math

AI and Math in 2026: What OpenAI Claimed and How to Try It Yourself

In 2026 OpenAI said models advanced the unit-distance problem (outside-reviewed) and produced a Navier-Stokes singularity argument (disputed, not peer-reviewed when last checked). This guide separates confirmed from contested claims and points you at Graph Lab and related calculators so you can try the small cases yourself.

By UtilBloom · Published 2026-11-08 · Updated 2026-11-08

How to read this page

This story is still moving. Last updated 2026-11-08. We do not say the problems are definitely solved, and we do not say they definitely aren't. We split what outside reviewers backed from what is still disputed.

Claim 1: the unit distance problem (May 2026)

Erdős asked in 1946: if you place n points on a plane, how many pairs can sit exactly distance 1 apart? The best known constructions barely moved for decades. OpenAI said an internal general-purpose reasoning model found arrangements past that limit, and called it the first time a prominent open problem central to a subfield was solved autonomously by AI. Outside mathematicians reviewed and backed the work, including Noga Alon (Princeton). Daniel Litt (University of Toronto) described a geometry problem that became a number theory problem. Caveats matter. One expert said this may be a problem where experts had tried and missed a trick. In Oct 2025, OpenAI staff claimed GPT-5 solved ten Erdős problems, then walked that back because humans had already solved them.

Small cases you can plot

Three points: an equilateral triangle gives 3 unit pairs. Four points: two equilateral triangles sharing an edge give 5 unit pairs. Try it in Graph Lab. Plot A(0,0), B(1,0), C(0.5,0.866), D(1.5,0.866). Distance is d = √((x₂−x₁)² + (y₂−y₁)²). Five pairs sit at distance 1 (A–B, A–C, B–C, B–D, C–D). A–D is about 1.73. Open the unit-distance 4-point demo. Challenge: add a fifth point and try to beat five unit pairs. Check any pair with the Distance Between Points calculator, or the Scientific Calculator for the square root by hand.

Claim 2: Navier-Stokes (Sept 8, 2026) — disputed

The Navier-Stokes equations describe fluid motion (water, air, blood). The open Millennium Prize question: can a smooth 3D flow develop a singularity in finite time? There are seven Millennium Prize Problems at $1M each; only the Poincaré conjecture has been solved so far. OpenAI says an unreleased model with a large group of agents produced a proof that a fluid at rest, with a smooth force applied, can form a singularity in finite time. It said the argument was checked in Lean, and that it won't claim the prize. Terence Tao called it a remarkable achievement. What stays disputed: it was not peer-reviewed or confirmed by the Clay Institute when last checked. There is debate over whether it meets prize criteria. NYU's Tristan Buckmaster and Anthropic's Levent Alpöge say they reached a related result first; there is a credit dispute (OpenAI says its staff and agents didn't see their work). OpenAI also said the same model solved 100+ other problems, and verification of that broader claim is an open question.

Reynolds number: a concrete fluid check

Re = (density × speed × length) ÷ viscosity. Example: water ≈20°C, pipe diameter 0.05 m, speed 1 m/s, density ≈998 kg/m³, viscosity ≈0.001 Pa·s → (998 × 1 × 0.05) ÷ 0.001 = 49,900, firmly turbulent. Pipe flow is usually smooth below about 2,300 and turbulent above about 4,000. Run it in the Reynolds Number Calculator. Convert units first with the Unit Converter if your inputs aren't already SI.

A 5-question checklist for "AI solved math" headlines

1. Who checked it, and are they independent of the lab making the claim? 2. Is it peer-reviewed, or only announced? 3. Is it formally verified (for example in Lean), and was that check audited? 4. Was the result already known under another formulation? 5. Who gets credit if similar work appeared first elsewhere? Want to poke at related algebra in a notebook? Open Code Lab for a browser Python notebook.

Why it matters (both views)

Supporters say this is a new path: general models proposing arguments that experts can check, not just regurgitating textbooks. Skeptics say announcements moved faster than community verification, credit is contested on Navier-Stokes, and past walk-backs show you should wait for outside confirmation. Both can be true at once: the unit-distance work has outside backing; the Navier-Stokes claim does not yet have the same status.

Sources

Reporting paraphrased from Live Science, Scientific American, Warp News, Breaking Math, ThePrint, YourStory, Free Press Journal, Analytics India Magazine, and The Decoder. We did not copy their text or reuse their images. Recheck primary announcements before you treat any claim as settled.

FAQ

Did AI solve the unit distance problem?

OpenAI announced a construction beyond the prior best-known bound in May 2026, and outside mathematicians reviewed and backed it. Treat that as outside-reviewed progress, and keep watching the literature for the full write-up.

Did AI solve Navier-Stokes?

OpenAI announced a singularity proof on Sept 8, 2026. It was not peer-reviewed or Clay-confirmed when last checked, and there is an active credit dispute. Do not treat it as settled.

Can I try the geometry myself?

Yes. Plot the four points in Graph Lab or check pairwise distances with the Distance Between Points calculator. Adding a fifth point that beats five unit pairs is a real challenge.

What is a Reynolds number?

A dimensionless ratio of inertial to viscous forces: (density × speed × length) ÷ viscosity. In pipes, flow is usually laminar below about 2,300 and turbulent above about 4,000.