forecasting
agent
← all forecasts

Will any AI lab solve a Millennium Prize Problem before 2027?

kalshiScience and Technology2026-10-30llm cost $1.43view on kalshi ↗
12%
Agent
29%
Market Price
-17.0pp
Edge
61%
Confidence
8,510
Volume
6.0c
Spread
62
Days to resolution
6
Markets in event
final rationale
average of 2 draws (11%, 13%), redrawn after an extra research pass
already resolved?
No source shows the condition met: Clay lists Navier–Stokes as 'Active' and the other five as 'Unsolved', with no journal acceptance of any AI Millennium paper. OpenAI's Navier–Stokes claim (Sept 8) predates the Sept 12 issuance and is likely excluded by 'another' and 'solved before time period do NOT count'.
time left
About two months remain, to Jan 1, 2027.
if nothing changes
Status quo is NO: the Oct 26 Yang–Mills claim is existence-only, the mass gap is explicitly open per OpenAI and the review panel, and the proof is withheld. The Riemann, Hodge and BSD releases are partial, and Clay's process takes years.
path to no
Yang–Mills stays existence-only and in panel custody, Navier–Stokes stays disputed (Lean/written-proof mismatch, priority dispute), and nothing meets the journal, Clay or two-Source-Agency test by Dec 31.
path to yes
A lab announces a full mass-gap or other full solution that two Source Agencies quickly report as resolved, or the resolver loosely counts post-issuance Navier–Stokes recognition. I discount the thin, volatile Kalshi price of 24–29% (range 9–74%), which likely conflates announcements with qualifying solutions, but keep weight on rules-ambiguity and surprise-claim tails.
price vs. prediction
market price agent forecast at t+0
22 follow-ups
earlier forecasts on this market
datepredictedmarket priceconfidence
2026-10-01 27% 44% 57%
1. Planner Opus 5.5
Tools chosen: kalshi_direct resolution_source kalshi_related polymarket_related article_search
Planner reasoning
The Kalshi price is the anchor, and the resolution source is the Clay Mathematics Institute's Millennium Problems page. Related markets on Kalshi and Polymarket give cross-venue signals on AI math milestones. Wikipedia gives background on the current status of the problems. With two months left and peer review required, the outcome is almost certainly No unless a near-recognized claim already exists, so one targeted search checks for any AI-credited claimed solution.
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): 29.00% (live; bid 24% / ask 30%) - Last daily close (2026-10-29): 24.00% - 7-day price change: +0.00% - 30-day price change: -43.00% - Average daily volume: 324 contracts - Price range: 9.00% - 74.00% - Data points: 26 days - resolution_source: already_resolved=unclear; matches_criteria=True; last_updated=unknown (page footer © 2026). The CMI Millennium Problems page (© 2026) lists Navier-Stokes under "Active problems" (the Navier-Stokes page itself is labelled "Active"). Birch–Swinnerton-Dyer, Hodge, P vs NP, Riemann and Yang-Mills are listed as "Unsolved problems", and only the Poincaré Conjecture is listed as "Solved". The CMI page sh - kalshi_related: 1 related markets / summaries. series KXAIMILLENNIUM: 0 markets (skipped 5 no-signal) | keyword 'Millennium Prize': ok | keyword 'AI math': no matches - polymarket_related: Polymarket search over 4 keyword(s), kept 20 matches. keyword 'Millennium Prize': 5 markets | keyword 'AI solve math': 5 markets | keyword 'Navier-Stokes': 5 markets | keyword 'Riemann': 5 markets - article_search: Fetched 9 unique articles across 2 queries (lookback=120d). 'AI lab claims proof Millennium Prize Problem Navier-Stokes DeepMind 2026': 8 articles | 'Clay Mathematics Institute statement AI proof': 8 articles Research loops, bottom lines - current state: As of 2026-10-30, no Millennium problem other than Navier–Stokes has even a lab claim of a full solution. Navier–Stokes was claimed before the market opened and has no official acceptance. The October Yang–Mills, Riemann, Hodge and BSD claims are explicitly partial, and Clay's process takes years, so a qualifying new full solution before 2027 would need an unexpected full claim plus rapid recognition. - base rate: The historical reference class is one MPP resolved in 26 years, and the fastest recognition after posting was years, not weeks. Navier–Stokes, the only AI claim so far, appears to be the excluded baseline and is still under dispute. A YES in about 2 months would need a new, near-instant, error-free AI solution of another problem, or a surprising early acceptance of Navier–Stokes if the resolver counts it. Markets price lab announcements higher, but this contract requires verified resolution. - yes case: The most likely YES path is the loose reading of the rules, in which Navier–Stokes recognition (a Clay statement citing OpenAI's paper, or two outlets reporting it solved) counts. After that comes a new full claim, most plausibly Yang–Mills with a mass gap, that wins quick two-source reporting. No source shows that any such trigger is scheduled before 2027, and the Yang–Mills claim explicitly excludes the mass gap. - no case: For a new problem to resolve YES by Dec 31, a full, error-free AI claim would have to appear and be adopted at once by two Source Agencies or by Clay. The only advanced claims are partial (Yang–Mills existence), unreleased, or under a priority and Lean-mismatch dispute (Navier–Stokes, probably excluded). A "reported unresolved error" clause and status-quo inertia both favour NO. Crux follow-up - The evidence favours NO. No journal publication or named-work recognition exists, and Clay's timeline runs far past year end. The one complete AI claim (NS) was announced before the window and is disputed, and the October Yang–Mills claim is explicitly partial and unreleased. The main YES risk is ambiguity in how resolvers treat NS. A resolver reading "at least two Source Agencies report resolved" loosely could count Quanta, Scientific American and Clay's "apparently settled" statement. A new fu
tools
toolstatustimeitemssummary
kalshi_direct OK 0.2s 1 ## This Market's Kalshi Data **Before 2027** (KXAIMILLENNIUM-28-27JAN01) - Current price (probability): 29.00% (live; bid 24% / ask 30%) - Last daily close (2026-10-29): 24.00% - 7-day price change: +0.00% - 30-day price change: -43.00% - Average daily volume: 324 contracts - Price range: 9.00% - 7
resolution_source OK 14.3s 1 already_resolved=unclear; matches_criteria=True; last_updated=unknown (page footer © 2026). The CMI Millennium Problems page (© 2026) lists Navier-Stokes under "Active problems" (the Navier-Stokes page itself is labelled "Active"). Birch–Swinnerton-Dyer, Hodge, P vs NP, Riemann and Yang-Mills are li
kalshi_related OK 2.2s 1 1 related markets / summaries. series KXAIMILLENNIUM: 0 markets (skipped 5 no-signal) | keyword 'Millennium Prize': ok | keyword 'AI math': no matches
polymarket_related OK 0.4s 20 Polymarket search over 4 keyword(s), kept 20 matches. keyword 'Millennium Prize': 5 markets | keyword 'AI solve math': 5 markets | keyword 'Navier-Stokes': 5 markets | keyword 'Riemann': 5 markets
article_search OK 0.5s 9 Fetched 9 unique articles across 2 queries (lookback=120d). 'AI lab claims proof Millennium Prize Problem Navier-Stokes DeepMind 2026': 8 articles | 'Clay Mathematics Institute statement AI proof': 8 articles
web research agents 4/4 ok, 18 searches, 2 page reads
dimensionstatustimesearchespage readsreport
current state OK 22s 5 0 4785 chars
base rate OK 24s 5 0 5889 chars
yes case OK 21s 4 0 5255 chars
no case OK 26s 4 2 5906 chars
report: current state
- [2026-09-08] OpenAI published "On the Navier–Stokes Millennium Prize Problem". It says an internal system produced a proof, with a Lean formalization, that smooth forced 3D Navier–Stokes can blow up in finite time. It says this settles statements C and D of the Clay formulation and that it does not intend to claim the prize. https://openai.com/index/navier-stokes-solution/ - [2026-09-11] Clay (CMI) said it contemplates "the announcement that the Navier-Stokes problem has apparently been settled". The Papex tracker quotes Clay as calling its evaluation process "deliberately unhurried". Clay relabelled the problem "Active" instead of "Unsolved". It has not declared the problem solved and has not named a work. https://www.claymath.org/news/navier-stokes-announcement/ - [2026-09-11] Under Clay's rules, a prize is considered only after refereed publication, two years and general acceptance. The Navier–Stokes claim (8 Sept) came before this market opened (12 Sept) and carries a priority dispute with Buckmaster and Anthropic's Alpöge. It is probably "solved before the period" for this contract, and the rules' "another" suggests the same. Its status is unresolved either way. https://en.wikipedia.org/wiki/Navier%E2%80%93Stokes_priority_controversy - [2026-10-15, updated 2026-10-28] Papex, citing Clay's pages, reports Clay recognises no AI solution. Poincaré is "Solved", Navier–Stokes is "Active", and P vs NP, Riemann, Hodge, BSD and Yang–Mills are "Unsolved". Clay still lists Yang–Mills and P vs NP as Unsolved as of 28 Oct. https://papex.org/blog/has-ai-solved-a-millennium-prize-problem - [2026-10-26] OpenAI announced a "proof of the existence of pure Yang–Mills theory". MLQ notes this is narrower than the full problem: Clay also requires a mass gap, which this does not settle. OpenAI itself describes it as part of the Clay problem. https://mlq.ai/news/openai-puts-three-ai-generated-breakthroughs-before-academic-review - [2026-10-26/27] OpenAI and Anthropic each claim to refute the Exponential Time Hypothesis. OpenAI's post is dated 17 Oct and revised 25 Oct. Anthropic's is dated 26 Oct. Papex says this does not settle P vs NP, and it found no AI claim on P vs NP. https://papex.org/blog/has-ai-solved-a-millennium-prize-problem - [2026-10-13 approx.] OpenAI released 722 AI-generated papers (372 result families) on GitHub. Science reports that one proves a zero-free region for Re(s) > 7/8. That is only a weaker "quasi-Riemann hypothesis" and not the Riemann hypothesis. https://www.science.org/content/article/ai-edges-closer-solving-riemann-hypothesis-one-math-s-hardest-problems - [2026-10-15] OpenAI's October releases on Hodge (special families, CM abelian varieties) and BSD (Selmer corank 0 or 1) are partial. Papex says three of the seven Hodge papers were withdrawn. OpenAI says unverified proofs may contain errors. https://papex.org/blog/has-ai-solved-a-millennium-prize-problem - [2026-10-26] Noema Magazine says another Clay problem was "felled by an AI system" on 26 Oct. That overstates the pure Yang–Mills existence claim. https://noemamag.com/a-new-dawn-for-mathematical-progress - [2026-10-27] A tweet from Diamandis says the Yang–Mills proof was finalized on 16 Oct and "five remain". It is social media and counts Navier–Stokes as solved. https://x.com/PeterDiamandis/status/2115074979373158516 - [2026-10-30] Wikipedia says only Poincaré is officially declared solved. It says OpenAI's Navier–Stokes result is unverified, and that Clay considers proposals only at least two years after publication. https://en.wikipedia.org/wiki/Millennium_Prize_Problems - No Clay or journal acceptance of any AI result was found. No scheduled event before 2027-01-01 that would trigger the resolution criteria was found. - The rules' third route needs two Source Agencies to report a problem as resolved by reference to a specific public work. Mass-gap-free Yang–Mills existence and the Riemann zero-free region are widely reported as partial, so they do not qualify. BOTTOM LINE: As of 2026-10-30, no Millennium problem other than Navier–Stokes has even a lab claim of a full solution. Navier–Stokes was claimed before the market opened and has no official acceptance. The October Yang–Mills, Riemann, Hodge and BSD claims are explicitly partial, and Clay's process takes years, so a qualifying new full solution before 2027 would need an unexpected full claim plus rapid recognition. NOT FOUND: Clay's pages for Yang–Mills and P vs NP, fetched directly (only secondary reports of their status). The Quartz article on OpenAI's 26 Oct release (403). Any full claim on P vs NP, the Riemann hypothesis, Hodge or BSD. Any peer-reviewed journal publication of an AI Millennium result. Any scheduled Clay or journal decision date before 2027-01-01.
report: base rate
- [2026-09-08] OpenAI posted a writeup and Lean formalization claiming a finite-time blowup for forced Navier–Stokes. It says this resolves the problem via statements C and D of Clay's formulation, and that it does not intend to claim the prize. This is the "first" AI claim, and it predates the market's 2026-09-12 opening. It is probably the baseline Navier–Stokes event; the "another" wording and "problems solved before time period do NOT count" likely exclude it. https://openai.com/index/navier-stokes-solution/ - [2026-09-11] Clay called Navier–Stokes "apparently settled" but said its evaluation is "deliberately unhurried". Wikipedia (as of Oct 2026) still lists the problem as "active" at Clay. Clay's rules have no consideration until at least two years after publication. https://navier-stokes.org/navier-stokes-problem-solved/ and https://en.wikipedia.org/wiki/Millennium_Prize_Problems - [2026-10-13] Scientific American: experts are still working through the 166-page OpenAI proof. They find it hard to read, and the Lean code is being used to help. The proof exploits a "loophole" in Clay's framing, namely an unphysical forcing term. Mathematicians dispute priority and the proof's relation to the Buckmaster/Alpöge work. https://www.scientificamerican.com/article/openais-blockbuster-navier-stokes-proof-isnt-what-we-thought-it-was/ - [2026-10-08] A Hacker News thread cites an arXiv paper, "Navier–Stokes Lost in Translation". It claims OpenAI's Lean proof does not correspond to its natural-language proof. I did not check the paper itself; this is a secondary report of an unreviewed preprint. https://news.ycombinator.com/item?id=49994145 - [Historical base rate] Only 1 of 7 Millennium problems has ever been officially solved: Poincaré. Perelman posted it in 2002–03, got the Fields Medal in 2006 and the Clay prize on 2010-03-18. That is about 7 years from posting to Clay recognition and 10 years from the problem's 2000 launch. No other problem had resolved in the roughly 26 years since 2000, before OpenAI's claim. https://plus.maths.org/news-world-maths-perelman-awarded-millennium-prize-proof-poincar-conjecture - [2026-09-08] CNN confirms only Poincaré is officially solved. Mathematicians have not independently verified OpenAI's claim, and Wikipedia says the same. https://edition.cnn.com/2026/09/09/business/openai-millennium-problems-navier-stokes-hnk - [Rule structure] The market's three YES paths are peer-reviewed journal publication, a CMI/academic body statement citing a written work, or at least two "Source Agencies" reporting resolution. A preprint or Lean proof alone does not count. A solution reported to contain an unresolved identified error does not count. Peer review plus a Clay review would normally take well over 2 months. The media-report path is the only fast one, and it is the one most exposed to an error report. This is from the question text. - [2026-10-30] Polymarket "Which Millennium Prize Problem will AI solve next?" (a different market, resolved on lab announcement with no Clay confirmation, ending 2027-12-31, opened 2026-09-09, about $320K volume). It prices "No solution by Dec 31, 2027" at about 38%. Yang-Mills is about 21%, Hodge about 19%, BSD about 16%, Riemann about 3–5% and P vs NP about 2%. The market description says Navier–Stokes is excluded, and it has a looser standard than this question. https://polymarket.com/event/which-millennium-prize-problem-will-ai-solve-next - [2026-09-11] Polymarket tweeted 71–72% for "AI lab announces another Millennium Prize solution" "this year". A commenter noted it was a 71% chance an AI lab announces, not that it actually solves one. https://x.com/Polymarket/status/2098488080827097355 - [~2026-07, snapshot undated, about 107 days old] Kalshi AI category: "Will OpenAI announce another Millennium Prize solution?" had Before 2027 at about 20% and Before 2028 at about 63%. "When will an AI lab solve another MPP?" had Before Jul 1 2027 at 47% and Before 2028 at 62%. These snapshots are stale and likely date from after Sept 8. They show the market treating "announce" and "solve" differently. https://kalshi.com/category/science/ai - [2026-09-10] On Manifold (play money), "all seven solved before 2030" was about 24%. A separate "another MPP solved in 2026" market was 38% on Sept 5. https://tech-insider.org/millennium-prize-problems-solved-2030-prediction-market/ - [2026-10-17] A thin Kalshi contract on Yang-Mills being the next MPP solved before 2045 jumped to about 20%. The report is from a secondary site and the volume is low. https://tech-insider.org/yang-mills-solved-next-before-2045-kalshi-odds/ - [2026-02] Octagon AI's model gave "next Millennium Prize awarded before 2027" about 2%, against a market price of 15%. Its track record is unknown and the page predates the OpenAI claim. https://www.octagonai.co/markets/science-and-technology/when-will-the-next-millennium-prize-be-awarded/ BOTTOM LINE: The historical reference class is one MPP resolved in 26 years, and the fastest recognition after posting was years, not weeks. Navier–Stokes, the only AI claim so far, appears to be the excluded baseline and is still under dispute. A YES in about 2 months would need a new, near-instant, error-free AI solution of another problem, or a surprising early acceptance of Navier–Stokes if the resolver counts it. Markets price lab announcements higher, but this contract requires verified resolution. NOT FOUND: Clay's primary statement page (claymath.org) was not read directly; its September 11 text comes from secondary sites. I found no announced AI claim on any of the other six MPPs. I found no peer-reviewed publication or acceptance of the OpenAI Navier–Stokes proof. I found no Kalshi page for this exact contract, and I did not look for its price. I found no expert-survey estimates, and no track record for any of the forecasters above.
report: yes case
YES means: some AI lab's Millennium Prize Problem solution, other than ones already solved before the market opened, counts as solved between now and 2026-12-31. It must meet one of three tests. The first is publication in a peer-reviewed journal. The second is a statement by Clay or an equivalent body that cites a specific work. The third is reports from at least two Source Agencies that cite a specific public work. It must also stay unretracted and free of an unresolved identified error. - [2026-10-14] Scenarica's scenario analysis (a forecasting newsletter, not a primary source) says the Navier–Stokes proof was "checked line by line" by Lean and had no published error after a month. It says Fefferman, who wrote Clay's official problem statement, told Quanta he was "thrilled" the problem was solved. It also describes the media route as louder than anything Navier–Stokes has received so far. It prices "a second problem accepted" at 17% for 2027-06-30 and "first claim unravels" at 2%. This is the strongest argument for the fast media path. https://scenarica.substack.com/p/the-referee-will-see-you-eventually - [2026-10-14] The same piece says some rival forecasting markets accept any two of two dozen outlets, and that their written rules leave Navier–Stokes in scope. A literal-minded resolver could count a Navier–Stokes event, for example a journal acceptance, if the "another" and "before period" exclusions are read loosely. That is a rules-ambiguity path to YES, not a mathematical one. https://scenarica.substack.com/p/the-referee-will-see-you-eventually - [2026-10-25] Scott Armstrong, a PDE mathematician, called the Navier–Stokes construction "unbelievable" and possibly beyond what a human would devise, though still understandable by people. This is expert praise, not a referee verdict. A preprint challenging the Lean-to-paper match also exists. https://runtimewire.com/article/scott-armstrong-says-openai-s-navier-stokes-result-may-exceed-human-invention - [2026-10-26 to 10-27] OpenAI's Yang–Mills existence proof was completed in natural language on 9 Sept and in Lean on 16 Oct. A QCD proof followed on 18 Sept. A five-person panel reviews it: Gross, Jaffe (who co-wrote Clay's Yang–Mills problem description), Douglas, Klebanov and Nekrasov. Jaffe's presence shows that existence was a real Clay requirement. The panel's 27 Oct statement says the mass gap remains open. The proof is withheld (1,147 pages per Woit). This is the nearest thing to a full-claim path, but only if a mass-gap claim follows. Nobody has made one. https://www.math.columbia.edu/~woit/wordpress/?p=15948 - [2026-10-26] Woit notes that the Alpöge/Sussman group, who work independently, say they have a 4D Yang–Mills existence proof on R^4. It runs to tens of thousands of unchecked pages. Alpöge is an Anthropic mathematician, so a second lab has a Yang–Mills route. Neither lab claims the mass gap. https://www.math.columbia.edu/~woit/wordpress/?p=15948 - [2026-10-06] OpenAI posted 722 manuscripts from 372 result families, produced at about three hours of thinking per problem across roughly 4,000 problems. About 42% (300 of 719) had Lean formalizations by 14 Oct. The rate is high, and the same model could be pointed at a Millennium problem that the same sweep has not yet tried. https://scenarica.substack.com/p/the-referee-will-see-you-eventually - [2026-09-11] OpenAI told the NYT it had made "substantial progress on another Millennium Prize problem". That turned out to be Yang–Mills, at the existence stage. This shows that OpenAI discloses on a weeks-long cadence, with its Navier–Stokes post on 8 Sept and Yang–Mills on 26 Oct. Another claim before year end is plausible on cadence alone. https://explainx.ai/blog/millennium-prize-problems-ai-solved-fact-check-2026 - [2026-10-12] Manifold's "AI solves a Millennium problem by EOY 2027" stood at 83%, with about 35% of its volume traded in 24 hours. It is play money and a looser standard than this contract. It does show the crowd treats further AI claims as likely. https://tech-insider.org/ai-millennium-problem-manifold-odds-eoy-2027/ - [2026-09-17] The Conversation notes that Anthropic also claimed a Claude-completed Fermat's Last Theorem formalization in 11 days. Two labs are now producing results at Millennium scale, so competition raises the chance of a rushed full claim. https://theconversation.com/openai-may-have-solved-the-navier-stokes-equation-but-mathematics-is-more-than-1-million-trophy-hunts-291963 BOTTOM LINE: The most likely YES path is the loose reading of the rules, in which Navier–Stokes recognition (a Clay statement citing OpenAI's paper, or two outlets reporting it solved) counts. After that comes a new full claim, most plausibly Yang–Mills with a mass gap, that wins quick two-source reporting. No source shows that any such trigger is scheduled before 2027, and the Yang–Mills claim explicitly excludes the mass gap. NOT FOUND: Any claim of a full solution of any Millennium problem other than Navier–Stokes. Any journal acceptance of an AI Millennium proof. Any Clay statement citing a specific AI work as resolving a problem. Any announced date for a Clay or journal decision. Clay's own pages, which I read only through secondary sources.
report: no case
YES means: an AI lab (another Millennium problem, not Navier–Stokes) is credited with a complete solution that, before 2027-01-01, is published in a peer-reviewed journal, officially recognised by CMI or an academic body, or reported as resolved by at least two Source Agencies. The solution must be unretracted and have no unresolved identified error. - [2026-10-27] King's College London and Cambridge (Bastounis, Circelli, Hansen) issued a press release on their Oct 6 arXiv preprint. It says OpenAI's Lean version of the Navier–Stokes proof does not fully match the written proof. In one place the written claim is stronger than the Lean claim. They say the written proof "still needs to be checked independently". The paper is not peer reviewed and does not show the proof is wrong. The release supplies an "identified error" or mismatch argument against the one AI claim that is furthest along. It also weakens the Lean-as-verification route for all later claims. https://www.kcl.ac.uk/news/openais-navier-stokes-proof-called-into-question - [2026-10-26/27] The Yang–Mills claim is not a full solution. Per Scientific American (via search snippets; the page was blocked when fetched), OpenAI claimed only "existence", and the mass gap remains open. The Clay problem requires both. No proof is public, and a "shepherding" panel is holding it back. Woit's blog (fetched) says the panel keeps it secret because "it might be wrong". It plans to validate the Lean definitions first and release the Lean proof only afterwards. The written proof is about 1,147 pages. A claim that is unpublished, partial and being validated cannot meet any of the three YES paths by Dec 31. https://www.math.columbia.edu/~woit/wordpress/?p=15948 - [2026-10-26] Woit notes that a competing human effort (Alpöge/Sussman) is public but is "tens of thousands of pages of unchecked" material. Even the most advanced Yang–Mills efforts are not at a stage where referees can finish in about 9 weeks. https://www.math.columbia.edu/~woit/wordpress/?p=15948 - [2026-10-26] Woit's reading of the Jaffe–Witten statement: even a rigorous T⁴ construction would not give a mass gap uniform in volume or the infinite-volume limit to R⁴. This is a technical reason the existence-only claim is far from a "complete proof ... as originally formulated". https://www.math.columbia.edu/~woit/wordpress/?p=15948 - [2026-09-22 to 2026-10-13] Scientific American says OpenAI's Navier–Stokes proof exploits a forcing "loophole" in the Clay statement. Several research groups are still simplifying and checking it. It would resolve only options C/D and not the unforced problem. If the resolver counts Navier–Stokes at all, the status is "disputed, unverified". If it excludes it (likely under "another" and "before period"), it adds no YES path. https://www.scientificamerican.com/article/did-openai-solve-the-wrong-navier-stokes-problem/ - [2026-10-28] Wikipedia's Navier–Stokes page still gives Clay's status as "active". It does not give "solved". Clay has not named a work. The second YES path (CMI or a recognised body naming a specific work) therefore has no trigger. The first path needs journal acceptance of a 166-page proof, a priority dispute and a Lean mismatch within weeks. https://en.wikipedia.org/wiki/Navier%E2%80%93Stokes_existence_and_smoothness - [2026-09-11] Clay's own process, quoted by implicator.ai, is "deliberately unhurried" and "absolutely rigorous". Clay's president (Bridson) said so. The same article says Clay "has not accepted" the result. This is the explicit institutional stance. https://www.implicator.ai/clay-institute-navier-stokes-openai-proof-claim/ - [2026-10-27] Science reports that Lean-translation weaknesses have been found, and that Lean software bugs have let invalid proofs through. OpenAI told New Scientist it knows of the gap. A "Source Agency" would have to treat any new claim as settled in this climate, which makes two agencies reporting resolution unlikely for any new claim. https://science.org/content/article/ai-shaking-math-can-its-proofs-be-trusted - [2026-10-26] AGMAI (the advisory panel OpenAI set up) objected that OpenAI's release approach broke its guidance. Scientific American reports this (snippet). Institutional friction over how AI results enter the literature adds delay. https://www.scientificamerican.com/article/dueling-proofs-from-openai-and-anthropic-astound-scientists/ - [2026-10-26] OpenAI's own framing: it says it will not seek the prize, and it says "existence", not "solved". Quantum Research Center says OpenAI addresses existence, not the mass gap. aiweekly says "no $1M prize is triggered". Under the contract, a claim still needs a complete resolution. https://quantumresearch.center/news/articles/openai-yang-mills-existence/ - [2026-10-30] No fast route exists. Perelman needed years. Clay's rules, as summarised on Manifold, include a 2-year wait after publication. A journal can run quickly (MDPI's median first decision is about 20 days), but a reputable mathematics journal will not referee a Millennium-level claim in weeks. Third-party sources give no sign that any journal has accepted any AI Millennium paper. https://www.mdpi.com/journal/ai BOTTOM LINE: For a new problem to resolve YES by Dec 31, a full, error-free AI claim would have to appear and be adopted at once by two Source Agencies or by Clay. The only advanced claims are partial (Yang–Mills existence), unreleased, or under a priority and Lean-mismatch dispute (Navier–Stokes, probably excluded). A "reported unresolved error" clause and status-quo inertia both favour NO. NOT FOUND: Clay's Yang–Mills and P vs NP pages read directly. A fetched copy of the Scientific American Yang–Mills article (403 error). Any AI claim of a full solution to P vs NP, Riemann, Hodge or BSD. Any journal acceptance of an AI Millennium paper. A published list of the contract's "Source Agencies".
3. Forecasters 2 draws, Opus 5.5, re-drawn after crux research
crux research
Crux research ran because the draws' mean 0.12 sits 0.16 from the same-question market "this kalshi market" at 0.29. One more research agent looked into that point, and every draw below was redone with its report.
first-round draws: 12%, 13% , crux research: 4 searches, 1 page read
report: crux
CRUX: Whether any problem other than Navier–Stokes (NS) could reach "solved" status under the rules before 2027. That mainly means Yang–Mills with the mass gap. The other question is whether the NS claim could count at all, given that it predates the Sept 12 opening and the rules say "another" and "solved before time period do NOT count". Evidence: - **2026-09-08, NS claim.** OpenAI announced a claimed resolution of the Navier–Stokes problem together with a Lean formalization . It addresses Clay statements C and D, a disproof of global smoothness. Quanta's headline said AI had solved the problem. (Sources: rits.shanghai.nyu.edu, quantamagazine.org) - **2026-09-11, Clay statement.** Clay said the Navier-Stokes problem "has apparently been settled" , and that the process is "deliberately unhurried" . This statement does not name a work, and it predates the Sept 12 issuance. (Sources: the-decoder.com, mixed-news.com) - **Clay's rules.** Clay requires publication in a qualifying outlet, at least two years since publication, and general acceptance . A formal Clay recognition of any AI work before 2027 is therefore very unlikely. The only open routes are a journal publication or two Source Agencies reporting a resolution. - **NS is contested.** Scientific American reported the proof "unambiguously" solves the problem as Clay originally formulated it , but only through the forced-case loophole. Science reports that researchers found the Lean code watered down some of OpenAI's original English claims (2026-10-28 snapshot). There is also a priority dispute with Buckmaster and Alpöge. (Sources: scientificamerican.com, science.org) - **2026-10-26, Yang–Mills.** OpenAI announced an existence proof for 4D Yang–Mills and QCD. It has not claimed the "mass gap" part. The Lean formalization was finished Oct 16, but the physics artifacts are not public. (Sources: mindstudio.ai, cellcog.ai) - **2026-10-27, Yang–Mills review.** A shepherding committee (Gross, Douglas, Jaffe, Klebanov, Nekrasov) holds custody of the proof. It plans to release the proof along with its assessment "in due course". The Institute for Advanced Study (IAS) advisory group demanded immediate release. No verification has happened. (Sources: kitp.ucsb.edu, ccleaks.com) - **Other problems.** A tracker lists no authoritative completed AI solution for BSD, Hodge or P vs NP . For Riemann it reports only progress on a related zero-density result, not a solution (kingy.ai, about 2026-09-12). An independent project says OpenAI's mass-publication totals count published claims, not validated proofs (mlq.ai, 2026-10-27). - **Pace.** OpenAI's disclosures have come every few weeks (Sep 8, Oct 6, Oct 26). Each Millennium-level claim so far has been either partial or already announced before Sept 12. BOTTOM LINE: The evidence favours NO. No journal publication or named-work recognition exists, and Clay's timeline runs far past year end. The one complete AI claim (NS) was announced before the window and is disputed, and the October Yang–Mills claim is explicitly partial and unreleased. The main YES risk is ambiguity in how resolvers treat NS. A resolver reading "at least two Source Agencies report resolved" loosely could count Quanta, Scientific American and Clay's "apparently settled" statement. A new full claim before year end is the other risk. NOT FOUND: - The text of the market's "Source Agencies" list. - Any statement on whether NS counts as "solved before" the period. - Any Clay page update after Sept 11, or the Clay "Active" status label (Clay's site could not be fetched). - Any journal submission or acceptance of the NS paper. - The Yang–Mills proof text, or any OpenAI mass-gap claim. - Any Anthropic or Google DeepMind Millennium claim.
claude-opus-5-5, high effort
62% conf
Yes 11%
No 89%
already resolved?
No source shows the condition met: Clay lists Navier–Stokes as 'Active' and the other five as 'Unsolved', with no journal acceptance of any AI Millennium paper. OpenAI's Navier–Stokes claim (Sept 8) predates the Sept 12 issuance and is likely excluded by 'another' and 'solved before time period do NOT count'.
time left
About two months remain, to Jan 1, 2027.
if nothing changes
Status quo is NO: the Oct 26 Yang–Mills claim is existence-only, the mass gap is explicitly open per OpenAI and the review panel, and the proof is withheld. The Riemann, Hodge and BSD releases are partial, and Clay's process takes years.
path to no
Yang–Mills stays existence-only and in panel custody, Navier–Stokes stays disputed (Lean/written-proof mismatch, priority dispute), and nothing meets the journal, Clay or two-Source-Agency test by Dec 31.
path to yes
A lab announces a full mass-gap or other full solution that two Source Agencies quickly report as resolved, or the resolver loosely counts post-issuance Navier–Stokes recognition. I discount the thin, volatile Kalshi price of 24–29% (range 9–74%), which likely conflates announcements with qualifying solutions, but keep weight on rules-ambiguity and surprise-claim tails.
claude-opus-5-5, high effort
60% conf
Yes 13%
No 87%
already resolved?
Not met: Clay lists Navier–Stokes as 'Active' and the other five open problems as 'Unsolved'. No journal publication or Clay statement naming a work exists. The Sept 8 Navier–Stokes claim predates the Sept 12 issuance and is likely excluded by 'another'/'solved before time period'; it is also contested (priority dispute, Lean-mismatch preprint).
time left
About 2 months remain, to Dec 31, 2026.
if nothing changes
Status quo is NO: October Yang–Mills (existence only, no mass gap, proof withheld by panel), Riemann (zero-free region only), Hodge and BSD (partial, some withdrawn) all fall short of complete solutions.
path to no
No new full claim, or a new claim draws immediate error reports; Clay's multi-year process precludes recognition.
path to yes
OpenAI or Anthropic announces a full solution (e.g. Yang–Mills mass gap) and two Source Agencies quickly report it as proven, or a resolver loosely counts Navier–Stokes. The market anchor is 24–29% after a 43-point 30-day drop; research shows no imminent full claim, so I price about 9–11% for a new solution plus 2–3% for rules ambiguity.
4. Final plain mean of 2 draws, second round
12% (61% confidence)
Yes: 12% No: 88%
average of 2 draws (11%, 13%), redrawn after an extra research pass
already resolved?
No source shows the condition met: Clay lists Navier–Stokes as 'Active' and the other five as 'Unsolved', with no journal acceptance of any AI Millennium paper. OpenAI's Navier–Stokes claim (Sept 8) predates the Sept 12 issuance and is likely excluded by 'another' and 'solved before time period do NOT count'.
time left
About two months remain, to Jan 1, 2027.
if nothing changes
Status quo is NO: the Oct 26 Yang–Mills claim is existence-only, the mass gap is explicitly open per OpenAI and the review panel, and the proof is withheld. The Riemann, Hodge and BSD releases are partial, and Clay's process takes years.
path to no
Yang–Mills stays existence-only and in panel custody, Navier–Stokes stays disputed (Lean/written-proof mismatch, priority dispute), and nothing meets the journal, Clay or two-Source-Agency test by Dec 31.
path to yes
A lab announces a full mass-gap or other full solution that two Source Agencies quickly report as resolved, or the resolver loosely counts post-issuance Navier–Stokes recognition. I discount the thin, volatile Kalshi price of 24–29% (range 9–74%), which likely conflates announcements with qualifying solutions, but keep weight on rules-ambiguity and surprise-claim tails.
5. Entry Decision Opus 5.5 trader
Side
nobetting no, wins if this resolves no
Position Size
$800
Contracts
1053
Trader Confidence
61%
Chance of Yes
  • market at entry 29%
  • market now 74%
  • agent 69%, nov 20
  • our side
Agent put yes at 12% with the market at 29%, so we bought no at $0.76 (15.0pp edge).
Trade rationale
This bet is with the favourite. The market prices YES at 27% and the forecast says 12%, so both put NO as the likely outcome, and that has been the book's profitable type of trade (44/7 closed). The 15pp edge rests on resolution mechanics: OpenAI's Navier–Stokes claim came on Sept 8, before the Sept 12 issuance, so it likely doesn't count. The Oct 26 Yang–Mills result reportedly proves existence only, with the mass gap still open. And Clay review, journal acceptance or two Source Agencies reporting a solution would all need to happen within 62 days. On the last scan the forecast was 27% against a 44% market, and the price has since fallen about 19pp toward it. The BBC (Oct 28) confirms the Navier–Stokes priority dispute is contested rather than settled. Forecaster confidence is a middling 0.61, and the two draws were close together (11% and 13%). I could not find the Oct 26 Yang–Mills claim in the news index, and labs here are clearly producing claims fast, so the risk of a surprise announcement is real.
Allocation logic
The edge is large enough for a bigger bet, but I'm sizing below the $1000 baseline. The book already holds about $2.3k of correlated 'AI/Millennium problem not solved' NO positions (Hodge x2, Riemann, P vs NP), the spread is 6 cents on a thin market, and one loosely judged resolution could hit all of these at once.
no contract: entry $0.76, now $0.26as of 21:39 utc
Status: OPEN
P&L: -$539.76
Pipeline Timing and Cost
Total pipeline time: 114.9s
LLM cost: $1.43 (Sonnet 5.5 $0.87, Opus 5.5 $0.38, 18 research-loop web searches $0.18)
Per-tool research timings shown in the Research section above.