Anthropic · Released, not on our provider
AI Riemann Hypothesis (2026): what Anthropic’s mathematician claims and where it stands
AI Riemann hypothesis refers to Anthropic mathematician Levent Alpöge’s claim, made on November 20, 2026, that he and Claude proved the Riemann hypothesis, released as 175,187 Lean files that MIT’s Andrew Sutherland reports compile and pass independent kernel checks, while mathematicians say the proof is not yet confirmed or understood.
Use it now
Chat
This chat starts with Claude Opus 5.5, not AI Riemann Hypothesis.
≈ 6 credits per reply · See all models
New accounts get 5 free credits once, valid for 7 days. They work on the lower-priced models here; the priciest ones need credits you bought.
Start chatting — 5 free credits- Maker
- Anthropic
- Type
- Text
- Released
- No date given
- OpenRouter id
- None
Checked
AI Riemann Hypothesis: facts, with sources
- Levent Alpöge, a member of Anthropic’s technical staff, announced on November 20, 2026 in a four-page note that he believes he has, working with Claude, discovered a proof of the Riemann hypothesis, and says the method also resolves the Generalized Riemann Hypothesis.Source: alpo.ge · checked 2026-11-22
- The note links a 14-part Lean archive instead of a paper. Alpöge writes that he chose to present only Lean rather than post a massive unreadable text, and that the argument is incredibly long at the moment.Source: alpo.ge · checked 2026-11-22
- In the note, Alpöge says he guided the work to a setting analogous to a simpler “toy” case and that near the end Claude “far outpaced” him; he adds that he may have built strange suboptimal choices into the argument and expects shorter arguments to follow. The note does not name a Claude model version.Source: alpo.ge · checked 2026-11-22
- The statement checked is Mathlib’s own definition of the Riemann hypothesis: every zero of the zeta function other than −2, −4, −6 and so on has real part 1/2. Sutherland’s page says con-leche’s compare mode accepts all five theorems against the challenge file, which imports only Mathlib.Source: math.mit.edu · checked 2026-11-22 · second-hand
- Nature quotes Oxford’s James Maynard, a Fields medallist: “I suspect that the proof will turn out to be correct”, but also that it is crazily complicated and will take mathematicians a long time to digest. Andrew Granville says the released document is garbled and that Alpöge should have prepared a clearer write-up first.Source: nature.com · checked 2026-11-22 · second-hand
- AP reports that hours after the announcement the Riemann hypothesis was still listed as unsolved on the Clay Mathematics Institute’s website, and quotes its president, Martin Bridson, saying the journey from the announcement to real understanding will be long and difficult.Source: bostonglobe.com · checked 2026-11-22 · second-hand
- Nature reports that hours after Alpöge’s note, Stanford’s Ravi Vakil posted a separate, entirely human-written preprint by Alessandro Maria Masullo that also claims a proof of the Riemann hypothesis.Source: nature.com · checked 2026-11-22 · second-hand
- Anthropic’s own news page, read on November 22, 2026, lists no post about Alpöge’s claim; the company’s IPO posts are there. Nature and AP both said Anthropic had not responded to their requests for comment.Source: anthropic.com · checked 2026-11-22
- Nature says the announcement came days after Anthropic’s IPO, and that Granville points to the company’s drop in valuation on November 19 as the reason for the timing; Alpöge’s note says the “obvious cynical view” of the timing “is false”. Anthropic did not respond immediately to Nature’s request for comment.Source: nature.com · checked 2026-11-22 · second-hand
Claim and checks
AI Riemann Hypothesis: the Anthropic mathematician’s Riemann claim and what has been checked
- Andrew Sutherland of MIT publishes check results on a verification page: all 175,187 Lean modules compile, the five headline theorems depend only on Lean’s three standard axioms, every module replays through the Lean kernel, the independent nanoda and con-leche kernels accept the export, and a sandboxed rebuild on a separate machine reproduces the export byte for byte. When we read it again on November 22, 2026 (page stamp 04:35 UTC) it said some checks (the lake comparator and the Lean default and paranoid kernel runs) were still running.Source: math.mit.edu · checked 2026-11-22 · second-hand
Where it stands
AI Riemann Hypothesis: the claim, the Lean checks and the human review, as reported
This is a claim, and the table keeps three things apart: what Alpöge says, what machines have checked, and what mathematicians have not yet done. It is not a statement that the Riemann hypothesis is proved. The same news is searched in Italian as “ipotesi di Riemann Anthropic” and in German as “Riemannsche Vermutung Anthropic” or “Riemann-Hypothese”. For the company behind the AI, see our Claude page; for the IPO the news stories mention, see Anthropic Aktien.
| Question | What is reported | Who said it, and when |
|---|---|---|
| Who claims what? | Levent Alpöge of Anthropic says he and Claude found a proof of the Riemann hypothesis, and that the method also resolves the Generalized Riemann Hypothesis. | Alpöge’s four-page note, November 20, 2026 |
| In what form? | Lean code only: a 14-part archive of 175,187 Lean files, no conventional paper. Alpöge calls the argument incredibly long. | Alpöge’s note; file count from Sutherland’s page, November 2026 |
| What have machines checked? | All 175,187 modules compile, the five headline theorems use only Lean’s three standard axioms, and independent kernels (nanoda, con-leche) accept the export, and a sandboxed rebuild reproduces it byte for byte. Some runs were still going when we read the page. | Andrew Sutherland (MIT), verification page read November 22, 2026 |
| What do mathematicians say? | James Maynard suspects it will turn out correct; Andrew Granville calls the released document garbled; both say it will take a long time to digest. | Nature, November 21, 2026 |
| Has the Clay Institute accepted it? | Not as of the AP report: the hypothesis was still listed as unsolved hours after the announcement, and the institute’s president said the journey to understanding will be long and difficult. | AP, November 21, 2026 |
| Is there another claimed proof? | Yes: a separate, human-written preprint by Alessandro Maria Masullo, posted by Ravi Vakil on the same day. Nature does not say it has been checked. | Nature, November 21, 2026 |
Source: alpo.geSource: math.mit.eduSource: nature.comSource: bostonglobe.com
Nearest available
Alternatives to AI Riemann Hypothesis you can use now
Claude Opus 5.5
Anthropic$4 / $20 (per million tokens)
- Released on September 22, 2026 (Anthropic’s model page marks it "Latest"); retirement not sooner than September 22, 2027.Source: platform.claude.com · checked 2026-11-15
- Anthropic describes it as "for long-running agentic coding and knowledge work": multihour autonomous coding agents, large-scale refactoring, complex systems engineering, vision-heavy workflows and computer use.Source: platform.claude.com · checked 2026-11-15
- Context window 1M tokens; maximum output 128K tokens; text and image input, text output. Adaptive thinking is always on (default effort "medium"); reliable knowledge cutoff June 2026.Source: platform.claude.com · checked 2026-11-15
Claude Sonnet 5.5
Anthropic$2 / $10 (per million tokens)
- Anthropic’s model page lists Claude Sonnet 5.5 as released on September 28, 2026, status "Active (latest)", with retirement not sooner than September 28, 2027.Source: platform.claude.com · checked 2026-11-17
- Anthropic’s model page says Sonnet 5.5 offers "the best combination of speed and intelligence", built for everyday coding, agent and enterprise workloads. The announcement calls it the second model in the Claude 5.5 family and a faster, lower-cost complement to Claude Opus 5.5, strongest at well-scoped everyday tasks, bug fixing and polished documents, slides and spreadsheets, and says Opus 5.5 remains stronger at complex, open-ended work.Source: anthropic.com · checked 2026-11-17
- Context window 1M tokens, maximum output 128K tokens (up to 300K on the Message Batches API with a beta header), text and image input, text output, adaptive thinking with default effort high on the API, and a reliable knowledge cutoff of June 2026. Setting temperature, top_p or top_k returns a 400 error, and the API does not accept turning thinking fully off (the lowest setting is between_tools).Source: platform.claude.com · checked 2026-11-17
FAQ
AI Riemann Hypothesis FAQ
Did Anthropic’s AI prove the Riemann hypothesis?
Not established. Levent Alpöge, the Anthropic mathematician in the Riemann news, claims he and Claude did, and his Lean code passes the machine checks reported so far, but the Clay Mathematics Institute still listed the problem as unsolved after the announcement and mathematicians say it will take a long time to understand. Treat it as a claim until they confirm it.
Who is the Anthropic mathematician behind the Riemann claim, and what did he say about Claude?
He is a mathematician on Anthropic’s technical staff. In his four-page note of November 20, 2026 he says he guided the work to a setting analogous to a simpler case and that Claude “far outpaced” him near the end. The note does not name a model version, so we cannot say which Claude it was. Anthropic’s own news page listed no post about it when we read it on November 22. Italian and German readers search the same story as “ipotesi di Riemann Anthropic” and “Riemannsche Vermutung Anthropic”.
What does “checked in Lean” mean?
Lean is a proof assistant: it accepts a proof only if every step follows from its rules. Andrew Sutherland reports that all 175,187 modules compile and that independent kernels accept the proof of Mathlib’s own statement of the hypothesis. That shows the code is consistent with the stated theorem; it does not explain why the proof works, which mathematicians still have to work out.
Has the Clay Institute accepted it, and is the $1 million prize paid?
No. The hypothesis is one of the seven Millennium Prize Problems, and AP reported that hours after the announcement the Clay Mathematics Institute still listed it as unsolved. Its president said the journey from the announcement to understanding will be long and difficult, and we found no sign that a prize decision has been made.
Why are some mathematicians cautious?
The proof is only code. Nature reports that Andrew Granville called the released document garbled and said Alpöge should have prepared a clearer write-up, and James Maynard, who suspects it is correct, says it is crazily complicated. Alpöge’s note itself says the argument is incredibly long and may contain suboptimal choices.
Is there another claimed proof of the Riemann hypothesis?
Yes. Nature reports that a separate preprint by Alessandro Maria Masullo, written by a human, was posted by Stanford’s Ravi Vakil on the same day. Nature does not say whether either claim has been confirmed, and we have not checked either one.
Can I use Claude on this site?
Yes, though we cannot say which Claude model was used for this work. A typical reply to Claude Opus 5.5 costs about 6 credits (a credit is about $0.01 in the $5 pack), billed by its real token cost, at least 1 credit; Claude Sonnet 5.5 about 3. Do not expect a chat to reproduce a proof: no one has shown that.
Is this Anthropic’s website?
No. This page is unofficial and not affiliated with or endorsed by Anthropic or Levent Alpöge. Every fact links to the page we read.