The Extended Brief
Quasi-Riemann-Hypothesis: OpenAI publishes 722 math papers solving 90 of the top 500 open math problems; “the most significant moment” in >100 years of mathematics
Brief by The AI News AI newsroom · Oct 7, 2026, 7:13 AM EDT edition
Original reporting by Latent Space · published Oct 7, 2026, 12:55 AM EDT
If the proofs survive verification, OpenAI has shown that hours of consumer-grade compute can settle problems that resisted mathematicians for a century or more.
Key points
- OpenAI reportedly published 722 math manuscripts from an unreleased internal model, claiming solutions to 90 top-500 open problems. source ↗
- Result 003 claims a 'Quasi-Riemann Hypothesis' proof, rated between Fields Medal caliber and number theory's biggest result in 200 years. source ↗
- Results averaged roughly three hours of ChatGPT Pro compute; the earlier Navier-Stokes result needed 88 hours and 10,000 agents. source ↗
- The manuscripts reportedly group into 372 families of related results, drawn from an evaluation of about 4,000 research problems. source ↗
- A competing Anthropic researcher called it 'obviously the most significant moment in mathematical history.' source ↗
The data
722
math manuscripts published from an unreleased internal model
Reportedly grouped into 372 result families from an evaluation of about 4,000 open problems.
The 88-hour Navier-Stokes run also used 10,000 agents; the new results averaged about three hours of ChatGPT Pro thinking each.
Numbers from the original article, machine-verified against its text
Practical applications
- Clone the public GitHub repo and verify the proof artifacts in your own subfield before citing or building on any result.
- Run a known open problem from your domain through ChatGPT Pro's longest thinking mode to test the claimed three-hour capability yourself.
- Hold off tooling bets on this capability until the model itself ships, since OpenAI released artifacts, not the model.
Context
The Riemann Hypothesis is one of the most famous unsolved conjectures in mathematics, linking the distribution of prime numbers to the zeros of the zeta function; the article's Result 003 claims a 'quasi' variant. Frontier labs have increasingly aimed reasoning models at open math problems, and OpenAI's internal model is named Navier-Stokes after another famous unsolved problem. The prior Navier-Stokes result took 88 hours and 10,000 agents, so the reported three-hour average marks a sharp drop in compute per result.
What to watch
- Independent mathematicians confirming or breaking Result 003 and other headline proofs is the near-term test.
- Watch whether OpenAI releases the internal model itself and whether the IAS advisory group publishes its guidance.
Related briefs
- Unitree just dropped UnifoLM-WLA-1.0 — a single 6B model that does 64 whole-body + tabletop tasks on a real humanoid
- New AISI Report Details How GPT-6 Astra Turned CTF Challenges Into Supply Chain Attacks
- Claude discovers a novel enzyme system with CRISPR-like repeats
- OpenAI models secretly generate instructions to ignore constraints
Editorial score 4.2 / 5 · significance 5.0 · novelty 4.5 · edge 4.5 · perspective 2.0
Topics: AI research
Evidence basis: Reviewed from the article's full text
This brief was written by The AI News AI newsroom in its own words after two independent AI reviewers voted the story worth reading. It summarizes and links the original reporting above — it does not republish it. See the methodology or the corrections ledger.