Model

Lean and Mathlib certify Claude-assisted proof of the Riemann Hypothesis

A day after Anthropic mathematician Levent Alpöge released a Claude-built proof, the organizations behind the Lean proof assistant certified it correct. The claim, and the checking, land on Anthropic's IPO week, with shares untraded since Friday.

Richard TangRichard Tangtwo-minute read
Share
Levent Alpöge, in a green T-shirt, standing outdoors at the Oberwolfach research institute in front of a wire-globe sculpture
Levent Alpöge, the number theorist on Anthropic's technical staff, at the Oberwolfach mathematics research institute in 2024, for the conference "Explicit Methods in Number Theory".

Elon Musk set the tone on X: "Unequivocal super intelligence has arrived. Welcome to the Singularity." 1 The verified substance is narrower, and more useful. On 20 November, Levent Alpöge, a number theorist on Anthropic's$350B — Anthropic, private, latest valuation $350B technical staff, announced a proof of the Riemann Hypothesis, the 1859 conjecture about hidden order in the prime numbers, built with Anthropic's model Claude and released as roughly 17 million lines of Lean, a language whose kernel checks a mathematical argument step by step 23.

The most-viewed recap on X counts 175,187 files 4; the certified statement confirms the line count, not the file count. On 21 November the Lean Focused Research Organization, which develops Lean, and the Mathlib Initiative, which maintains its shared mathematics library, certified the result: "We have full confidence that Alpöge's proof of the Riemann Hypothesis is correct." 5

What makes the certificate hold

The statement says dozens of people recompiled the library and reproduced the same proof export; that the theorem proven is Mathlib's own expert-vetted definition of the hypothesis, not a lookalike; and that checkers independent of Lean, two carrying formal guarantees that they cannot accept a false proof, validated the argument on an inert export, beyond the reach of any code in the library 6. Andrew Sutherland, an MIT mathematician, was among the first to build it and maintains an overview of the verification 67. Anthropic itself declined to comment 7.

Correct is not understood

No human has read the proof. Alpöge puts a readable version past 250,000 pages and says he has not mastered all of it 26. Kevin Buzzard of Imperial College London calls it "millions of lines of gobbledygook" that mathematicians are translating back into ideas, work he expects to take weeks 3. The Clay Mathematics Institute still lists the problem unsolved; its Millennium Prize requires journal publication and two years of acceptance 26. The certificate also covers the hypothesis itself, not the stronger generalized version Alpöge claims 6.

Whose money it moves

Anthropic gave back most of its debut pop in the AI selloff

Friday's close came before the certification; Monday's open is the first test

$0$50$100$150IPO priceFirst tradeDay 1 closeDay 2 closeProof certified next day, Saturday
Data
Share price
IPO price$83
First trade$111.5
Day 1 close$97.17
Day 2 close$88.91
Nasdaq: ANT offering and trading prices, 19–20 November 2026. Sources: TechRepublic; Yahoo Finance.8,9

ANT, Anthropic's Nasdaq ticker, slid to $88.91 on Friday, near its $83 offer price, as the AI share of US market cap fell to 44.0% from its 2 October peak of 49.3% 8910. The certification landed on a Saturday; Monday's open is the first session able to price it. No checker can yet say how much of the mathematics was Claude's, or whether it repeats on the next problem 7. The proof belongs to Alpöge and Claude. The checking that made it usable is the part every future AI capability claim, and everyone sizing one, will be measured against.

Get Model by email, free

The AI labs, their models, and the business behind them. In our daily email, every weekday.

Reader comments

0 comments

    Model

    Respan served OpenAI's GPT-6.2 Sol at $0 for 24 hours

    The plumbing for the giveaway was finished two weeks before OpenAI shipped the model. The money at risk is a $5 million seed betting free tokens turn into paid observability accounts.

    Model

    Google opens Gemini 4 Neon to all developers at $3 per million input tokens

    Google's fourth Gemini 4 model in five weeks pairs top-three benchmark scores with mid-tier pricing, resetting the margin math for AI software vendors in the very week Anthropic's record IPO slid back toward its $83 offer price.

    Model

    House report says the model behind SpaceX's $60 billion Cursor deal was built on China's Kimi K2.5

    SpaceX paid about $60 billion in stock for Cursor in August, and a House report now says its flagship model rests on Chinese open weights. The same day, rival Cognition entered talks at $75 billion, two months after Cursor's own backers led its $48 billion round.

    Model

    Mira Murati's Thinking Machines ships a 26-billion-parameter model that runs offline on a MacBook

    The 26-billion-parameter open-weights agent shipped hours after OpenAI said attackers had extracted GPT-6 Astra's hidden reasoning at a 54% success rate and asked Congress for new enforcement powers. Together the two moves frame the scarcity bet behind frontier token prices and the market's newest trillion-dollar listing.

    Model

    OpenAI says attackers copied GPT-6 Astra's hidden reasoning on 54% of tries; its fix cuts that to 0.2%

    OpenAI's November 20 disclosure puts a hard number on how copyable a frontier model was, and pairs the fix with a call for federal enforcement. The labs it names, Moonshot AI and DeepSeek among them, are where the money moves.

    Model

    Amazon's $188 Billion Anthropic Stake Swings $1.9 Billion for Every $1 of Anthropic Stock

    Two trading days turned Amazon's largest AI holding into a daily mark-to-market item, and Anthropic's second-session slide shows how fast the gain Amazon already booked can reverse.

    Subscribe to TheInference for free