← all forecasts
Will P vs NP be resolved before 2028?
4%
Agent
8%
Market Price
-3.5pp
Edge
72%
Confidence
6,298
Volume
5.0c
Spread
426
Days to resolution
6
Markets in event
final rationale
- already resolved?
- The condition has not been met. The Clay page (© 2026) labels P vs NP "Unsolved". MathWorld, Cairn Commons and ProofAtlas list no accepted proof, and the late-October ETH refutation from OpenAI and Anthropic explicitly "does not bear on P vs NP".
- time left
- About 14 months remain before the Jan 1, 2028 deadline.
- if nothing changes
- If nothing changes, the market resolves No. The Clay prize route is effectively closed by the two-year waiting rule, so only a peer-reviewed proof, recognition by an authoritative body, or two Source Agencies reporting an unretracted proof could trigger Yes.
- path to no
- The likely No path is that AI labs keep producing algorithmic upper-bound results (ETH, Subset Sum) that never cross the relativization, natural-proofs and algebrization barriers, while fringe claims resting on axioms keep failing review.
- path to yes
- The Yes path is an AI lab releasing a Lean-checked P≠NP or P=NP proof that a top theorist like Fortnow quickly endorses, two outlets report as a proof, and no identified error surfaces before expiry. The ETH and Navier–Stokes precedents show this route is mechanically possible. Kalshi trades at 7% (bid 5%/ask 10%) on thin volume, down from 16%. Given the rapid AI-math progress in 2026 I sit somewhat below the market, not far below. Expert consensus (Fortnow: "not in my lifetime"), the absence of any credible claim, and the retraction/error clause support pricing this near 4%.
price vs. prediction
market price
agent forecast at t+0
14 follow-ups
1. Planner Opus 5.5
Tools chosen:
kalshi_direct
resolution_source
kalshi_related
polymarket_related
wikipedia
Planner reasoning
P vs NP being resolved by 2028 is a near-zero base-rate event, so the market price is the anchor. The Clay Millennium Prize page and Wikipedia give the official status of the problem. Kalshi and Polymarket markets on related conjectures help gauge how much a longshot premium is inflating the price.
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 2028** (KXPNP-32-28JAN01) - Current price (probability): 7.00% (live; bid 5% / ask 10%) - Last daily close (2026-10-29): 16.00% - 7-day price change: -6.00% - 30-day price change: -6.00% - Average daily volume: 57 contracts - Price range: 13.00% - 71.00% - Data points: 5 days
- resolution_source: already_resolved=unclear; matches_criteria=True; last_updated=unknown (page footer © 2026). The official Clay Mathematics Institute P vs NP page (© 2026) labels the problem "Unsolved". It says "so far no one has managed to prove" that these problems are as hard as they appear. The page lists no proof, disproof or award.
- kalshi_related: 2 related markets / summaries. series KXPNP: 0 markets (skipped 6 no-signal) | keyword 'millennium prize': ok | keyword 'conjecture proven': ok
- polymarket_related: Polymarket search over 3 keyword(s), kept 10 matches. keyword 'P vs NP': 0 markets | keyword 'millennium prize': 5 markets | keyword 'Riemann hypothesis': 5 markets
- wikipedia: Fetched 2 Wikipedia entries (0 missing pages).
Research loops, bottom lines
- current state: As of Nov 6, 2026, P vs NP is open. The recent AI news (ETH refutation) is a weaker-than-P=NP result, and the experts reviewing it and the labs themselves say it does not address P vs NP. No scheduled event before Jan 2028 is known that would resolve it, though AI-lab claims are arriving at a faster pace.
- base rate: The outside view is a very low base rate. In 26 years, 0 of 7 Millennium problems besides Poincaré were resolved, and P vs NP has resisted about 55 years. Experts expect it to take decades or longer, and recent AI-assisted claims target adjacent questions or rest on axioms. The main residual risk is a loosely worded Kalshi clause, such as two media outlets reporting a proof, and 2026 AI-math hype raises that risk slightly.
- yes case: The best YES case is one AI-lab release of a complete P≠NP proof, with Lean verification, that two outlets report as a proof before 2028. The ETH and Subset Sum results and the Navier–Stokes precedent show that route is mechanically possible. But no source shows a credible claim yet, and the ETH result is explicitly not a P vs NP step.
- no case: Under the literal criteria, every route to YES needs a vetted, unretracted proof of P vs NP by Jan 2028. The only 2026 complexity breakthrough is an upper-bound result far from polynomial time. The Clay two-year wait, the retraction clause and the expert consensus all point to NO. The residual YES risk is a sloppy media or resolution reading, or an unexpected AI-generated proof.
tools
| tool | status | time | items | summary |
|---|---|---|---|---|
| kalshi_direct | OK | 1.0s | 1 | ## This Market's Kalshi Data **Before 2028** (KXPNP-32-28JAN01) - Current price (probability): 7.00% (live; bid 5% / ask 10%) - Last daily close (2026-10-29): 16.00% - 7-day price change: -6.00% - 30-day price change: -6.00% - Average daily volume: 57 contracts - Price range: 13.00% - 71.00% - Data |
| resolution_source | OK | 5.7s | 1 | already_resolved=unclear; matches_criteria=True; last_updated=unknown (page footer © 2026). The official Clay Mathematics Institute P vs NP page (© 2026) labels the problem "Unsolved". It says "so far no one has managed to prove" that these problems are as hard as they appear. The page lists no proo |
| kalshi_related | OK | 8.8s | 2 | 2 related markets / summaries. series KXPNP: 0 markets (skipped 6 no-signal) | keyword 'millennium prize': ok | keyword 'conjecture proven': ok |
| polymarket_related | OK | 0.4s | 10 | Polymarket search over 3 keyword(s), kept 10 matches. keyword 'P vs NP': 0 markets | keyword 'millennium prize': 5 markets | keyword 'Riemann hypothesis': 5 markets |
| wikipedia | OK | 0.2s | 2 | Fetched 2 Wikipedia entries (0 missing pages). |
web research agents
4/4 ok,
17 searches,
6 page reads
| dimension | status | time | searches | page reads | report |
|---|---|---|---|---|---|
| current state | OK | 24s | 5 | 2 | 4623 chars |
| base rate | OK | 21s | 4 | 0 | 4308 chars |
| yes case | OK | 22s | 4 | 2 | 5600 chars |
| no case | OK | 27s | 4 | 2 | 6542 chars |
report: current state
- [2026-11-06] No resolution of P vs NP has been found. Wolfram MathWorld's table, dated to Sep 11, 2026, shows CMI listing P vs NP as "unsolved". Wikipedia's Millennium Prize page says only Poincaré has been officially declared solved. https://mathworld.wolfram.com/MillenniumPrizeProblems.html
- [2026-10-26/27] OpenAI and Anthropic each posted AI-generated papers claiming a subexponential-time (2^o(n)) algorithm for 3-SAT. They say this refutes the Exponential Time Hypothesis (ETH). Both say it was Lean-checked. https://www.scientificamerican.com/article/dueling-proofs-from-openai-and-anthropic-astound-scientists/
- [2026-10-27] The ETH result is not a resolution of P vs NP. OpenAI's paper says it "does not bear on the P vs NP question directly". Anthropic says it does not directly address P=NP. The algorithms would not beat brute force until about 10^10,000 variables. https://www.webpronews.com/openai-anthropic-exponential-time-hypothesis-refuted-sat/ ; https://dataconomy.com/2026/11/04/openai-and-anthropic-claim-ai-disproved-the-exponential/
- [2026-11-04] The ETH claims are still unverified. Review is led by Impagliazzo and Paturi (UCSD) and Ryan Williams (MIT). They say their role does not certify the result. Neither paper is peer-reviewed or in a journal. https://dataconomy.com/2026/11/04/openai-and-anthropic-claim-ai-disproved-the-exponential/
- [2026-10-27] A Manifold play-money market, "Does P = NP?", was at about 12% YES after the ETH news. It resolves on a "widely-accepted proof" with no deadline. Per tech-insider, it traded about 85% of its lifetime volume in a day. This is a sibling venue and a different contract, so it is a weak guide. https://tech-insider.org/does-p-np-manifold-market-odds-volume-spike-2026/
- [2026-09-08] OpenAI claimed a Navier–Stokes Millennium solution, with a Lean proof. CMI wrote that it "shares in the excitement". Wikipedia says the claim awaits validation. The Quanta article says a "forced" variant falls short of the Millennium criteria. This is a precedent for AI claims, not P vs NP. https://openai.com/index/navier-stokes-solution/ ; https://www.quantamagazine.org/ai-has-solved-one-of-maths-1-million-millennium-prize-problems-20260908/
- [2026-11-06] CMI policy, as reported on Wikipedia: CMI takes no direct submissions. A solution must be published in a qualifying outlet, wait at least two years, and gain general acceptance. So a CMI prize before 2028 is unlikely, but the contract's other triggers (peer-reviewed publication, an authoritative body's recognition, or two Source Agencies reporting it as proven) do not depend on the prize. https://en.wikipedia.org/wiki/Millennium_Prize_Problems
- [2026-06/08] Fringe proof claims exist. Arthanari's arXiv 2606.03194 (2026-06-02) claims a Lean-verified P=NP, but its repository declares key results as axioms. A July package claims P≠NP with two axioms. ProofAtlas (accessed Aug 7) found no accepted machine-checked resolution. https://www.proofatlas.ai/collaboration/p-vs-np/
- [2026-06-10] Lance Fortnow wrote that he does not expect a P vs NP proof "in my lifetime". He wrote that AI cannot yet produce even trivial full Lean proofs. https://blog.computationalcomplexity.org/2026/06/respect-p-v-np-problem.html
- [2026-10-03] Cairn Commons lists the status as open, with no accepted proof either way. https://cairn-commons.com/problems/p-vs-np
- [Undated; likely Sep–Nov 2026] The Kalshi series KXPNP-32 opened Sep 12, 2026. A bitcoin.com mirror showed the "before 2028" strike at 7%, with only $11.6K volume. An earlier cached copy showed 16%. Both snapshots are likely stale. The mirror's rules text says "before Jan 1, 2027", with close on Jan 7, 2027. That is inconsistent with this question's 2028 wording, so check the live rules. https://www.bitcoin.com/prediction-markets/kalshi/when-will-p-vs-np-be-resolved--KXPNP-32/
- [2026-11-02] Manifold's "another Millennium problem claimed in 2026" market was at 27%. This is a different question. https://manifold.markets/HumanClanker/will-another-millennium-prize-probl
BOTTOM LINE: As of Nov 6, 2026, P vs NP is open. The recent AI news (ETH refutation) is a weaker-than-P=NP result, and the experts reviewing it and the labs themselves say it does not address P vs NP. No scheduled event before Jan 2028 is known that would resolve it, though AI-lab claims are arriving at a faster pace.
NOT FOUND: any claim of P vs NP resolution by a credible group; a CMI statement on P vs NP in 2026 (I could not open claymath.org); the live Kalshi price or rules text for the 2028 strike; any announced resolution event for the question window.
report: base rate
- [2026-06-10] Lance Fortnow, a leading complexity theorist, wrote on the Computational Complexity blog that he does not expect a P vs NP proof "in my lifetime", by humans or machines. He also said we lack even a viable approach. https://blog.computationalcomplexity.org/2026/06/respect-p-v-np-problem.html
- [2026-08-07] ProofAtlas reviewed two 2026 Lean-based claims. These were an Arthanari arXiv preprint (2026-06-02) claiming P=NP and a July Lean package claiming P≠NP. Both declare key results as axioms. It found no accepted machine-checked resolution, and the Clay page still lists the problem as unsolved. https://www.proofatlas.ai/collaboration/p-vs-np/
- [2026-09-11] Per MathWorld, Clay lists 5 problems as unsolved, labels Navier-Stokes "Active", and lists Poincaré as solved. P vs NP is "unsolved". https://mathworld.wolfram.com/MillenniumPrizeProblems.html
- [2026-09-11] Clay's rules require publication in a qualifying outlet, general acceptance, and at least two years since publication before it considers a solution. A Clay award before 2028 would therefore need a paper already public by about early 2026. The Kalshi rules are looser than Clay's, because they accept a peer-reviewed article, an authority's recognition, or two news sources. https://mathworld.wolfram.com/MillenniumPrizeProblems.html
- [2010-03-18] Reference class: 1 of 7 Millennium problems solved in 26 years. Poincaré was posed in 1904, Perelman's preprints came in 2002–03, and the Clay award in 2010, about 7 years after the preprints. https://www.claymath.org/millennium/poincare-conjecture/
- [2012] Expert polls by Gasarch: in 2002, 62% thought P vs NP would be resolved by 2100. In 2012 that fell to 53%. Individual predictions ranged from "never" to roughly 2030 to 2050, with several centuries-out answers. https://www.cs.umd.edu/users/gasarch/BLOGPAPERS/pollpaper3.pdf
- [2018] Per Wikipedia, 88% of respondents to Gasarch's third poll believed P≠NP (124 responses). A P≠NP proof is the likelier form of resolution, and the field's view is that it would be a long haul. I did not find a more recent poll with timing data. https://en.wikipedia.org/wiki/P_versus_NP_problem
- [2017-08-14] Failed-claim history: Norbert Blum's 2017 arXiv P≠NP claim is one example of the many such claims, and most are wrong. https://www.goodreads.com/author_blog_posts/15564405-norbert-blum-on-p-versus-np
- [2026-10-27] OpenAI claims an AI-produced Lean-verified disproof of the Exponential Time Hypothesis. This is not P vs NP, and OpenAI's president said P probably still ≠ NP. The claim is awaiting expert review. https://www.mindstudio.ai/blog/exponential-time-hypothesis-disproven
- [2026-10-27] A Manifold "Does P = NP?" market was at about 12% YES per a secondary article. That is a different question, since it asks about the P=NP direction and has no deadline in the snippet. https://tech-insider.org/does-p-np-manifold-market-odds-volume-spike-2026/
- [2026-10-17] Kalshi runs sibling Millennium markets, including "next solved before 2045" lines per problem. The Yang-Mills line was about 20% on thin volume, according to a secondary article. Bitcoin.com's aggregator, about 243 days old, showed Kalshi "When will P vs NP be resolved?" at 49% before 2040 and 39% before 2035. These are stale and thin, and I did not find the before-2028 strike. https://tech-insider.org/yang-mills-solved-next-before-2045-kalshi-odds/ and https://www.bitcoin.com/prediction-markets/
- [2026-10-03] Cairn Commons lists no accepted proof either way, and the relativization, natural-proofs and algebrization barriers still stand. https://cairn-commons.com/problems/p-vs-np
BOTTOM LINE: The outside view is a very low base rate. In 26 years, 0 of 7 Millennium problems besides Poincaré were resolved, and P vs NP has resisted about 55 years. Experts expect it to take decades or longer, and recent AI-assisted claims target adjacent questions or rest on axioms. The main residual risk is a loosely worded Kalshi clause, such as two media outlets reporting a proof, and 2026 AI-math hype raises that risk slightly.
NOT FOUND: Any expert poll or forecast specifically for resolution before 2028. A post-2018 Gasarch poll. The current price of the Kalshi before-2028 strike. Any peer-reviewed or officially recognized P vs NP resolution.
report: yes case
YES means: P vs NP (Clay formulation) is resolved before the deadline by a peer-reviewed proof or disproof, by written recognition from an authoritative body, or by at least two Source Agencies reporting it as proven or disproven from a specific public work, with no retraction or identified uncorrected error. The live rules text may say Jan 1, 2028.
- [2026-10-28] The most likely YES path is an AI-lab release of a Lean-checked P≠NP proof. Lance Fortnow's blog says OpenAI's ETH paper was published in a public repository with a Lean proof. He says Anthropic and others (O'Donnell, Kintali) followed within days with simplified proofs. His post calls the ETH "toast" even though he had not verified it himself. This shows a claimed result can be treated as settled by a leading complexity theorist within about two days, so the third trigger (two Source Agencies plus a public written work) can be met with no journal and no Clay action. https://blog.computationalcomplexity.org/2026/10/the-exponential-time-refutation.html
- [2026-10-28] The ETH result moves toward P vs NP in a limited way. Fortnow's post says it gives a subexponential algorithm for k-SAT and counting, and the fix also yields a new simulation: NTIME(T) ⊆ DTIME(2^{O(T(loglog T)^3/√log T)}). That is a nontrivial deterministic simulation of nondeterminism. It is still far from a separation, and Fortnow says full NP subexponentiality would need 2^{n^{o(1)}} for 3-SAT. https://blog.computationalcomplexity.org/2026/10/the-exponential-time-refutation.html
- [2026-11-04] Ryan Williams, who is reviewing the claims, said theorists should put somewhat more weight on P=NP being plausible, adding that "one's weight may vary". This is a sign that the expert prior has moved slightly. It is not evidence of an imminent resolution. https://dataconomy.com/2026/11/04/openai-and-anthropic-claim-ai-disproved-the-exponential
- [2026-11-04] The pace of AI results is the strongest YES-side mechanism. Thore Husfeldt (ITU Copenhagen) said AI-generated results arrive every week or two and disrupt ongoing projects. Anthropic's Claude found the core 3-SAT algorithm after 14 million tokens with no human input. Neither lab's capability was credibly forecast a year ago. https://dataconomy.com/2026/11/04/openai-and-anthropic-claim-ai-disproved-the-exponential
- [2026-10-28] A Subset Sum algorithm in time 2^{0.49n} was posted by OpenAI on 2026-10-04, per a blog comment. It is a previously unbroken exponent barrier (≈2^{n/2}) cracked within the same month. Fortnow calls it "impressive". This is a second example of an AI lab breaking a classic NP-problem bound within a month. https://blog.computationalcomplexity.org/2026/10/the-exponential-time-refutation.html
- [2026-09-08 to 2026-09-11] The Navier–Stokes precedent supports the loose-trigger path. Clay said the problem was "apparently settled" and that its review would be unhurried. Scientific American said OpenAI "unambiguously" solved the problem under the Clay wording (options C and D, with forcing). Media framed it as a Millennium problem solved, though Clay still lists it as "active". The same pattern for P vs NP would only need two Source Agencies and a public manuscript. https://www.tuftsdaily.com/article/2026/09/mathematicians-still-checking-the-navier-stokes-proof-that-openai-claims-to-have-solved ; https://www.scientificamerican.com/article/did-openai-solve-the-wrong-navier-stokes-problem/
- [2026-09-08] Clay's wording was an exploitable gap in Navier–Stokes. The official problem statement allowed an external force that most researchers consider contrived. P vs NP's Clay statement (Cook's) is a standard formal one, so I found no comparable loophole. This weakens the analogy. https://www.scientificamerican.com/article/did-openai-solve-the-wrong-navier-stokes-problem/
- [2026-11-05] A November dossier written by Claude said P vs NP remains open and has no known proof path. Any proof must be non-relativizing, non-algebrizing and non-natural. This is an AI-written source of uncertain reliability, but it states the barrier list the AI results have not crossed. https://github.com/Todd7777/P-vs-NP-Problem
- [2026-09-12 snapshot] A bitcoin.com mirror of the Kalshi strikes showed before-2028 at 16%, before-2029 at 25%, before-2030 at 32%, and 93–95% for before 2035 and before 2040. Volume was about $8K. The snapshot is undated and likely stale. The curve implies traders saw real mass in the next few years, but it may be thin-market noise. https://www.bitcoin.com/prediction-markets/kalshi/when-will-p-vs-np-be-resolved--KXPNP-32/
- [2026-11-06] Nothing found is scheduled to resolve P vs NP before the deadline. The YES case therefore depends on a surprise claim arriving, with a "complete" public manuscript, two outlets, and no unaddressed error. That last condition is hard because the ETH reviewers say Lean checking does not confirm the statement matches the claim. https://dataconomy.com/2026/11/04/openai-and-anthropic-claim-ai-disproved-the-exponential
BOTTOM LINE: The best YES case is one AI-lab release of a complete P≠NP proof, with Lean verification, that two outlets report as a proof before 2028. The ETH and Subset Sum results and the Navier–Stokes precedent show that route is mechanically possible. But no source shows a credible claim yet, and the ETH result is explicitly not a P vs NP step.
NOT FOUND: any credible claimed P vs NP resolution; any lab statement saying it is working on P vs NP; any 2026 expert poll with a timeline; the live Kalshi price or the 2028-strike rules text; the Clay page (not fetched directly).
report: no case
YES means: P vs NP, in the official Clay formulation, is resolved before the deadline under the contract's triggers. Those are a peer-reviewed proof or disproof, a recognition by an authoritative body, or two Source Agencies reporting it as proven. The resolution must not be retracted before expiry.
- [2026-10-28] Fortnow, who is not a sceptic of AI math, wrote that the OpenAI ETH algorithm applies to k-SAT but "gives nothing new" for Subset Sum or TSP. The reductions from those problems to 3-SAT blow up the variable count too much. Getting subexponential algorithms for all of NP would need 3-SAT in 2^(n^o(1)), which this result "doesn't get close to". The bound is about 2^(n·loglog n/√log n). So the strongest AI result so far is a long way from even a subexponential algorithm for all of NP, let alone polynomial time. https://blog.computationalcomplexity.org/2026/10/the-exponential-time-refutation.html
- [2026-10-28] The same post says the ETH result is checkable by humans: O'Donnell and Kintali posted simplified proofs within days. This does not show AI is near P vs NP, because ETH-type results are algorithmic upper-bound constructions. P≠NP would need a super-polynomial lower bound for all algorithms. That is the barrier-laden side (relativization, natural proofs, algebrization). Other complexity-theory work in 2026 (Range Avoidance, hardness magnification, meta-complexity) is described as not resolving it alone. https://github.com/Todd7777/P-vs-NP-Problem
- [2026-11-05] A Claude-written dossier (not peer-reviewed) says any proof must be non-relativizing, non-algebrizing and non-natural. It marks the status as "open" and says no 2026 frontier result resolves P vs NP alone. It is a weak source, but it shows that even AI-assisted surveys find no current route. https://github.com/Todd7777/P-vs-NP-Problem
- [2026-10-27] Even the one AI complexity claim that did reach the news is still unverified. Review is by Impagliazzo, Paturi and Williams, and Williams said the group is working to see if it holds up. The papers are GitHub preprints, not journal articles. So no AI complexity paper has yet cleared peer review. A P vs NP claim would also need a journal article, or two reputable outlets citing a specific work with no unaddressed error. https://tech-insider.org/does-p-np-manifold-market-odds-volume-spike-2026/
- [2026-09-11] Deadline arithmetic for the Clay route, per MathWorld. On Navier–Stokes, CMI called the problem "apparently" settled but awarded nothing and kept it "Active". The two-year wait means a hypothetical P vs NP paper posted in 2027 could not reach CMI consideration before the deadline. Clay's rules, quoted in an older mirror, also bar direct submission and require a journal "of world-wide repute". The prize route is therefore closed for any new 2027 paper. Only the looser Kalshi triggers remain. https://mathworld.wolfram.com/MillenniumPrizeProblems.html
- [2026-11-06] Date-clause technicality. The bitcoin.com mirror shows the Kalshi series rules as "before Jan 1, 2027" with a close of Jan 7, 2027. It lists the strikes as before 2027, 2028, 2029, 2030, 2035 and 2040. The Jan 2027 wording belongs to the earliest strike. The mirror also shows the before-2028 strike at 7%, with the 2029 strike at 18% and 2030 at 26%, on only $11.6K total volume. The curve implies mass in 2029–2040 that the 2028 contract cannot collect. If the 2028 contract's own rules say "before Jan 1, 2028", any event in Jan 2028 does not count. Check which expiration text the live contract uses. https://www.bitcoin.com/prediction-markets/kalshi/when-will-p-vs-np-be-resolved--KXPNP-32/
- [2026-11-06] The retraction clause favours NO. Any triggering claim must survive to expiry without a retraction, repudiation, or an identified uncorrected error. The two 2026 Lean-based P vs NP claims rest on declared axioms (per the Aug 7 ProofAtlas review). ProofAtlas also says neither was audited or published, so neither changes the open status. A new claim late in 2027 would have little time to be vetted before expiry. The "identified error" clause lets any credible refutation block a media-based resolution. https://www.proofatlas.ai/collaboration/p-vs-np/
- [2026-06-02 to 2026-07-19] Where AI-generated material on P vs NP does appear, it is labelled as unreviewed. A July Zenodo survey, for example, states "internal automated adversarial review; not externally peer reviewed". Volume of such items is growing, which raises noise, not credible resolutions. https://zenodo.org/records/21440860
- [2026-10-03] A tracker for this exact problem lists 0 claims, 0 verified, 0 disputed, 0 refuted, and says a full solution is not expected. The status is open, with no accepted proof either way. https://cairn-commons.com/problems/p-vs-np
- [2026-09-12 onward] The status-quo base rate is strong. The contract opened Sep 12 and has been open about eight weeks. In that time the most celebrated AI-math event, ETH, did not touch P vs NP, and the labs themselves said so. About 14 months remain, and nothing scheduled (a journal issue, a prize ceremony, an IMU or AMS statement) points to a P vs NP resolution in that window. https://blog.computationalcomplexity.org/2026/10/the-exponential-time-refutation.html
- [2026-10-27] The ETH result does not even count as a partial precedent for a media trigger. One report noted that ETH is a stronger statement than P≠NP and that refuting it leaves P vs NP open. A careless "AI disproves" headline could mislead a reader. But the contract requires reporting that describes P vs NP itself, as in its official Clay formulation, as proven or disproven, and the major outlets have said the opposite. https://tech-insider.org/does-p-np-manifold-market-odds-volume-spike-2026/
BOTTOM LINE: Under the literal criteria, every route to YES needs a vetted, unretracted proof of P vs NP by Jan 2028. The only 2026 complexity breakthrough is an upper-bound result far from polynomial time. The Clay two-year wait, the retraction clause and the expert consensus all point to NO. The residual YES risk is a sloppy media or resolution reading, or an unexpected AI-generated proof.
NOT FOUND: The live Kalshi rules text for the 2028 strike (the mirror shows 2027 wording). Any credible P vs NP resolution claim, or any journal acceptance or authority statement on it. Any post-2018 expert poll with timing data. Any scheduled event in the window. I did not read Fortnow's newer posts "Open No More" and "3SUM and APSP and ETH", and I did not open the claymath.org rules page.
3. Forecasters 2 draws, Opus 5.5
claude-opus-5-5, high effort
72% conf
| Yes | 4% |
| No | 96% |
- already resolved?
- The condition has not been met. The Clay page (© 2026) labels P vs NP "Unsolved". MathWorld, Cairn Commons and ProofAtlas list no accepted proof, and the late-October ETH refutation from OpenAI and Anthropic explicitly "does not bear on P vs NP".
- time left
- About 14 months remain before the Jan 1, 2028 deadline.
- if nothing changes
- If nothing changes, the market resolves No. The Clay prize route is effectively closed by the two-year waiting rule, so only a peer-reviewed proof, recognition by an authoritative body, or two Source Agencies reporting an unretracted proof could trigger Yes.
- path to no
- The likely No path is that AI labs keep producing algorithmic upper-bound results (ETH, Subset Sum) that never cross the relativization, natural-proofs and algebrization barriers, while fringe claims resting on axioms keep failing review.
- path to yes
- The Yes path is an AI lab releasing a Lean-checked P≠NP or P=NP proof that a top theorist like Fortnow quickly endorses, two outlets report as a proof, and no identified error surfaces before expiry. The ETH and Navier–Stokes precedents show this route is mechanically possible. Kalshi trades at 7% (bid 5%/ask 10%) on thin volume, down from 16%. Given the rapid AI-math progress in 2026 I sit somewhat below the market, not far below. Expert consensus (Fortnow: "not in my lifetime"), the absence of any credible claim, and the retraction/error clause support pricing this near 4%.
claude-opus-5-5, high effort
72% conf
| Yes | 5% |
| No | 95% |
- already resolved?
- The condition is not met. The Clay page fetched today labels P vs NP "Unsolved", and trackers (Cairn Commons, ProofAtlas, MathWorld) show no credible accepted claim. The only 2026 complexity breakthrough, the AI-lab refutation of the Exponential Time Hypothesis, is explicitly not a P vs NP result. OpenAI, Anthropic and Fortnow all say so, and the reviewers (Impagliazzo, Paturi, Williams) have not yet verified even that claim.
- time left
- About 14 months remain until the Jan 1, 2028 deadline.
- if nothing changes
- The problem stays open and the market resolves No. The Clay prize route is effectively closed by its two-year waiting rule, so YES would need a peer-reviewed proof, written recognition from an authoritative body, or two Source Agencies reporting a specific public proof that survives without an identified error.
- path to no
- AI labs keep producing algorithmic upper-bound results like ETH and Subset Sum, but the barrier-laden lower-bound side (relativization, natural proofs, algebrization) does not fall, or fringe Lean claims that rest on axioms get debunked.
- path to yes
- An AI lab releases a Lean-checked P≠NP proof in 2027, and two outlets report it as solved, as happened quickly with the Navier–Stokes and ETH claims. A loose media trigger could fire before full vetting, provided no error is identified.
The Kalshi anchor is 7% (bid 5 / ask 10) on very thin volume. The unusually fast 2026 AI-math pace and the loose media clause justify more than the near-zero historical base rate. Still, expert consensus and the lack of any proof path argue slightly below the market, so I set 5%.
4. Final plain mean of 2 draws
4%
(72% confidence)
Yes: 4%
No: 96%
- already resolved?
- The condition has not been met. The Clay page (© 2026) labels P vs NP "Unsolved". MathWorld, Cairn Commons and ProofAtlas list no accepted proof, and the late-October ETH refutation from OpenAI and Anthropic explicitly "does not bear on P vs NP".
- time left
- About 14 months remain before the Jan 1, 2028 deadline.
- if nothing changes
- If nothing changes, the market resolves No. The Clay prize route is effectively closed by the two-year waiting rule, so only a peer-reviewed proof, recognition by an authoritative body, or two Source Agencies reporting an unretracted proof could trigger Yes.
- path to no
- The likely No path is that AI labs keep producing algorithmic upper-bound results (ETH, Subset Sum) that never cross the relativization, natural-proofs and algebrization barriers, while fringe claims resting on axioms keep failing review.
- path to yes
- The Yes path is an AI lab releasing a Lean-checked P≠NP or P=NP proof that a top theorist like Fortnow quickly endorses, two outlets report as a proof, and no identified error surfaces before expiry. The ETH and Navier–Stokes precedents show this route is mechanically possible. Kalshi trades at 7% (bid 5%/ask 10%) on thin volume, down from 16%. Given the rapid AI-math progress in 2026 I sit somewhat below the market, not far below. Expert consensus (Fortnow: "not in my lifetime"), the absence of any credible claim, and the retraction/error clause support pricing this near 4%.
Pipeline Timing and Cost
Total pipeline time: 62.3s
LLM cost: $1.09
(Sonnet 5.5 $0.73, Opus 5.5 $0.19, 17 web searches $0.17)
Per-tool research timings shown in the Research section above.