Is the Riemann hypothesis solved? Not confirmed. On 20 November 2026, Anthropic mathematician Levent Alpöge claimed a proof made with Claude, covering both the original hypothesis and its generalized form. He released it only as Lean code, a format a computer can check line by line, and experts started checking it within hours. Until that check finishes and mathematicians accept the argument, the most famous open problem in mathematics stays officially open.
This guide explains what Alpöge claimed and what Claude did, and where the verification stands as of 18:00 UTC on 20 November. We also downloaded the full proof archive and opened it ourselves, so you can see what is actually in the code that most coverage only describes secondhand. It also covers the questions people are asking: what the Riemann hypothesis is, whether the proof wins the $1 million prize, and what would change if it holds.
The Key Takeaways
- The claim: Levent Alpöge wrote that "working with Claude" he found a proof of the Riemann hypothesis and the Generalized Riemann Hypothesis. He posted it on 20 November 2026.
- The status: claimed, not confirmed. Lean experts are checking the code, and no mathematician has yet confirmed the argument.
- The archive: we counted 175,187 Lean files and 17.4 million lines in the 239 MB download. Its top theorem targets the standard Lean library's own definition of the Riemann hypothesis.
- No readable paper: Alpöge says a written-out version ran past 250,000 pages, so he released only the Lean code.
- The prize: the Clay Institute's $1 million requires publication, two years of waiting and general acceptance, so no award is possible before late 2028.
Riemann Hypothesis Solved? Where the Checks Stand
No, not yet. The Riemann hypothesis is claimed solved, and the claim is serious, but none of the checks that turn a claim into an accepted proof has finished. On the Fello AI proof ladder, our four-step scale for AI math results, the Alpöge proof sits on rung 1, "claimed". The table shows each check and where it stood at 18:00 UTC on 20 November 2026.
| Check | What it rules out | Status, 20 Nov 2026, 18:00 UTC |
|---|---|---|
| Proof published | Nothing yet | Done: four-page note plus Lean archive |
| Lean statement matches the hypothesis | Proving the wrong theorem | Early yes: Elliot Glazer of Principia Labs says the formal statement "is in fact GRH" |
| Lean proof compiles | Logical gaps in the formal proof | In progress on the Lean community's Zulip forum |
| Mathematicians understand the argument | Misreading what was proved | Not started: there is no readable paper |
| Clay Prize conditions | Almost everything | Not possible before late 2028 |
The Lean check is the fastest of these, because a computer does the checking.
Mathematicians understanding why the proof works will take far longer.
What Levent Alpöge Claimed
Levent Alpöge claimed that he and Claude proved the Riemann hypothesis and its generalized form, and that both are fully formalized in Lean. "I believe I have, working with Claude, discovered a proof of the Riemann Hypothesis," he wrote in a four-page note on his website. He added that "Claude and I have also formalized the arguments without assumptions" and that he expects the method to reach the Extended Riemann Hypothesis "with more time".
Alpöge is a number theorist, not an outsider with a theory. He works at Anthropic, is a junior fellow at Harvard's Society of Fellows and wrote his PhD thesis at Princeton. In July he posted a Claude-made counterexample to the Jacobian conjecture.
What the Note Says Claude Did
The note gives Claude a large share of the credit. Alpöge writes that he chose the strategy and "guided us to the right situation", but that "this time Claude far outpaced me near the finish line". He singles out one step: Claude suggested how long a chain of mathematical constructions called theta lifts should be. Alpöge also says he has "little mastery" of the higher-rank automorphic forms the argument relies on, the part Claude carried.
The note does not say which Claude model was used. We found no statement about the claim from Anthropic itself as of 18:00 UTC on 20 November. The public Claude models, and which plans include them, are listed in our Claude pricing guide.
The 41-Day Timeline
The note ends with a 20-step timeline that runs from 6 October to 16 November 2026, whose last entry reads simply "GRH". Eight of the 20 steps are dated 15 November. Alpöge says Claude compiled the timeline "after a thorough search", and he warns that some descriptions are paraphrased from an AI summary "which might make zero sense".
Why the Timing Raised Questions
The claim landed one day after Anthropic's shares began trading on the Nasdaq on 19 November, covered in our Anthropic IPO guide. Alpöge addresses this directly. He says the result came together "late Sunday night" and that he asked Dario Amodei about announcing it. "There is an obvious cynical view, to do with the timing, but it is false," he writes. The proof's validity does not depend on the timing either way. Only the checks in the status table can settle that.
What We Found Inside the Lean Archive
The Alpöge proof is enormous, and its headline theorem targets the standard definition of the Riemann hypothesis. We ran the Fello AI archive check on 20 November 2026 at 17:28 UTC. We downloaded the 14 archive parts from Alpöge's site, joined them and read the files without running any code. The figures below are our own counts.
| What we measured | Result |
|---|---|
| Archive size | 239,275,554 bytes in 14 parts |
| Lean source files | 175,187 |
| Lines of Lean code | 17,374,215 |
| Lean version | Lean 4 v4.33.1, with the Mathlib library pinned to one fixed version |
| Main theorem file | 12 lines, proving five statements |
| Shortcuts found by text search | 0 sorry, 0 admit, 0 axiom declarations, 0 native_decide |
The Theorem the Code Claims to Prove
The first theorem in the archive's main file proves RiemannHypothesis, a definition that lives in Mathlib, Lean's shared math library, and not in Alpöge's code. That matters, because the most common way a formal proof goes wrong is proving a statement that only looks like the real problem. Mathlib has carried that definition since at least 2024, and its own comment says that "constructing a term of this type is worth a million dollars". The second restates the Riemann hypothesis in the archive's own terms, and the other three cover Dirichlet L-functions, the functions the Generalized Riemann Hypothesis is about.
Here is the archive's entire main file, copied from our download. Each line names a statement and the proof that settles it.
import Statements
import Supplements.Kernels.MainTheoremsAll
theorem riemannHypothesis : RiemannHypothesis := rhOfAll
theorem riemannZeta_eq_zero_re : RiemannZetaZerosOnLine := rhOfAll
theorem LFunction_ne_zero_of_half_lt_re : DirichletLNonvanishingRightOfHalf := allChi
theorem LFunction_eq_zero_re : DirichletLZerosOnLine := claimAll.2.1
theorem LFunction_eq_zero_re_of_isPrimitive : PrimitiveDirichletLZerosOnLine := critAll
What Our Check Does Not Prove
The Fello AI archive check rules out the obvious shortcuts, not hidden errors. Lean marks an unproven step with words like sorry or admit, and an unproven assumption with axiom. We found none in any of the 175,187 files. We did not build the project. Only a full build, plus a report of which axioms the final theorem depends on, confirms the formal proof. That is the check now running on the Lean community's Zulip forum.
How Mathematicians Are Checking the Riemann Hypothesis Proof
Mathematicians are checking the Riemann hypothesis proof in two stages: the computer check of the Lean code first, then a human reading of the mathematics. The news spread fast. Mathematician Harry Stoltz was among the first to flag the note, at 13:49 UTC on 20 November.
Well, looks like @__alpoge__ & @claudeai are claiming a proof of the Riemann Hypothesis
— Harry Stoltz (@HarryStoltz1) November 20, 2026
https://alpo.ge/remark.pdf
The Computer Check
The first question for any Lean proof is whether the formal statement says what the claim says. Elliot Glazer of Principia Labs answered it the same afternoon: the statement is the Generalized Riemann Hypothesis. He also put the chance that the result rests on a bug in Lean itself at well under 1%.
1) The formalized statement in Levent's drop is in fact GRH.
— Elliot Glazer (@ElliotGlazer) November 20, 2026
2) Still compiling, will take a few hours, but c'mon, it's gonna compile, obv that was checked ahead of time
3) Technically it could be a Lean bug but frontier models virtually never do that on accident. <<1% it's a bug
Glazer's point about a compile is about the code, not the mathematics. A compiled Lean proof means the formal statement is true. It says nothing about whether any person understands why.
The Human Check
The human check has barely started, because there is nothing for a human to read. Alpöge writes that it was "obviously inappropriate to attempt to post some massive unreadable travesty of 'slop'", and a footnote says his written-out version "hit 250,000+ pages". Thomas Bloom, a Royal Society research fellow at the University of Manchester, summed up the reaction.
I don't know whether I'm more surprised that (a) AI claims to have proved the Riemann Hypothesis or (b) that it claims to have produced a coherent proof that, written out in human-readable format, is 250,000+ pages long.
— Thomas Bloom (@thomasfbloom) November 20, 2026
Alpöge expects the argument to shrink. "My intuition insists it will be massively simplified, likely immediately," he writes, and he says he plans to write it up properly. Until a version exists that mathematicians can follow, even a clean compile leaves the proof on rung 2 of the ladder, machine-checked but not understood.
What Is the Riemann Hypothesis, and Why Does a Proof Matter?
The Riemann hypothesis says that every non-trivial zero of the Riemann zeta function lies on one vertical line, where the real part equals exactly 1/2. Bernhard Riemann proposed it in 1859. It matters because the zeros control how evenly the prime numbers are spread out, and many results in number theory are proved only on the assumption that it is true. The Clay Mathematics Institute says it has been checked for the first 10,000,000,000,000 zeros, but no number of examples can prove it for all of them.
The Generalized Riemann Hypothesis makes the same claim for a whole family of related functions, the Dirichlet L-functions. Those govern how primes split between different remainder classes, for example primes ending in 1, 3, 7 or 9. Proving the generalized form would settle every result that currently assumes it.
What a Proof Would Change
A confirmed proof would change mathematics far more than daily life. "If the proof is correct and understood, this completes a dictionary between different branches of math where the connections were only partial," mathematician Hector Pasten told Scientific American. "Something like a Rosetta Stone." The same article notes that the result "won't upend global banking or make the internet run faster". Its effect would be inside mathematics, where results now proved only by assuming the hypothesis would stand on their own.
Earlier Claims That Failed
Claimed proofs of the Riemann hypothesis have failed before. In September 2018, Fields medallist Michael Atiyah presented what he called a "simple proof" at the Heidelberg Laureate Forum, and experts quickly doubted it. The Alpöge claim differs in one important way: it arrived as a formal proof a computer can check, not as a talk or a paper. That makes it much faster to confirm or refute.
Riemann Hypothesis Solved by Claude? August vs November
Claude has two separate Riemann results in 2026, and search results often mix them up. The August result is a checked step toward the hypothesis. The November result is a claimed proof of the whole thing.
| August 2026 result | November 2026 claim | |
|---|---|---|
| What it says | At least 67.2% of zeta zeros lie on the critical line, up from 41.6% | All non-trivial zeros lie on the line, and the generalized form holds |
| Who | An unreleased research version of Claude, prompted by Anthropic's Jarred Sumner | Levent Alpöge with Claude |
| Checked by | Anthropic mathematicians Levent Alpöge and Ralph Furman, plus a Lean formalization | Lean check in progress, no human review yet |
| Proves the hypothesis? | No | Claims to |
The August bound came from an unreasonable request. According to Anthropic's research post, Sumner asked Claude to "take a real stab at the Riemann hypothesis", and the model found the improved bound along the way. Alpöge was one of the two mathematicians who checked it. For OpenAI's separate October claim, the quasi-Riemann hypothesis, see our record of every AI math claim in 2026.
Does the Riemann Hypothesis Proof Win the $1 Million Prize?
Not yet, and not for at least two years. The Clay Institute's rules set three conditions before it will even consider a solution. The proof must be published in a qualifying outlet, at least two years must pass after publication, and it must have "received general acceptance in the global mathematics community". Clay does not accept direct submissions.
A Lean archive on a personal website meets none of those conditions yet. If Alpöge publishes a paper in 2027, the earliest Clay could act is two years after that.
What Happens Next
The next verdict will come from the Lean check, which is already running. If the archive compiles with only the standard axioms, the claim moves to rung 2, and the Riemann hypothesis would have its first machine-checked proof. If the build fails or the axiom report shows a hidden assumption, the claim collapses just as fast. After that comes the slow part: mathematicians turning 17 million lines of code into an argument they can follow.
Until then, treat any headline saying "AI solved the Riemann hypothesis" as a claim with a computer check running.
To understand the argument while you wait, ask two models to explain the note and compare their answers before you trust either. Fello AI puts Claude, GPT and Gemini in one app, so a second opinion is one click away instead of another subscription. Our best AI for math guide compares the public models on hard math. Our AGI prediction tracker follows how fast they are improving.
FAQ
Has the Riemann hypothesis been solved in 2026?
Not confirmed. On 20 November 2026, Anthropic mathematician Levent Alpöge claimed a proof made with Claude, released only as Lean code. Experts are checking whether it compiles and states the right theorem, and no mathematician has confirmed the argument yet.
Did Claude solve the Riemann hypothesis?
Alpöge says Claude did much of the work, writing that it "far outpaced me near the finish line". He chose the strategy, and Claude supplied key steps in areas he says he has not mastered. The proof is unverified, and the note does not say which Claude model was used.
Why is there no readable paper?
Alpöge says the fully written-out argument ran past 250,000 pages, which no one could read. He released the Lean formalization instead, 175,187 files by our count, and says he will write a proper, shorter version.
What is the Generalized Riemann Hypothesis?
The Generalized Riemann Hypothesis extends the original claim to Dirichlet L-functions, which describe how primes are distributed across remainder classes. The Alpöge archive states three theorems about these functions alongside two versions of the original hypothesis.
Would a proof of the Riemann hypothesis break encryption?
No, not directly. Scientific American put it plainly: the result "won't upend global banking or make the internet run faster". Its main effect would be inside mathematics, on the many results that are currently proved only by assuming the hypothesis.
L’image principale de cet article a été créée avec l’aide de l’IA.