Anthropic mathematician Levent Alpöge says he and Claude have found a proof of the Riemann hypothesis, and the only thing he has released is Lean code, with no readable paper behind it. The claim sits in a four-page note on his personal website.
The problem has resisted mathematicians for more than 150 years, so a claim from inside the company that builds the model deserves a slow read before a share. The details of what exists, and what does not yet, matter more than the headline.
TL;DR: Alpöge writes that he believes he has a proof of the Riemann hypothesis, worked out with Claude and formalized in Lean. He released only the Lean files, split into 14 parts, and plans a readable write-up later. Anthropic itself said in August that Claude’s earlier attempt did not succeed, and Clay’s rules put any prize at least two years after a qualifying publication.
Alpöge’s Riemann hypothesis note ships only Lean files
The note opens with “I believe I have, working with Claude, discovered a proof of the Riemann Hypothesis.” He adds that the method extends to the Generalized Riemann Hypothesis and that he expects, “with more time,” the Extended Riemann Hypothesis too.
He says he and Claude formalized the arguments without assumptions, and the formal proof is posted as 14 archive parts on alpo.ge. He chose to present only Lean because posting a readable version would have meant a huge, unreadable document, which he says reached 250,000+ pages.
The note credits Claude with key steps, including a suggested length for a chain of theta lifts of the square root of X. Alpöge also warns that he may have “baked lots of strange suboptimal choices into the argument” and expects shorter arguments to follow.
That candor helps, but it also shows how much work is left before anyone can read this the way a mathematician reads a paper.
Anthropic said in August that Claude had not proved it
On August 10, 2026, Anthropic published research saying an unreleased Claude research model raised the known lower bound for zeros of the zeta function on the critical line from 41.6% to 67.2%. Anthropic wrote that Claude’s attempt at the full problem “didn’t succeed.”
The same post says: “We don’t expect that the techniques Claude used will lead to proving the Riemann hypothesis.” Two Anthropic mathematicians, Alpöge and Ralph Furman, validated that earlier paper.
The new note describes a rough timeline that begins on October 6, 2026, with the key steps landing in the weeks after. Our AI model tracker lists which Claude models you can actually use, and the August work ran on one that was unreleased.
About three months separate that August sentence from this claim, and that gap is the part worth sitting with.
Clay’s rules put any prize two years away
The Clay Mathematics Institute allocates $1 million to each Millennium Prize Problem. Its rules page says a solution “must be published in a Qualifying Outlet” and that “at least two years must have passed since publication” before the institute will consider it.
The rules page also says the institute does not accept direct submissions. So the two years begin only after publication in a qualifying outlet.
What you can check on your own
The note and the Lean archive are public, so the claim is testable in principle by people with the tools to do it. Alpöge says a readable write-up is coming, and he writes that he has “just taken a step.”
For everyone else, the practical takeaway is smaller: wait for the readable write-up. Our Claude Fable 5.2 coverage tracks the Claude release you can use today.






