OpenAI Open-Sources 722 AI-Generated Math Manuscripts, Sparking Debate

The algorithmic complexity record for matrix multiplication has been dramatically lowered to O(N^2.25), a significant improvement over the previous record of roughly O(n^2.37). Matrix multiplication is the core operation in neural network training and inference, and every drop in the exponent directly affects large-scale compute efficiency, which is why this progress has drawn so much attention.

Confirmed

• Stanford professor and Matrix$$ founder Reza Zadeh posted announcing the discovery of a faster matrix multiplication algorithm, with links to the paper and code, noting it could be further improved with more compute.
• @Pascallisch relayed RokoMijic's claim: OpenAI proved matrix multiplication can be done in just O(n^2.25) operations, breaking the previous record of roughly O(n^2.37).
• Historical context: Strassen brought the exponent down to 2.81 in 1969; after Coppersmith–Winograd reached 2.376 in 1990, the record remained nearly stagnant until a string of recent breakthroughs (e.g., AlphaEvolve's result seven weeks ago).

Not yet confirmed

• Reza Zadeh speculates that major AI labs already possess matrix multiplication methods approaching the theoretical limit of exponent 2, though this is purely personal speculation.
• OpenAI's O(n^2.25) result reportedly lacks a concrete algorithmic construction, and posts disagree on who holds the record, so details await further verification.

Why it matters

• The exponent omega sets the theoretical speed ceiling for all matrix computation, directly affecting the cost of training and inference for large models.
• The record had been nearly stagnant for three decades, and has now been repeatedly broken in a short span—with the breakthroughs coming from the AI side, highlighting AI's potential to feed back into fundamental algorithm research.

2026-10-07 ~ 2026-10-08 · 246 related posts

Full story(2 episodes)→

• Episode 1: OpenAI Open-Sources 722 AI-Generated Math Manuscripts, Sparking Debate(2026-10-07, 246 posts)
• Episode 2: OpenAI's Fast Matrix Multiplication Proof Generalized to Arbitrary Fields(2026-10-07, 2 posts)

• Reza Zadeh claims faster matrix multiplication algorithm, suspects labs near exponent 2 — Reza_Zadeh · 2026-10-07
• [source] OpenAI releases 722 AI-generated math manuscripts with partial Lean proofs — OpenAI · 2026-10-07
• OpenAI Releases Mathematical Proofs, Adding Substance to Math Breakthrough Buzz — imjustnewatai · 2026-10-07
• OpenAI's openai/math repo: 722 model-generated math manuscripts, some unformalized — aran_nayebi · 2026-10-07
• OpenAI releases new math results from an internal frontier model — kimmonismus · 2026-10-07
• OpenAI publishes 722 math manuscripts by unreleased internal frontier model — kimmonismus · 2026-10-07
• Mistral Researcher Hails OpenAI Models Solving Important Math Problems — charliermarsh · 2026-10-07
• OpenAI details AI math progress, publishes manuscripts and proof artifacts publicly — AMBNNJ · 2026-10-07
• OpenAI releases 722 math papers from an internal frontier model, including a claimed quasi-Riemann proof — imjustnewatai · 2026-10-07
• Mathematician celebrates AI model cracking a problem he chased for a decade — ctjlewis · 2026-10-07
• OpenAI's frontier model claims progress on Riemann, Hodge, BSD and a faster FFT — apples_jimmy · 2026-10-07
• OpenAI releases broad new mathematical results from internal frontier model, advised by IAS group — IgorCarron · 2026-10-07
• OpenAI releases novel math results from unreleased frontier model, ~3h Pro compute per problem — gaganghotra_ · 2026-10-07
• OpenAI appears to have dropped nearly 1,000 math preprints at once — airkatakana · 2026-10-07
• OpenAI researcher on internal model solving long-revered problems: 'a Copernican shift' — weijie444 · 2026-10-07
• After Navier-Stokes, OpenAI's internal math model now claims pieces of two more Millennium Problems — physiopeng · 2026-10-07
• Dev asks GPT 6 Pro to rank 3 years of discoveries: 81% were AI-made, released today — willdepue · 2026-10-07
• AI accounts for 87% of top math discoveries in 3 years, humans just 13% — willdepue · 2026-10-07
• Two counterexamples to the Shafarevich conjecture spotted, littmath jokes about double-dipping — littmath · 2026-10-07
• 'If you're a mathematician, a nuke just hit ground zero': reactions to OpenAI's math release — inductionheads · 2026-10-07
• OpenAI releases hundreds of advanced math proofs from internal model on GitHub — Pascallisch · 2026-10-07
• Mathematicians Ask: Where Exactly Is OpenAI's Claimed Millennium Problem Progress? — littmath · 2026-10-07
• OpenAI's Frontier Model Drops Matrix Multiplication Exponent to 2.25, Shattering 2.37 Record — inductionheads · 2026-10-07
• Mathematician's student fumes after OpenAI releases results scooping his conjecture work — littmath · 2026-10-07
• Ethan Mollick mocks AI accounts pretending to grasp the latest math breakthroughs — emollick · 2026-10-07
• OpenAI math drop stuns researchers: 7/8 quasi-Riemann proof hailed — Singularitarian · 2026-10-07
• OpenAI Publishes Preprint Claiming Integer Multiplication Faster Than n log n — Singularitarian · 2026-10-07
• OpenAI releases new math results from an internal frontier model, some already solved by outside agents — repligate · 2026-10-07
• Terence Tao says Q_p section conjecture work is the most interesting in OpenAI's math dump — littmath · 2026-10-07
• Only 372 hits out of 4,000 problems: analysts probe failure rates in OpenAI's math output — basedjensen · 2026-10-07
• OpenAI Releases Frontier Model's New Math Results, Claiming 90 of Top 500 Open Problems Solved — inductionheads · 2026-10-07
• OpenAI uploads hundreds of model-solved open math problems with Lean proofs to GitHub — AI寒武纪 · 2026-10-07
• OpenAI's Lean Formalization Proves Weaker Bound for Mutually Unbiased Bases in Dimension Six — MarioKrenn6240 · 2026-10-07
• 'STOC, FOCS, SODA in One Shot': Scholar Sizes Up OpenAI's Batch of Mathematical Results — giannis_daras · 2026-10-07
• OpenAI's Frontier Model Produces 'Quasi-Riemann' Result; Mathematician Says It'd Win a Fields Medal — soumitrashukla9 · 2026-10-07
• Unique Games conjecture rumor confirmed, a major theoretical CS breakthrough — felpix_ · 2026-10-07
• OpenAI's math repo claims frontier-model proofs of quasi-Riemann, Hilbert's tenth over Q, and more — repligate · 2026-10-07
• OpenAI's math drop: BSD proven in many cases, L-functions zero-free for Re(s)>7/8 — snikolov · 2026-10-07
• OpenAI CDO Kevin Weil hails math repo release, points to physical sciences next — kevinweil · 2026-10-07

11 near-duplicate retellings: Reza_Zadeh · apples_jimmy · felpix_ · nabla_theta · rand_longevity · willdepue · AlexTensor · Justin_Halford_ · ctjlewis · kristoph · ZeroStateReflex