value.”
Carl T. Bergstrom, Bluesky · 02:34 UTC“mathematicians are angry about how the field is being treated; meanwhile, OpenAI workers consider math dead as a field of inquiry.”
Tim Marchman, Bluesky · 18:03 UTC“OpenAI Releases the Final Ten Minutes of 500 Previously Unreleased Films, Ushering in a New Era of Movie Watching”
Terence Tao, Bluesky · 14:53 UTC“OPENAI/MATH IS AI SLOP”
aegisi👻ndure, Bluesky, skeptic · 02:08 UTC““There's a perception of mobster behavior” from leading AI companies, one mathematician tells WIRED as OpenAI prepares to release more than 100 new solutions to unsolved problems.”
WIRED, Bluesky · 17:33 UTC“They systematically took our motivational questions, perhaps with the…”
Alexander Kasprzyk, Bluesky · 06:16 UTC
More Middle Ground reactions (12)
“I am curious if any of the results have immediate applications in any kind of engineering or science.”
dekhn, Hacker News · 22:49 UTC“I think they should put human names on the papers as someone who has reviewed the result, even if just a preliminary review.”
ks2048, Hacker News · 22:44 UTC“That is some crazy exponent -- plus an interestingly old computational model to boot”
NotOscarWilde, Hacker News · 23:24 UTC“OpenAI just blooped hundreds of proofs and counterexamples to open problems”
Zach Weinersmith, Bluesky · 22:46 UTC“I guess we are about to find out how much duplicate work there has been during postponed releases.”
Pekka Lund, Bluesky · 11:02 UTC“We're fucked, math is over! Also me: I can't even understand the statements of these results! Now we need more mathematicians than ever before! It's the golden age for math now!!”
Dulwich Quantum Computing, Bluesky · 02:13 UTC“sorry, 722 problems but even more interesting might be the 3278 that it didn’t solve”
mr. TIM, Bluesky, skeptic · 00:26 UTC“Despite the scientific advance this must be a difficult day for the thousands (?) of people that were working on these.”
Pedro Beltrao, Bluesky · 06:39 UTC“OpenAI solved math tonight or something; anyway I have an early flight to SF for unrelated reasons so hopefully the world keeps turning a bit longer”
Chaparral Coyote, Bluesky, skeptic · 01:30 UTC“it's funny to think that openai is showing off how cheaply they can do math compared to their peer”
value plus discounted, Bluesky · 23:17 UTC“The best thing we could do would be to start promoting the continuing value of mathematics, targeting 14-16 year olds. We need an advertising campaign. Large. Soon. I don’t see this as a “fight” between us and OpenAI: frankly I”
Alexander Kasprzyk, Bluesky · 07:27 UTC“Using countless tokens, OpenAI brute forced a bunch of Lean code that *may* approach a solution.”
Hoon, Bluesky, skeptic · 02:21 UTC
Sources
13 articles from 11 outlets- The Neuron😺 OpenAI’s AI produced 722 math manuscripts
- The DecoderOpenAI dumps 372 AI-generated math proofs on GitHub, telling the academic world to keep up
- The DecoderOpenAI dumps 372 AI-generated math proofs on GitHub, telling the academic world to keep up
- GizmodoOpenAI Dumps 377 New Math Results on GitHub, Publishes Hand-Wringing Blog Post
- The Verge AIOpenAI drops another batch of mathematical breakthroughs
- Hacker News front page (AI)Mathematical manuscripts and supporting proof artifacts produced by OpenAI
- OpenAISharing AI progress in mathematics
- Hacker News front page (AI)Sharing AI Progress in Mathematics
- Scientific AmericanOpenAI unleashes hundreds more math results upon a field already in shock
- Crypto BriefingOpenAI previously released solutions to 10 open math problems on GitHub
- Wired AIOpenAI Is Pissing Off a Bunch of Mathematicians—Again
- WIREDOpenAI Is Pissing Off a Bunch of Mathematicians—Again
- OpenAI newsSharing AI progress in mathematics
Related Splits
All stories →
Common Sense Media calls ChatGPT for Teens an unacceptable risk
On October 7, 2026, Common Sense Media, a nonprofit focused on youth safety, published an assessment concluding that OpenAI's ChatGPT for Teens poses an "unacceptable risk"...
Jump Trading uses ChatGPT to scale quantitative research workflows
Jump Trading, a quantitative trading firm, has adopted ChatGPT to expand its quantitative research operations, according to a case study published by OpenAI on October 6,...
OpenAI and Ironclad partner to train AI agents on contracting work
OpenAI and Ironclad announced a partnership on October 6, 2026, focused on training and evaluating AI agents on complex contracting workflows. The collaboration is described as...