TL;DR
What did OpenAI's AI discover?
On Monday, October 26, 2026, OpenAI announced three results it says its models produced: a proof of the local smoothing conjecture in math, a disproof of the Exponential Time Hypothesis in computer science, and an existence proof for quantum chromodynamics (QCD) in physics. These are OpenAI's claims, and outside experts are only now starting to examine them. The math and computer science proofs were checked by a computer in a proof language called Lean, and OpenAI has published both on GitHub. OpenAI says it is not seeking academic credit, and it is handing the work to groups of academics to explain. There's no product to try, and no direct effect on your business yet.
What happened
OpenAI posted the announcement on Monday, October 26, 2026, and OpenAI researcher Dan Roberts shared it on X at 5:36 PM ET. OpenAI says the results came from its internal models, with outside experts brought in afterward to organize groups that will study and explain them.
OpenAI also set out four principles for these groups: transparency, no academic credit for OpenAI, recognition for the researchers who explain the work, and voluntary, unpaid participation. Members won't sign nondisclosure agreements and can publish their own conclusions.
What OpenAI says its AI found
All three are OpenAI's claims, from its announcement:
- Math: local smoothing. In 1991, mathematician Sogge proposed that waves can't stay extremely focused for long. OpenAI says its models proved this in every dimension. If it holds, it also settles the Kakeya conjecture and Stein's restriction conjecture, two long-open problems.
- Computer science: the Exponential Time Hypothesis is false. The ETH is a long-standing guess that some hard problems can't be solved much faster than by brute force. OpenAI says hundreds of papers have relied on it. Its AI found a disproof on October 16, 2026, and checked it in Lean the next day.
- Physics: QCD exists, mathematically. OpenAI says its models produced a rigorous existence proof for certain quantum field theories, including QCD, the theory of how quarks and gluons stick together. OpenAI says this includes a proof of pure Yang–Mills existence, part of one of the Clay Mathematics Institute's Millennium Prize problems.
Who is checking the results?
Each result has its own plan. For the math proof, OpenAI simply released the AI's writeup and Lean proof, so researchers can organize themselves.
The computer science result goes to a group led by Russell Impagliazzo and Ramamohan Paturi of UC San Diego and Ryan Williams of MIT. Impagliazzo and Paturi first proposed the ETH, with Francis Zane. The group asked for all raw data to be public, and OpenAI says it will fund a conference on what the result means.
The physics result goes to a group led by Nobel laureate David Gross. That group chose to hold back the AI's materials and write the result up themselves first.
Why a computer-checked proof matters
A normal math proof is checked by people. Experts read it line by line, which can take months or years for a long, unusual proof. Mistakes can slip through.
Lean is a proof language that works more like a strict spell-checker for logic. You write each step in a form the computer understands, and Lean refuses to accept the proof unless every step follows from the last. If Lean accepts it, the logic holds.
There's one catch. Lean checks that the proof matches the statement as written, so experts still need to confirm the statement says what everyone thinks it says. They also need to understand why it works, which is the job the new groups have taken on.
The jargon, in one line each
| Term | What it means |
|---|---|
| Conjecture | A statement mathematicians believe is true but haven't proved |
| Hypothesis (in computer science) | A widely assumed claim that other research builds on, like the ETH |
| Disproof | A proof that a claim is false |
| Lean | A programming language for writing proofs that a computer checks step by step |
| Formalization | Rewriting a human-language proof in Lean so a computer can check it |
| Harmonic analysis | The math of waves and signals |
| Complexity theory | The study of how hard problems are for computers to solve |
| Quantum field theory (QFT) | The physics framework for subatomic particles and forces |
| QCD (quantum chromodynamics) | The theory of the strong force that holds quarks and gluons together |
| Peer review | Other experts checking research before the field accepts it |
X reactions
OpenAI's Dan Roberts announced the results on X in the same words OpenAI used on its page.
Our take
If these hold up, this is the strongest sign yet that AI can produce new knowledge. Handing the work to the very people who proposed the ETH, and putting the proofs on GitHub for anyone to check, is the right way to make a claim this big.
Still, three big claims in one announcement is a lot, and the field gets the final word. Lean helps a great deal, but understanding a result takes longer than checking it. We'll treat these as OpenAI's claims until the expert groups report back.
For a café owner or a solo creator, nothing changes this week. A faster algorithm, if one ever comes out of the ETH result, would take years of further research, by OpenAI's own account. Read it as a signal of where these models are heading, and keep judging AI tools by what they do for you today.
— Artificially Nerdy
What to do this week
- If you're curious: read OpenAI's announcement. The section headings explain each result in a few paragraphs. For everyday research help, see our AI research and search tools.
- If you teach or tutor: use the Lean idea as a classroom example of how a computer can check reasoning step by step.
- If you write or post about AI: say "OpenAI says" or "OpenAI claims" until the expert groups weigh in.
- If you run a business: no action needed. File this under "worth knowing" and carry on. For an OpenAI launch you can actually use, read about GPT-6.1 Terra, its new computer-use model.
FAQ
Did OpenAI's AI really disprove the Exponential Time Hypothesis?
OpenAI says so, and says the disproof was checked in the Lean proof language. A group led by Russell Impagliazzo, Ramamohan Paturi and Ryan Williams will verify and explain it, so the field hasn't given its verdict yet.
What is the Exponential Time Hypothesis?
It's a long-standing assumption that certain hard computer problems, like Boolean formula satisfiability, can't be solved much faster than trying every option. OpenAI says hundreds of papers have built on it.
What is the local smoothing conjecture?
A 1991 math conjecture about how waves focus. It says extreme focusing must be short-lived. OpenAI says its models proved it in every dimension, which would also settle the Kakeya conjecture.
Why does it matter that the proof was checked in Lean?
Lean only accepts a proof if every step follows logically, so a Lean-checked proof is much less likely to hide an error. Experts still need to confirm the statement and explain why the proof works.
Is OpenAI taking credit for these discoveries?
No. OpenAI says it is not seeking academic credit, and the outside groups get recognition for the explanations they write.
Will this change how I use ChatGPT?
Not right now. OpenAI announced research results, not a new product or feature.
Sources
- OpenAI: Working with academic communities to build scientific understanding — the October 26, 2026, announcement with all three results, dates, working groups and the no-credit principle.
- OpenAI math repository on GitHub — where OpenAI published the AI writeups and Lean proofs for the math and computer science results.
- The Verge: OpenAI's science announcement — news coverage of the three claims.
- Dan Roberts on X — the October 26, 2026, 5:36 PM ET announcement post.
