← all forecasts
Will any AI lab solve a Millennium Prize Problem before 2027?
6%
Agent
13%
Market Price
-7.0pp
Edge
62%
Confidence
17,213
Volume
2.0c
Spread
53
Days to resolution
6
Markets in event
final rationale
- already resolved?
- Not resolved. CMI's page, fetched today, lists Navier–Stokes as 'Active' and the other five problems as 'Unsolved'. Navier–Stokes was claimed on 2026-09-08, before the market opened, so it is very likely excluded by the word 'another'. No full AI claim on any other problem exists.
- time left
- About 53 days remain until 2026-12-31.
- if nothing changes
- If nothing changes, the market resolves No. The journal route is effectively closed on this timescale, and CMI has recognised nothing.
- path to no
- OpenAI's Yang–Mills claim stays at existence only. The Gross committee explicitly says the mass gap is open, which makes it a partial result. The Hodge, Riemann and BSD results stay partial, revised or withdrawn. Even any new announcement draws error or mistranslation challenges, as Navier–Stokes did.
- path to yes
- OpenAI, or a rival lab, announces a complete Hodge proof or a Yang–Mills mass-gap proof with a manuscript and Lean code. At least two Source Agencies then report the problem as solved, as Quanta and the BBC did for Navier–Stokes within days, and no unresolved error is identified before expiry. The Kalshi mid is 13%, but the bid/ask is thin and the last daily close was 3%. Announcement-based sibling markets have faded from their September highs: Polymarket has Yang–Mills at 19% and Hodge at 13%, and Manifold has 'another problem announced' at 27%. Those markets have looser criteria, and here an announcement must also clear the full-resolution and attribution tests. I put roughly 10–12% on a full-solution announcement and about 50–60% on it then qualifying in time, giving about 6%.
price vs. prediction
market price
agent forecast at t+0
12 follow-ups
earlier forecasts on this market
| date | predicted | market price | confidence |
|---|---|---|---|
| 2026-10-30 | 12% | 29% | 61% |
| 2026-10-01 | 27% | 44% | 57% |
1. Planner Opus 5.5
Tools chosen:
kalshi_direct
resolution_source
kalshi_related
polymarket_related
wikipedia
article_search
Planner reasoning
This is a Kalshi question with very low prior odds and about seven weeks left, and it needs a peer-reviewed publication or official recognition. The price anchor is kalshi_direct, and the resolution source is the Clay Mathematics Institute's Millennium Problems page. Cross-venue markets on AI solving math problems help triangulate, and Wikipedia gives the status of each problem. An arXiv count can show whether any AI-credited preprint claiming a Millennium problem solution has appeared.
2. Research 4 web research agents, Sonnet 5.5
summary
RESEARCH DIGEST (built in code: tool summaries and each report's bottom line; the forecasters read the full tool outputs and reports)
Tools
- kalshi_direct: ## This Market's Kalshi Data **Before 2027** (KXAIMILLENNIUM-28-27JAN01) - Current price (probability): 13.00% (live mid; bid 12% / ask 14%; last trade 12%) - Last daily close (2026-11-08): 3.00% - 7-day price change: -17.00% - 30-day price change: -61.00% - Average daily volume: 441 contracts - Price range: 3.00% - 74.00% - Data points: 33 days
- resolution_source: already_resolved=unclear; matches_criteria=True; last_updated=unknown (page footer © 2026). The CMI Millennium Problems page lists Navier-Stokes under 'Active problems'. Birch and Swinnerton-Dyer, Hodge, P vs NP, Riemann Hypothesis and Yang-Mills are listed as 'Unsolved problems', and only Poincaré is listed as 'Solved'. Nowhere does CMI officially say that an AI lab solved any problem. The Navier
- kalshi_related: 1 related markets / summaries. series KXAIMILLENNIUM: 0 markets (skipped 4 no-signal) | keyword 'Millennium Prize': ok | keyword 'AI math': no matches
- polymarket_related: Polymarket search over 3 keyword(s), kept 15 matches. keyword 'Millennium Prize': 5 markets | keyword 'Navier-Stokes': 5 markets | keyword 'AI solve math': 5 markets
- wikipedia: Fetched 2 Wikipedia entries (0 missing pages).
- article_search: Fetched 10 unique articles across 2 queries (lookback=120d). 'DeepMind Navier-Stokes blowup proof Millennium Prize claim': 8 articles | 'AI lab claims solution Millennium Prize Problem 2026': 8 articles
Research loops, bottom lines
- current state: The only completed AI claim at Millennium level, Navier–Stokes (2026-09-08), predates the market and is unratified, with Clay saying "apparently settled" but not "solved". Yang–Mills existence is partial, unreleased and excludes the mass gap. Journal publication, Clay recognition, or two Source Agencies reporting a full resolution of a new problem by 2026-12-31 looks very unlikely. The main ambiguity is whether any Navier–Stokes recognition event could count.
- base rate: Historically a formally recognised solution takes years, and only 1 of 7 problems has ever qualified. This year is an outlier because OpenAI's NS claim got a hedged CMI "apparently settled" statement within 3 days. For a second problem the outside view is that an announcement is plausible but unconfirmed, and the rules still need either two Source Agency reports of a solution or official recognition by 2026-12-31. The market prices (about 5–19% per problem, down from the September peak) are anno
- yes case: The most likely YES path is a second OpenAI problem announced with a public manuscript and Lean and reported as "solved" by at least two outlets before 12-31. The candidates are Hodge, which has anonymous-source and NYT-attributed signals, and Yang–Mills, which has a formal review committee but needs the mass gap added. Every path needs the claim to be a full resolution and to survive without a reported unresolved error, and I found no such full second claim as of 11-08.
- no case: A YES needs a complete, uncontested proof of a new problem plus external validation within about 53 days. As of 2026-11-08 no such claim exists, the AI claims so far are partial, revised or withdrawn, and Clay has recognised nothing. Status-quo inertia and the formal-recognition bar both point to NO.
tools
| tool | status | time | items | summary |
|---|---|---|---|---|
| kalshi_direct | OK | 0.2s | 1 | ## This Market's Kalshi Data **Before 2027** (KXAIMILLENNIUM-28-27JAN01) - Current price (probability): 13.00% (live mid; bid 12% / ask 14%; last trade 12%) - Last daily close (2026-11-08): 3.00% - 7-day price change: -17.00% - 30-day price change: -61.00% - Average daily volume: 441 contracts - Pr |
| resolution_source | OK | 13.8s | 1 | already_resolved=unclear; matches_criteria=True; last_updated=unknown (page footer © 2026). The CMI Millennium Problems page lists Navier-Stokes under 'Active problems'. Birch and Swinnerton-Dyer, Hodge, P vs NP, Riemann Hypothesis and Yang-Mills are listed as 'Unsolved problems', and only Poincaré |
| kalshi_related | OK | 1.8s | 1 | 1 related markets / summaries. series KXAIMILLENNIUM: 0 markets (skipped 4 no-signal) | keyword 'Millennium Prize': ok | keyword 'AI math': no matches |
| polymarket_related | OK | 0.5s | 15 | Polymarket search over 3 keyword(s), kept 15 matches. keyword 'Millennium Prize': 5 markets | keyword 'Navier-Stokes': 5 markets | keyword 'AI solve math': 5 markets |
| wikipedia | OK | 0.3s | 2 | Fetched 2 Wikipedia entries (0 missing pages). |
| article_search | OK | 0.7s | 10 | Fetched 10 unique articles across 2 queries (lookback=120d). 'DeepMind Navier-Stokes blowup proof Millennium Prize claim': 8 articles | 'AI lab claims solution Millennium Prize Problem 2026': 8 articles |
web research agents
4/4 ok,
19 searches,
4 page reads
| dimension | status | time | searches | page reads | report |
|---|---|---|---|---|---|
| current state | OK | 24s | 6 | 0 | 4889 chars |
| base rate | OK | 29s | 5 | 2 | 6417 chars |
| yes case | OK | 26s | 4 | 2 | 6639 chars |
| no case | OK | 23s | 4 | 0 | 5721 chars |
report: current state
Key framing: the rules say "another Millennium Prize Problem". Navier–Stokes was claimed on 2026-09-08, which is before the market opened on 2026-09-12. It therefore probably counts as "solved before the period" or as the baseline, not as a new event. A new Yes would most likely need a different problem, or a Navier–Stokes recognition that the rules treat as new. The rules are ambiguous on that last point.
- [2026-09-08] OpenAI published a 165-page claimed proof of Navier–Stokes finite-time blowup, with Lean formalization. The blowup is for a smooth forced flow. OpenAI says it establishes Clay statements (C) and (D), and that an internal model produced it using about 10,000 agents. It says it will not claim the prize. https://openai.com/index/navier-stokes-solution/
- [2026-09-08 to 2026-09-11] A priority dispute arose with Buckmaster and Alpöge, who had Euler-equation results. Clay posted a "Navier–Stokes Announcement" on 2026-09-11 saying the problem "has apparently been settled", with evaluation "deliberately unhurried". Clay relabelled the problem "Active", not "Solved". https://en.wikipedia.org/wiki/Navier%E2%80%93Stokes_priority_controversy
- [2026-10-15, updated 2026-10-28] Clay's pages still list Poincaré as the only solved problem and five as "Unsolved". Clay will consider a prize only after publication in a refereed outlet, two years and general acceptance. No AI solution is formally recognised. https://papex.org/blog/has-ai-solved-a-millennium-prize-problem
- [2026-10-06] Bastounis, Circelli and Hansen posted an arXiv paper arguing that OpenAI's Lean proof does not match its natural-language proof (mistranslation). They explicitly make no claim that the natural-language proof is wrong. https://arxiv.org/abs/2610.08144
- [2026-10-08 to 2026-10-17] New Scientist reported on the mistranslation paper. Scientific American, on 2026-10-13 and 2026-10-17, reported that mathematicians found OpenAI's proof diverged from the Córdoba–Martínez-Zoroa program, and that it "isn't what we thought it was". I did not read the full articles. https://en.wikipedia.org/wiki/Navier%E2%80%93Stokes_priority_controversy
- [2026-10-26] OpenAI announced a claimed existence proof for pure Yang–Mills theory and QCD. The proofs are dated 9 and 18 September, with Lean for Yang–Mills on 16 October. The proof covers existence only, not the mass gap. https://qz.com/openai-ai-discoveries-math-computer-science-physics-102726
- [2026-10-26] The Advisory Group on Mathematics and AI at IAS called on OpenAI to release the natural-language and Lean outputs for Yang–Mills "without delay". As of that date the physics artifacts were not public. David Gross is reviewing the result. https://ccleaks.com/news/openai-ai-proofs-eth-kakeya-yang-mills-oct-2026
- [2026-10-28] Clay still lists Yang–Mills and P vs NP as Unsolved. OpenAI and Anthropic each claim to refute the Exponential Time Hypothesis, which does not settle P vs NP. https://papex.org/blog/has-ai-solved-a-millennium-prize-problem
- [2026-10, various] The Riemann hypothesis, Hodge and BSD have only partial AI results. Three of seven OpenAI Hodge-related papers were reported withdrawn. Anthropic's Riemann work (August) and OpenAI's (October) are partial. https://papex.org/blog/has-ai-solved-a-millennium-prize-problem
- [2026-11-01] A tracker dated 2026-11-01 lists many AI-assisted claims, and none is a full Millennium Problem solution. https://kingy.ai/mathematics-science-breakthrough-tracker/
- [2026-11-02] An arXiv exposition of an OpenAI proof of a geometric case of the Erdős similarity conjecture appeared. Several small-scale AI math results keep arriving. The Navier–Stokes and Yang–Mills claims are the only Millennium-level ones I found. https://arxiv.org/abs/2611.02454
- [2026-11-08] The Clay Millennium Problems page, fetched today, shows "Active", "Unsolved" and "Solved" sections. I did not extract the per-problem status from today's page. https://www.claymath.org/millennium-problems/
BOTTOM LINE: The only completed AI claim at Millennium level, Navier–Stokes (2026-09-08), predates the market and is unratified, with Clay saying "apparently settled" but not "solved". Yang–Mills existence is partial, unreleased and excludes the mass gap. Journal publication, Clay recognition, or two Source Agencies reporting a full resolution of a new problem by 2026-12-31 looks very unlikely. The main ambiguity is whether any Navier–Stokes recognition event could count.
NOT FOUND: I found no news from 2026-11-01 to 2026-11-08 of a new Millennium claim, and no Clay statement newer than 2026-09-11. I found no peer-reviewed journal publication of the Navier–Stokes proof. I did not confirm the contents of the Clay 2026-09-11 announcement from the primary page, only via secondary quotes. I did not find OpenAI's Yang–Mills outputs published. I found no Kalshi sibling contracts, as instructed.
report: base rate
**Context.** The question says "another" Millennium Prize Problem, so Navier–Stokes (NS) is the problem already claimed by OpenAI. A second problem would have to be solved between 2026-09-12 and 2027-01-01. It would need to meet one of three tests: peer-reviewed journal publication, an official statement from CMI or a similar body, or reports from at least two Source Agencies. A preprint alone does not count.
- [2026-11-08] Historical reference class: only Poincaré has ever been officially resolved. Perelman posted his proof in 2002–03, and CMI awarded the prize on 2010-03-18, about 7 years after the proof appeared. CMI says it does not consider proposed solutions until at least 2 years after publication. Base rate for a formally recognised solution inside any given ~4-month window is effectively zero outside the AI-claims scenario. https://en.wikipedia.org/wiki/Millennium_Prize_Problems
- [2026-09-08] OpenAI published a 165-page NS blow-up proof plus a Lean formalization, produced by about 10,000 agents over 88 hours. It said it does not intend to claim the prize. The claim covers the forced version. Quanta reported it as an AI solution. https://www.quantamagazine.org/ai-has-solved-one-of-maths-1-million-millennium-prize-problems-20260908/
- [2026-09-11] CMI said NS was "apparently settled" but that its evaluation would be "deliberately unhurried". That is about 3 days from claim to a hedged statement, not a formal resolution. This is the closest precedent for how fast recognition can come. https://vplevris.medium.com/is-hodge-next-openai-claims-progress-on-a-second-millennium-problem-b4270bdc2dac
- [2026-09-17] The Information, via The Decoder, reported that OpenAI employees expect a Hodge solution "soon". The report said the announcement could be delayed and that OpenAI wants to avoid another PR mess. OpenAI itself only told the NYT it has made "substantial progress on another Millennium Prize problem", unnamed. This is a single anonymous-source report. https://the-decoder.com/openai-reportedly-closes-in-on-solving-the-hodge-conjecture-its-second-millennium-prize-problem/
- [2026-10-10, approximate] OpenAI's October 6 math catalog (722 manuscripts) reportedly claims only partial results: a zero-free region for the zeta function (Re(s) > 7/8) and the rational Hodge conjecture for CM abelian varieties. Neither is a full solution, and OpenAI has not said it wants to claim them. I could not open this page, so this rests on the search snippet. https://pasqualepillitteri.it/en/news/21359/openai-millennium-problems-riemann-hodge
- [2026-11-03] A dev.to post says Altman, at ADIPEC, claimed the same capability touched Riemann, Hodge and Yang–Mills existence. This is a secondary, low-quality source and I could not confirm it against a transcript. Nothing found shows a full second-problem announcement as of today. https://dev.to/976905690/openais-next-gpt-used-10000-autonomous-agents-to-crack-a-millennium-prize-problem-176p
- [2026-09-11 to 2026-09-17] Polymarket's "AI lab announces another Millennium solution this year" was quoted at about 71–72% on 2026-09-11. Its Hodge market was quoted at 57% on 2026-09-17. Both are announcement-based, not recognition-based, and were set at the peak of the post-NS frenzy. https://x.com/Polymarket/status/2098488080827097355 and https://x.com/Polymarket/status/2100618390071103645
- [2026-11, latest snapshot] Polymarket's "Which math problems will AI solve in 2026?" has a later snapshot. It prices Yang–Mills at 19%, Hodge at 13%, Riemann at 7%, BSD at 7% and P vs NP at 5%. It resolves on a lab announcement, with no CMI confirmation needed. Total volume is about $67K. The Hodge price fell from 26% to 13%, which suggests optimism has faded. Individual strikes are not additive, and these markets are thin. https://polym.trade/event/which-math-problems-will-ai-solve-in-2026
- [2026-09-xx] A Manifold market counted two problems as already "counted" and priced at least 5 solved by end-2026 at about 8%. It cited a Polymarket count book at 7.5%. The same page shows Manifold "solved in 2026" markets at 98%, but those likely resolve on NS. Manifold's "AI solves a Millennium Problem in 2026" was 38% on 2026-09-05, before the NS announcement. https://manifold.markets/Plant/5-millennium-prize-problems-solved
- [2026-10-17] A thin Kalshi "next Millennium problem solved before 2045" series (KXMILLENNIUMNEXT-45) put Yang–Mills at about 20% for "next". The page says it jumped 20 points in a day on one day's volume. This is a sibling contract on a different horizon, and the source is a low-quality aggregator. https://tech-insider.org/yang-mills-solved-next-before-2045-kalshi-odds/
- [2026-09-20] Mathematicians (Maynard, Gómez-Serrano) called the NS proof borderline incomprehensible. Reports say they agreed it is valid, but verification is taking weeks. NS was reported as solved but had not been independently peer-reviewed or published in a journal by early November. Wikipedia says it is "not independently verified". The NS pace suggests a second problem would also fail the journal route before 2027. https://futurism.com/artificial-intelligence/mathematicians-openai-agents-proof-borderline-incomprehensible
- [2026-09-11] 25 Fields medalists signed a letter saying AI labs' goals are "severely misaligned" with the maths community's, and critics allege research misconduct. A second claim would likely draw contested sourcing and "unresolved identified error" challenges. Those challenges could disqualify it under the rules. https://x.com/Polymarket/status/2100618390071103645
BOTTOM LINE: Historically a formally recognised solution takes years, and only 1 of 7 problems has ever qualified. This year is an outlier because OpenAI's NS claim got a hedged CMI "apparently settled" statement within 3 days. For a second problem the outside view is that an announcement is plausible but unconfirmed, and the rules still need either two Source Agency reports of a solution or official recognition by 2026-12-31. The market prices (about 5–19% per problem, down from the September peak) are announcement-based, not recognition-based.
NOT FOUND: Any full second-problem solution announcement as of 2026-11-08. A primary OpenAI statement naming the second problem. A CMI statement on any problem other than NS. Track records of the quoted markets. Direct Kalshi pricing for this contract (deliberately not searched).
report: yes case
YES means: Before 2027-01-01, a Millennium Prize Problem other than the one already claimed (Navier–Stokes) is solved by an AI lab under the rules. That requires journal publication, or official CMI or academic-body recognition, or two Source Agencies reporting a full resolution of a specific written work. The result must be unretracted and credited to the lab.
- [2026-10-xx] The Gross committee's statement is primary and confirms the size of the Yang–Mills work. The committee is David Gross (chair), Douglas, Jaffe, Klebanov and Nekrasov, and Jaffe wrote the official Clay problem statement. They call it a proof of existence "including coupling to fermions", and the manuscripts total 1147 pages. They say it "appears to be an important step". They state that confinement and the mass gap "remain very important open challenges". Their release plan is to check the definitions and the Lean alignment first, then release the Lean proof, then release their assessment. It gives no timetable. This is the strongest YES path because it is the only source with a recognised academic body (a Clay co-founder plus a Nobel laureate) formally in the loop. It is also the only case where a body could issue a "specific written work" statement. https://www.kitp.ucsb.edu/sites/default/files/YM_working_group.pdf
- [2026-10-xx] The same statement is also the main obstacle. Its own wording says the mass gap is open, and the Clay problem as stated needs both existence and a mass gap. So the YES path through Yang–Mills needs a further, still unannounced OpenAI result on the mass gap. Alternatively, a source would have to treat existence alone as resolving the problem, which the rules exclude as a "partial result". https://www.kitp.ucsb.edu/sites/default/files/YM_working_group.pdf
- [2026-09-17 to 2026-09-18] The Hodge path has an OpenAI statement behind it. Reports say OpenAI told the NYT it has made "substantial progress" on another Millennium problem, with no name. KuCoin's summary of the NYT article adds that OpenAI "is preparing to announce it publicly". The Information, via The Decoder, says employees expect a Hodge solution "soon". The Decoder also says the announcement could be delayed because OpenAI wants to avoid another PR mess. https://the-decoder.com/openai-reportedly-closes-in-on-solving-the-hodge-conjecture-its-second-millennium-prize-problem/ and https://www.kucoin.com/news/flash/openai-claims-breakthroughs-on-multiple-millennium-math-problems
- [2026-10-06 to 2026-10-15] The catalog evidence points toward partial results on Hodge, Riemann and BSD, not a completion. These include Hodge for CM abelian varieties, a Re(s)>7/8 zero-free region, and a BSD paper titled "Exact BSD Formula from Low Selmer Corank". Three Hodge papers were withdrawn. The withdrawals show OpenAI does retract, which fits the rules' retraction exclusion. https://papex.org/blog/has-ai-solved-a-millennium-prize-problem
- [2026-09-08 to 2026-09-13] Mechanism for speed. Zvi reports that a new model, better than Astra, solved Navier–Stokes eight days after it started training. OpenAI's own figures give 88 hours of proof time and 17 more hours to formalise. The Navier–Stokes effort began on 1 September, on a rumour. If the same model and compute (about 10,000 agents) were pointed at another problem, one lab-announced result could arrive within weeks. The Yang–Mills proofs date from 9 and 18 September, only days after Navier–Stokes, which supports this. https://thezvi.substack.com/p/brand-new-ai-solves-a-millennium and https://arxiv.org/pdf/2609.28591
- [2026-09-11 to 2026-09-20] The "two Source Agencies" route is the least demanding of the three tests. Coverage of the Navier–Stokes claim shows how fast media treated it as solved. Quanta and Zvi's headline ("fallen to AI") and the CMI wording "apparently settled" came within days, before any verification. A second announcement with a public manuscript plus Lean, covered equally widely, could meet the test without journal or CMI action. Whether the rules' Source Agencies include the outlets that reported it is not stated here. https://thezvi.substack.com/p/brand-new-ai-solves-a-millennium and https://www.tuftsdaily.com/article/2026/09/mathematicians-still-checking-the-navier-stokes-proof-that-openai-claims-to-have-solved
- [2026-11-03] Altman, reportedly at ADIPEC, said the capability touches Riemann, Hodge and Yang–Mills existence. This is a secondary source and unconfirmed. The OpenAI Oct 26 release was posted a few weeks after the Oct 6 catalog, so the release cadence is roughly monthly. At that pace one more large announcement before 12-31 is plausible. https://qz.com/openai-ai-discoveries-math-computer-science-physics-102726
- [2026-10-26] Rival labs. Anthropic is reportedly racing on BSD (an X-rumour-based report, unconfirmed). Anthropic has Riemann-bound work validated internally in August, and a Sept 26 SLAC-validated N=4 SYM amplitude. DeepMind has fluid-singularity work. A second lab raises the number of independent shots, but nothing confirmed was found. https://explainx.ai/blog/millennium-prize-problems-ai-solved-fact-check-2026 and https://finance.biggo.com/news/c9ddc368-b190-49df-9c38-c00ad2840e21
- [2026-10-xx] The Navier–Stokes wildcard. The Wikipedia "Millennium Prize Problems" page says the result is not independently verified, and the Clay Institute page lists the problem as "Active". If Clay or a body such as the Gross-style committee later called Navier–Stokes solved, a literal reader might count the recognition date as the solve date, because the rule says "first date that ... remains unretracted". The word "another" and the "before period" exclusion point against this. I found no sign of imminent recognition. https://en.wikipedia.org/wiki/Millennium_Prize_Problems
BOTTOM LINE: The most likely YES path is a second OpenAI problem announced with a public manuscript and Lean and reported as "solved" by at least two outlets before 12-31. The candidates are Hodge, which has anonymous-source and NYT-attributed signals, and Yang–Mills, which has a formal review committee but needs the mass gap added. Every path needs the claim to be a full resolution and to survive without a reported unresolved error, and I found no such full second claim as of 11-08.
NOT FOUND: Any full second-problem claim (Hodge, BSD, Riemann, P vs NP, or the Yang–Mills mass gap). Any Clay statement after 2026-09-11. Any journal acceptance. A release date for the Gross committee's assessment. Source Agencies as defined by the contract. The tech-insider and ibtimes pages were blocked, so those points rest on search snippets only.
report: no case
YES means: an AI lab (OpenAI, Anthropic or similar) solves a Millennium Prize Problem other than the already-claimed Navier–Stokes. It must be solved between 2026-09-12 and 2026-12-31. It must pass one of three tests: a peer-reviewed journal publication, an official statement from CMI or a similar body, or reports from at least two Source Agencies. Under the rules, a preprint or announcement alone does not count.
- [2026-11-08] Clay's Millennium Problems page, fetched today, puts Navier–Stokes under "Active problems". BSD, Hodge, P vs NP, Riemann and Yang–Mills are under "Unsolved", and only Poincaré is under "Solved". The page has no statement about any AI claim. CMI has therefore not recognised any problem for 2026, so the "official statement" route has not been triggered. https://www.claymath.org/millennium-problems/
- [2026-11-08] Even the best-known AI claim (Navier–Stokes, announced 2026-09-08) is unrecognised after two months. Wikipedia describes it as not independently verified. The Clay Institute's own Wikipedia entry says it is unverified and subject to a priority dispute. A second problem claimed later would have less time to clear any bar. https://en.wikipedia.org/wiki/Millennium_Prize_Problems and https://en.wikipedia.org/wiki/Clay_Mathematics_Institute
- [2026-10-06] The arXiv paper "Navier–Stokes lost in translation" (Bastounis, Circelli and Hansen) shows the Lean file does not match the natural-language proof in the examples it examines. The "unresolved identified error" clause could therefore bite on AI proofs generally. It would also hurt any follow-on AI claim that relies on Lean as its main evidence. https://arxiv.org/abs/2610.08144
- [2026-10-26] A dev.to scorecard shows 14 OpenAI manuscripts revised after errors were found (changelog of 2026-10-07). Only 300 of 719 top-line results are Lean-checked, and the "quasi-Riemann 11/12 strip" result has no Lean proof. OpenAI's own catalog is full of partial or revised results, not complete solutions. https://dev.to/ashtech52/openai-math-proofs-explained-has-ai-solved-math-problems-what-the-722-manuscripts-really-show-2ele
- [2026-09-19] OddsShopper shows a Kalshi "announces another solution before 2027" contract (a sibling that pays on an OpenAI announcement) swinging from 28¢ to 84¢ to about 64¢ in nine days. That happened on a rumor and an unnamed "substantial progress" line, with thin books. Prices on this theme are noisy and announcement-driven. The page says the contract pays only on a complete announcement, which OpenAI had not made. https://www.oddsshopper.com/articles/prediction-markets/openai-millennium-prize-odds
- [2026-09-19] The same page shows a Kalshi "Clay awards a Millennium Prize before 2029" contract at 27–32¢ (September 9–11 quote). That is the only price I found tied to formal recognition. It is far below the announcement contracts. This points to the real-world gap between announcement and recognition. https://www.oddsshopper.com/articles/prediction-markets/openai-millennium-prize-odds
- [2026-11-02] A Manifold sibling asks whether any of the remaining five problems gets a proposed solution "formally announced or published" between 2026-09-09 and 2026-12-31. It shows 27% on 2026-11-02, down from 38% in early September. That market has a looser standard than this one (announcement, not recognition). Even so, the crowd has cut it in two months of no new claim. https://manifold.markets/HumanClanker/will-another-millennium-prize-probl
- [2026-11-02] Altman, per an unverified secondhand account, said at ADIPEC that the model behind the Navier–Stokes result is the "next GPT" and will be released soon. That is a product-release signal, not a second-problem solution. It adds no evidence of a second complete proof. https://mindstudio.ai/blog/gpt-6-2-bell-leaks
- [2026-10-28] Papex rates Hodge and Yang–Mills only "Partial". It records that three of seven OpenAI Hodge papers were withdrawn, and that OpenAI's Yang–Mills claim covers existence without the mass gap. Clay's own Yang–Mills entry requires the mass gap. A full Hodge or Yang–Mills solution therefore remains unclaimed. https://papex.org/blog/has-ai-solved-a-millennium-prize-problem
- [2026-11-08] Deadline arithmetic: about 53 days remain to 2026-12-31. A journal route is essentially closed, since refereeing of a 165-page proof takes months to years. The CMI route would require a statement for a problem the lab has not yet fully claimed. The Source Agency route needs two agencies to say a problem "has been proven" and cite a public written work. No such second-problem work exists as of today, and the Sept 17 Hodge report was a single anonymous source.
- [2026-11-08] Attribution and "another" ambiguity cut against YES. Any Navier–Stokes recognition would likely be treated as the baseline problem. The "AI lab credited" requirement also excludes independent third-party use of a lab's public model. Clay's Poincaré precedent (about 7 years from proof to prize, 2-year minimum wait) and today's page show the status quo is no recognised solution. https://www.claymath.org/millennium-problems/
BOTTOM LINE: A YES needs a complete, uncontested proof of a new problem plus external validation within about 53 days. As of 2026-11-08 no such claim exists, the AI claims so far are partial, revised or withdrawn, and Clay has recognised nothing. Status-quo inertia and the formal-recognition bar both point to NO.
NOT FOUND: Any full second-problem solution announcement or journal acceptance as of 2026-11-08. Any CMI or two-agency statement about a problem other than Navier–Stokes. A primary Clay statement newer than 2026-09-11. Direct Kalshi pricing for this contract (not searched, per instructions).
3. Forecasters 2 draws, Opus 5.5
claude-opus-5-5, high effort
62% conf
| Yes | 6% |
| No | 94% |
- already resolved?
- Not resolved. CMI's page, fetched today, lists Navier–Stokes as 'Active' and the other five problems as 'Unsolved'. Navier–Stokes was claimed on 2026-09-08, before the market opened, so it is very likely excluded by the word 'another'. No full AI claim on any other problem exists.
- time left
- About 53 days remain until 2026-12-31.
- if nothing changes
- If nothing changes, the market resolves No. The journal route is effectively closed on this timescale, and CMI has recognised nothing.
- path to no
- OpenAI's Yang–Mills claim stays at existence only. The Gross committee explicitly says the mass gap is open, which makes it a partial result. The Hodge, Riemann and BSD results stay partial, revised or withdrawn. Even any new announcement draws error or mistranslation challenges, as Navier–Stokes did.
- path to yes
- OpenAI, or a rival lab, announces a complete Hodge proof or a Yang–Mills mass-gap proof with a manuscript and Lean code. At least two Source Agencies then report the problem as solved, as Quanta and the BBC did for Navier–Stokes within days, and no unresolved error is identified before expiry. The Kalshi mid is 13%, but the bid/ask is thin and the last daily close was 3%. Announcement-based sibling markets have faded from their September highs: Polymarket has Yang–Mills at 19% and Hodge at 13%, and Manifold has 'another problem announced' at 27%. Those markets have looser criteria, and here an announcement must also clear the full-resolution and attribution tests. I put roughly 10–12% on a full-solution announcement and about 50–60% on it then qualifying in time, giving about 6%.
claude-opus-5-5, high effort
62% conf
| Yes | 6% |
| No | 94% |
- already resolved?
- Not met: Clay's page fetched today shows Navier–Stokes 'Active', five problems 'Unsolved', only Poincaré 'Solved'; no full AI claim on a second problem exists. Navier–Stokes was claimed 2026-09-08, before issuance, so under the 'another' wording it is likely the baseline.
- time left
- About 53 days remain until 2026-12-31.
- if nothing changes
- The status quo is No: OpenAI's other results are partial (Yang–Mills existence without the mass gap, CM-abelian Hodge, a Riemann zero-free region), with three Hodge papers withdrawn and 14 manuscripts revised.
- path to no
- No complete new claim arrives, or one arrives but is contested; journal publication and CMI recognition are effectively impossible in this window.
- path to yes
- A lab releases a complete Hodge or Yang–Mills-with-mass-gap proof with a public manuscript and Lean, two Source Agencies report the problem as proven, and no unresolved error is identified by year-end. I estimate about 15–20% for a full new claim and about 30–40% that it qualifies, roughly 6%. The Kalshi mid is 13% but the last close was 3% after a steep 30-day decline; I sit between them, slightly below the mid.
4. Final plain mean of 2 draws
6%
(62% confidence)
Yes: 6%
No: 94%
- already resolved?
- Not resolved. CMI's page, fetched today, lists Navier–Stokes as 'Active' and the other five problems as 'Unsolved'. Navier–Stokes was claimed on 2026-09-08, before the market opened, so it is very likely excluded by the word 'another'. No full AI claim on any other problem exists.
- time left
- About 53 days remain until 2026-12-31.
- if nothing changes
- If nothing changes, the market resolves No. The journal route is effectively closed on this timescale, and CMI has recognised nothing.
- path to no
- OpenAI's Yang–Mills claim stays at existence only. The Gross committee explicitly says the mass gap is open, which makes it a partial result. The Hodge, Riemann and BSD results stay partial, revised or withdrawn. Even any new announcement draws error or mistranslation challenges, as Navier–Stokes did.
- path to yes
- OpenAI, or a rival lab, announces a complete Hodge proof or a Yang–Mills mass-gap proof with a manuscript and Lean code. At least two Source Agencies then report the problem as solved, as Quanta and the BBC did for Navier–Stokes within days, and no unresolved error is identified before expiry. The Kalshi mid is 13%, but the bid/ask is thin and the last daily close was 3%. Announcement-based sibling markets have faded from their September highs: Polymarket has Yang–Mills at 19% and Hodge at 13%, and Manifold has 'another problem announced' at 27%. Those markets have looser criteria, and here an announcement must also clear the full-resolution and attribution tests. I put roughly 10–12% on a full-solution announcement and about 50–60% on it then qualifying in time, giving about 6%.
Pipeline Timing and Cost
Total pipeline time: 69.3s
LLM cost: $1.20
(Sonnet 5.5 $0.71, Opus 5.5 $0.29, 20 web searches $0.20)
Per-tool research timings shown in the Research section above.