Riemann hypothesis proof with Claude: Lean certifies it (Nov 2026)
Yes, the computer check passed: on Nov 21 the Lean and Mathlib communities certified Levent Alpöge's Claude-made Riemann hypothesis proof. Human review is next.
Checked and updated 21 November 2026 at 21:20 UTC
Updates
- The Lean and Mathlib communities certified the statement and the proof as correct; added Andrew Sutherland's MIT verification results (175,187 modules, independent kernels nanoda, con-leche, lean4lean).
- Added the rival all-human Stanford draft (Alessandro Masullo, posted by Ravi Vakil), the problems reported in it, the 250,000-page size estimate and first expert reactions (Brian Conrey, Nestor Guillen).

Key points
- Levent Alpöge, a mathematician at Anthropic, posted a note on his website saying that, working with Claude, he believes he has proved the Riemann hypothesis and the Generalized Riemann hypothesis.
- Instead of a paper, he posted the proof as computer code for the Lean proof checker, split into 14 archive files. No readable write-up exists yet; he says one will come later.
- On November 21, 2026, the Lean and Mathlib communities, backed by the Lean FRO and the Mathlib Initiative, certified that both the statement and the proof are correct. MIT's Andrew Sutherland reports all 175,187 Lean modules compile and several independent proof checkers accept it.
The computer check is done and it passed. On November 21, 2026, the Lean and Mathlib communities certified that Levent Alpöge's proof of the Riemann hypothesis, made with Claude, proves the right statement and is valid. Alpöge, a mathematician who works at Anthropic, posted the proof a day earlier as 17 million lines of code for the Lean proof checker, with no conventional paper. What is still missing is the human side: a readable write-up, mathematicians understanding the new ideas, and journal review.
"Riemann hypothesis" jumped into Google's US trending searches on November 20, 2026, and the note topped Hacker News. Here is what was actually posted, why it is hard to check, and what it means if you use Claude.
Lean community certifies the proof (Nov 21)
In a joint statement on the Lean community blog, endorsed by the Lean FRO and the Mathlib Initiative, the communities behind the Lean language and its math library wrote that they have "full confidence" the proof is correct and congratulated Alpöge. Their reasoning follows a four-question test that the Xena Project blog laid out on November 12 for AI-generated Lean proofs:
| Check | What they found |
|---|---|
| Is the statement right? | The target is Mathlib's own, expert-vetted definition of the Riemann hypothesis, not one written by Alpöge or Claude |
| Does the code build? | Dozens of people compiled it with no errors and got the same proof export file |
| Does it prove that exact statement? | Two separate tools (comparator and con-leche) confirmed the proved theorem matches Mathlib's statement |
| Is the proof valid? | Lean's official kernel plus several independent checkers in other languages accepted it, including two that are themselves formally proved unable to accept a false proof |
The checks ran on an inert export of the proof, so hidden malicious code in the source files could not fool them.
MIT mathematician Andrew Sutherland published a detailed results page (checks run with Claude Code). As of 12:55 UTC on November 21 it reported:
- All 175,187 Lean modules compile, and the five main theorems rely only on Lean's three standard axioms.
- Every module was replayed through the Lean kernel with zero rejections, and the independent kernels nanoda (written in Rust) and con-leche accept the full proof.
- The five theorems cover the Riemann hypothesis itself and the Generalized Riemann hypothesis for Dirichlet L-functions.
- A couple of slower runs (Lean's paranoid kernel on the full export, one comparator run) were still going at that time.
New to Lean? Our explainer on what Lean is and how it checks proofs covers it in plain English.
What Alpöge posted
The note, titled "Remark", opens with the line: "I believe I have, working with Claude, discovered a proof of the Riemann Hypothesis." In it he says:
- The method also proves the Generalized Riemann hypothesis, a broader version, and he expects it to reach the Extended Riemann hypothesis with more time.
- He and Claude have formalized the whole argument in Lean "without assumptions". The files sit on his site in 14 archive parts.
- The argument is "incredibly long". Rather than post what he calls an unreadable document, he released only the Lean code and plans a proper write-up later.
- Claude led near the end. He says he steered the approach, then Claude "far outpaced" him near the finish line.
- He addresses the timing, the week of Anthropic's stock market debut (see our Anthropic IPO page). He says the work finished late on a Sunday night, that he discussed the announcement with company leadership first, and that a cynical reading of the timing is false.
According to OfficeChai's reading of the note, the approach borrows a method used for a simpler version of the problem (Stepanov's method over function fields) and replaces its key step with a long chain of "theta lifts". The same day, OfficeChai reports, Alpöge and NYU mathematician Tristan Buckmaster posted a separate claimed result about the Navier-Stokes equations.
The rival "all human" draft from Stanford
A few hours after Alpöge's note, Stanford professor Ravi Vakil, who is also president of the American Mathematical Society, put up what he called a hurried post on his blog. It shared a draft proof of the Generalized Riemann hypothesis by Alessandro Masullo, a postdoctoral researcher in Stanford's math department. Vakil described it as entirely human work, with no AI.
According to The Decoder's report:
- Stanford number theorist Brian Conrad had been reading the draft for months and giving detailed feedback.
- Masullo had planned another year of polishing before going public. The Anthropic news pushed the release forward.
- About eight hours later, Vakil updated his post: several people had reported problems in the draft, some found with the help of AI models. He wrote that every issue must be settled to the community's satisfaction and that he makes no claim the strategy is correct.
So in one day there were two claims: one made with AI and checked by a computer, one written by a human and checked by a human. A day later, the computer-checked one has passed.
| Alpöge (with Claude) | Masullo (Stanford) | |
|---|---|---|
| Posted | Nov 20, his website | Nov 20, Ravi Vakil's blog |
| Covers | Riemann hypothesis and the Generalized version | Generalized Riemann hypothesis |
| Format | Lean code only (over 250,000 pages if printed, per The Decoder) | Written draft, unfinished |
| AI used | Yes, Claude did much of the work | No, per Vakil |
| Status (Nov 21) | Certified by the Lean and Mathlib communities | Problems reported, under revision |
What experts said first
- Brian Conrey, a leading Riemann hypothesis specialist, told AFP it would be "a monumental moment in human history" if it holds. He had only seen the short note, asked for the manuscript, and said he expects simpler proofs to follow.
- Nestor Guillen, a visiting professor at NYU, praised Alpöge for saying openly that he does not yet fully understand every part of the proof.
- Anthropic did not answer AFP's request for comment, so there is still no company statement on the claim.
What the Riemann hypothesis is, in one minute
Prime numbers (2, 3, 5, 7, 11…) look scattered at random, but mathematicians can estimate how many there are below any number. The Riemann hypothesis, from 1859, is a statement about where the zeros of a function called the zeta function sit. If it is true, those prime estimates are as accurate as they can possibly be. Thousands of other results assume it is true. It is one of the Clay Mathematics Institute's seven Millennium Prize Problems, each with a $1 million prize.
What was open on day one, and what is answered now
Lean is a programming language in which every step of a proof is checked by the computer, so a finished Lean proof is very strong evidence. On November 20, Scientific American and OfficeChai listed what still needed checking. By November 21:
| Question | Why it matters | Status (Nov 21) |
|---|---|---|
| Do the files build in Lean? | A proof that fails to compile proves nothing | Yes, dozens of independent builds |
| Does the final theorem state the real Riemann hypothesis? | A small slip in the statement can make a "proof" of something else | Yes, matched to Mathlib's vetted definition |
| Does it rely only on standard axioms? | Extra assumptions hidden in the code would break it | Yes, only Lean's three standard axioms |
What the certification does not do is explain the proof. The Lean statement says the communities look forward to mathematicians working out the new ideas, and Alpöge has said a readable write-up will come later.
How this fits with Claude's earlier math work
This is not Claude's first brush with the problem. In August 2026, Anthropic said a research version of Claude raised a known lower bound on the share of zeta zeros that obey the Riemann hypothesis, from 41.6% to 67.2%. Alpöge and Ralph Furman checked that work, and outside experts Brian Conrey and Dan Goldston examined it. Anthropic said then that it did not expect those techniques to lead to a full proof. This week Anthropic also published two black hole physics results found by Claude agents.
What it means for you
- Nothing changes in the Claude app today. The note does not name a model you can pick, and this was months of work by a specialist, not one chat.
- Read headlines carefully. The accurate version today is "an Anthropic mathematician's proof, made with Claude, has passed full computer verification; human understanding and journal review come next".
- The prize is years away even if the proof holds. Clay's rules need journal publication, two years' wait and general acceptance.
If you are curious what Claude can do for everyday work, our guide on how to prompt Claude Opus 5.5 and our Claude Fable 5.2 overview are better starting points. We will update this page when a written paper appears, when Anthropic comments, or when there is news on the Masullo draft.
AI Tools This Week is not affiliated with Anthropic.
Frequently asked questions
Has the Riemann hypothesis been proved?
The computer-checked proof has now passed. On November 20, 2026, Anthropic mathematician Levent Alpöge posted a proof made with Claude as Lean code, and on November 21 the Lean and Mathlib communities certified that the Lean statement is the real Riemann hypothesis and that the proof is valid. Mathematicians still need to understand the ideas, and no readable paper or journal review exists yet.
Who verified the Riemann hypothesis proof?
The Lean and Mathlib communities, in a joint statement endorsed by the Lean FRO and the Mathlib Initiative, plus MIT mathematician Andrew Sutherland, who published a page of checks run with Claude Code. Dozens of people compiled the code and got the same proof file.
Did Claude prove the Riemann hypothesis?
Alpöge says Claude did much of the work and 'far outpaced' him near the end, while he steered the approach. Anthropic itself has not put out a statement in the reports we read. In August, Anthropic said a research version of Claude only raised a known bound on the problem and was not expected to lead to a full proof.
Will Alpöge win the $1 million Millennium Prize?
Not soon, even if the proof holds. The Clay Mathematics Institute's rules require publication in a qualifying journal, at least two years after publication, and general acceptance by mathematicians.
Sources
- Lean verification of the Riemann Hypothesis (Lean community blog, Nov 21, 2026)
- RH Lean proof: verification files (Andrew Sutherland, MIT)
- ...and the proof is checked in Lean (Xena Project, Nov 12, 2026)
- Hours after Anthropic claims a Claude proof of the Riemann hypothesis, Stanford releases an all-human competing draft (The Decoder)
- Anthropic researcher says he and Claude proved the Riemann hypothesis (AFP via Phys.org, Nov 20, 2026)
- Remark: claimed proof of the Riemann Hypothesis with Claude (Levent Alpöge, PDF)
- Anthropic mathematician claims to prove Riemann hypothesis (Scientific American, Nov 20, 2026)
- Claimed Riemann Hypothesis proof made with Claude appears on Levent Alpöge's website (OfficeChai)
- Claude and the zeros of the Riemann zeta function (Anthropic, Aug 2026)
- Rules for the Millennium Prize Problems (Clay Mathematics Institute)
Related
Claude black hole physics: 2 open problems solved (Nov 2026)
Anthropic says Claude Fable 5.2 agents solved two open black hole problems: the Hawking-Page transition is abrupt, and a new 3D gravity mass limit.
Claude Fable 5.2: what's new, price and who gets it
Anthropic released Claude Fable 5.2 and Mythos 5.2 on November 9, 2026. Here is what changed, what it costs and which one you can actually use.
Gemini 4 Neon vs Claude Opus 5.5: price, limits, which to pick
Gemini 4 Neon costs $3/$15 per 1M tokens, Claude Opus 5.5 $4/$20. Both have 1M-token context. Side-by-side specs, costs and where each one works.