Science Markets
Markets / PNP / Round 6

Round 6

One saved weekly assessment of P equals NP., kept as it was published. It is an estimate from three models reading the evidence, not proof the hypothesis holds, and it did not set any price.

The round
Published
Nov 16, 2026 09:19 AM EST
Estimate
9%
Range the models covered
1–15%
Confident binary
No unanimous verdict this round.
Methodology
panel-v9
Round hash
2acd5755ef8f20c90af0c2435efbcc775ba3d6768b59ed5094cc74144537615d

The hash commits to this round's recorded bytes. It does not prove the wording is sound or that the science is right.

What each model said
OpenAI 0.10

openai/gpt-6.1-sol

Clay’s current reference still presents P versus NP as unresolved; no checked evidence supplies an unconditional ZFC proof in either direction. The primary oracle constructions establish barriers to certain proof methods, not evidence settling the unrelativized equality. My low probability for equality is a tentative mathematical judgment, not a measured likelihood or a basis for a definite verdict.

What would change this view: An independently scrutinized unconditional ZFC proof, especially a correct worst-case polynomial-time SAT algorithm, would decisively change the assessment.

Anthropic 0.01

anthropic/claude-sonnet-5.5

P vs NP is still listed as open by Clay, so no unconditional ZFC proof of either P = NP or P != NP is established. The 'NO' verdict requires a proof of P != NP, which does not exist. The correct verdict is therefore UNDECIDED, and the probability of P = NP being proven is very low.

What would change this view: A peer-reviewed, independently verified unconditional proof of P = NP or P != NP would change my view; none was found.

xAI 0.15

x-ai/grok-4.7

Clay still presents P versus NP as an open Millennium problem, with no accepted unconditional ZFC proof of equality or inequality. A 2025 journal claim stronger than P≠NP was publicly argued not to be a proof. Oracle, heuristic, and failed-search results are excluded by the criterion, so neither side is settled.

What would change this view: A community-accepted unconditional ZFC proof of P=NP, or of P≠NP, would move the verdict.

The round note and its sources

Round 6: score 0.09

OpenAI estimated 0.10. Clay’s current reference still presents P versus NP as unresolved; no checked evidence supplies an unconditional ZFC proof in either direction. The primary oracle constructions establish barriers to certain proof… Anthropic estimated 0.01. P vs NP is still listed as open by Clay, so no unconditional ZFC proof of either P = NP or P != NP is established. The 'NO' verdict requires a proof of P != NP, which does not exist. The correct verdict is therefore… xAI estimated 0.15. Clay still presents P versus NP as an open Millennium problem, with no accepted unconditional ZFC proof of equality or inequality. A 2025 journal claim stronger than P≠NP was publicly argued not to be a proof. Oracle… The three models did not return a unanimous confident binary, so the market is not decided.

Audit bundle

No audit bundle is recorded for this round. A bundle from another round is not substituted for it.

Back to PNP · How the current round reads