OpenAI Releases 722 AI-Generated Mathematics Papers, Including Advances on Riemann Hypothesis

A collection of 722 mathematics papers generated by OpenAI's unreleased artificial intelligence model went live on GitHub this week. The papers represent a significant computational contribution to mathematical research, with findings that establish new proofs, refute existing conjectures, and constrain the solution space for long-standing unsolved problems across approximately 20 distinct mathematical subfields.

Among the most prominent results is work related to the Riemann hypothesis, a conjecture more than 150 years old that concerns the distribution of prime numbers—integers divisible only by themselves and one. The hypothesis posits that prime numbers follow a pattern governed by the Riemann zeta function. While a complete proof remains elusive, OpenAI's model succeeded in proving the quasi-Riemann hypothesis, an important intermediate result. The significance of resolving the full Riemann hypothesis extends beyond pure mathematics: countless subsequent mathematical papers assume its truth, meaning a proof would validate an entire body of dependent research.

Theoretical computer science emerged as another major research focus, with the AI model producing more than 80 papers in this domain. Three papers specifically addressed matrix multiplications, the mathematical operations underlying how AI systems process information. For decades, researchers have pursued faster and more hardware-efficient multiplication methods, operating under the assumption that computational speed improvements have an upper bound. OpenAI's model clarified the nature of this limit with greater precision. In parallel work, the system developed a novel algorithm for multiplying integers, the fundamental data structures holding whole numbers that serve as building blocks in numerous computational applications.

The research also yielded findings relevant to physics and engineering. OpenAI published more than a dozen proofs addressing partial differential equations, or PDEs—mathematical formulations essential to semiconductor design, structural engineering, quantum mechanics and related disciplines. The model proved a variant of De Giorgi's conjecture, which pertains to equations used in studying metal alloy behavior. Additionally, it resolved several open questions surrounding the Navier–Stokes equations, which engineers employ to model fluid dynamics. In September, the same model had previously solved a separate Navier–Stokes problem ranked among mathematics' most challenging open questions.

Many of the newly published papers include Lean files—code segments that enable computational verification of proofs without manual review. OpenAI intends to expand Lean proof availability across additional papers and has committed to funding research conferences and initiatives dedicated to evaluating AI-generated mathematical discoveries.