Anthropic mathematician Levent Alpöge said Friday that he and the company’s AI model Claude have proven the Riemann hypothesis, the 167-year-old conjecture that carries a $1 million Millennium Prize.
Alpöge, 34, released a four-page note and 14 compressed Lean 4 archives totaling 239 megabytes and 175,187 source files — a formalization he says exceeds 250,000 printed pages. The claim also covers the Generalized Riemann Hypothesis, a vastly broader statement that underpins thousands of results in number theory.
Independent verification of the Lean code is underway. A clean compile confirms only that the logical steps follow Lean’s axioms, not that the formal statement matches the mathematical claim.
Second Millennium Claim Same Day
Hours earlier, Alpöge and New York University mathematician Tristan Buckmaster posted a 79-page paper asserting finite-time blow-up for the unforced three-dimensional Navier–Stokes equations — the second Millennium Prize problem the pair has targeted in 2026.
In August they verified a related result for the 3D Euler equations using both Claude and OpenAI’s Codex. When OpenAI announced its own Navier–Stokes solution on September 8, Buckmaster alleged OpenAI researchers may have accessed his private Codex session, sparking a priority dispute that remains unresolved.
Lean experts are currently investigating the validity of the claimed formalised proof of the Riemann Hypothesis – follow along at https://t.co/GwuQHMlxvx
“This is indeed an historic moment,” MIT mathematician Andrew Sutherland told me.
Proof Attempts Stepanov-Inspired Route
Alpöge’s note describes a route inspired by Stepanov’s proof of the Riemann hypothesis over function fields. The decisive step, he writes, was a long chain of theta lifts whose length Claude suggested setting at roughly the square root of the primes involved.
If this proves to be true, it will be a monumental breakthrough. An ode to mathematics, to AI, and above all, to human ingenuity. After all, there is no such thing as AI born out of AI only.https://t.co/3NJmjnIJcD
Alpöge acknowledges he has not mastered every part of the argument and that no human-readable paper exists yet; the Lean formalization is the only version released for scrutiny.
Verification Gap Remains
The Clay Mathematics Institute requires any Millennium Prize solution to be published in a qualifying peer-reviewed journal and survive two years of community scrutiny. Lean can check each logical step, but it cannot explain why the result is true. A MathOverflow discussion in October highlighted a recurring failure mode: a model can formalize a headline theorem while the natural-language proof contains steps that do not survive close reading.
For 167 years, the Riemann hypothesis eluded mathematicians. One day after the company’s IPO, an Anthropic employee says Claude helped discover a proof. https://t.co/qvhaizPoMh via @WSJ
In September, 25 Fields medallists signed a declaration warning that rushed announcements leave no time for proper write-ups, isolation of new ideas, or citation of earlier work — arguing that solving problems is a proxy for the real goal, which is understanding.
The announcement follows Anthropic’s record IPO on Wednesday. The company priced at $83 per share, upsized the offering to about 980 million shares, and raised $81.3 billion — the largest IPO ever — giving it a fully diluted market capitalization near $1.5 trillion under the ticker ANT. Alpöge said he consulted CEO Dario Amodei before going public and acknowledged the timing could be viewed cynically.
Mathematicians Weigh Impact
“If true, even from a first high level scan of the summary of the proposed argument, it is very plausible to me that the new ideas in the proof will open up brand new avenues of thinking in number theory and have an immense impact beside the obviously extremely influential result in itself,” said Carlo Pagano, mathematician Carlo Pagano, a professor at Concordia University and a visiting researcher at Google DeepMind.
“If the proof is correct and understood, this completes a dictionary between different branches of math where the connections were only partial. Something like a Rosetta Stone,” said Hector Pasten, a mathematician at the Pontifical Catholic University of Chile.
Holy cow. Anthropic is claiming to have solved the Riemann Hypothesis!!! If true, huge.#RiemannHypothesis https://t.co/80OxaFhmP6https://t.co/X7XMnFp7OK
On X, NYU and Sorbonne mathematician Scott Armstrong summarized the mood: “In lesser news, Tristan and Levent claim to have proved blow-up of UNFORCED Navier-Stokes.” His post drew roughly 33,000 views within hours.
Levent Alpöge, born April 1, 1992, on Long Island, New York, earned a BA in mathematics and an MA in physics from Harvard in 2014, a Master of Advanced Studies from Cambridge as a Churchill Scholar in 2015, and a PhD from Princeton under Manjul Bhargava in 2020. He held an NSF postdoctoral fellowship and was a Junior Fellow of the Harvard Society of Fellows from 2021 to 2025 before joining Anthropic.
In lesser news, Tristan and Levent claim to have proved blow-up of UNFORCED Navier-Stokes
In 2022, with Bhargava and Ari Shnidman, he proved the first bounds on the density of integers expressible as sums of two rational cubes. In 2025, with Bhargava, Wei Ho, and Shnidman, he gave a second proof that Hilbert’s tenth problem has a negative answer over the ring of integers of every algebraic number field.
The first test is mechanical: does the Lean formalization compile cleanly against the standard math library, and does the formal statement correctly encode the Riemann and Generalized Riemann hypotheses? A group of volunteers has begun that check.
If it passes, the community must still decide whether the formal proof constitutes a human-understandable argument — and whether the ideas inside it can be extracted, written up, and submitted to a qualifying journal. Under Clay rules, the two-year clock starts only after publication.