Research updates

Live panel did not return all three estimates (OpenAI, Anthropic, xAI failed or gave an unusable answer). A round is only live when all three are valid, so this round is a placeholder at the previous score rather than an average of two.

Live panel did not return all three estimates (OpenAI, Anthropic, xAI failed or gave an unusable answer). A round is only live when all three are valid, so this round is a placeholder at the previous score rather than an average of two.

Live panel did not return all three estimates (OpenAI, Anthropic, xAI failed or gave an unusable answer). A round is only live when all three are valid, so this round is a placeholder at the previous score rather than an average of two.

Live panel did not return all three estimates (OpenAI, Anthropic, xAI failed or gave an unusable answer). A round is only live when all three are valid, so this round is a placeholder at the previous score rather than an average of two.

Live panel did not return all three estimates (OpenAI, Anthropic, xAI failed or gave an unusable answer). A round is only live when all three are valid, so this round is a placeholder at the previous score rather than an average of two.

Live panel did not return all three estimates (OpenAI, Anthropic, xAI failed or gave an unusable answer). A round is only live when all three are valid, so this round is a placeholder at the previous score rather than an average of two.

Live panel did not return all three estimates (OpenAI, Anthropic, xAI failed or gave an unusable answer). A round is only live when all three are valid, so this round is a placeholder at the previous score rather than an average of two.

Live panel did not return all three estimates (OpenAI, Anthropic, xAI failed or gave an unusable answer). A round is only live when all three are valid, so this round is a placeholder at the previous score rather than an average of two.

Live panel did not return all three estimates (OpenAI, Anthropic, xAI failed or gave an unusable answer). A round is only live when all three are valid, so this round is a placeholder at the previous score rather than an average of two.

Live panel did not return all three estimates (OpenAI, Anthropic, xAI failed or gave an unusable answer). A round is only live when all three are valid, so this round is a placeholder at the previous score rather than an average of two.

Live panel did not return all three estimates (OpenAI, Anthropic, xAI failed or gave an unusable answer). A round is only live when all three are valid, so this round is a placeholder at the previous score rather than an average of two.

OpenAI estimated 0.50. I interpret the claim strictly as unforced flow on R³. The September 2026 announcement describes forced Navier–Stokes blow-up, which does not contradict this hypothesis, even if independently verified. Neither the… Anthropic estimated 0.03. This is the unforced Millennium problem on R^3, and I found no accepted proof or disproof. OpenAI's recent claim is blow-up with a contrived external force, which this criterion excludes. Follow-up work reportedly shows… xAI estimated 0.55. The Clay problem on R^3 remains unverified: the 11 Sep 2026 CMI note says it was “apparently” settled but evaluation is unfinished and does not state proof versus blow-up. No independently checked ZFC proof for every… The panel disagreed by 0.52, so read this average as soft. The three models did not return a unanimous confident binary, so the market is not decided.

OpenAI estimated 0.78. You et al.’s ultrafast scattering and heat-capacity signatures support a critical region, while molecular simulations independently support plausibility within the specified domain. However, the retrieved primary… Anthropic estimated 0.42. You et al. (Science, March 2026) report ultrafast-heating x-ray evidence of a liquid-liquid critical point near 210±8 K. It is a single peer-reviewed experiment, and the criterion requires an independent reproduction.… xAI estimated 0.63. You et al. (Science 2026) report critical-like fluctuations and a heat-capacity rise at 210±8 K before ice forms, and accurate simulations place an LLCP near 198 K and ~127 MPa. That supports the claim but does not… The panel disagreed by 0.36, so read this average as soft. The three models did not return a unanimous confident binary, so the market is not decided.

OpenAI estimated 0.45. I interpret “can” as chemical feasibility under the specified conditions, not merely partial copying. Primary results support important components, but QT45 self-synthesis required a prohibited hexamer and achieved only… Anthropic estimated 0.01. The evidence does not show the required ten-transfer, two-lab, isotope-traced self-replication. QT45 copies itself at about 0.2% yield over 72 days. The freeze–thaw work replicates only fragments, and its dilution… xAI estimated 0.28. Named primary papers show trinucleotide-fed polymerase activity and partial exponential copying, but not the required integrated result. QT45 copies itself and its complement only in separate low-yield reactions… The panel disagreed by 0.44, so read this average as soft. The three models did not return a unanimous confident binary, so the market is not decided.

OpenAI estimated 0.90. I interpret the hypothesis as the universal mathematical statement, not merely its finite verification. Rigorous interval-arithmetic checks support it over a large bounded range, but Clay still lists it as unsolved; the… Anthropic estimated 0.02. I found no accepted proof or disproof of the Riemann hypothesis. Clay still lists it as unsolved. The recent AI-assisted result only raises the proven fraction of zeros on the critical line to about 67%, which is not a… xAI estimated 0.88. Clay still lists the classical Riemann hypothesis as an open Millennium Problem. The first 10^13 zeros lie on Re(s)=1/2, but finite checks do not meet the YES criterion, and no independently checked unconditional ZFC… The panel disagreed by 0.88, so read this average as soft. The three models did not return a unanimous confident binary, so the market is not decided.

OpenAI estimated 0.85. I interpret the hypothesis as an unrelativized separation under the stated uniform, total-variation approximation definition. Aaronson–Arkhipov provides substantial but conditional support: approximate-sampling hardness… Anthropic estimated 0.02. An unconditional ZFC separation of SampBQP and SampBPP would imply major open separations such as P≠PSPACE. No such proof exists. Aaronson–Arkhipov and Aaronson–Chen give only conditional hardness (PH noncollapse… xAI estimated 0.60. No unconditional ZFC proof separates or equates SampBQP and SampBPP under the stated uniform total-variation definition. Aaronson’s equivalence reduces the separation to FBQP versus FBPP, which is open, and… The panel disagreed by 0.83, so read this average as soft. The three models did not return a unanimous confident binary, so the market is not decided.

OpenAI estimated 0.10. Clay currently lists P versus NP as unsolved, and the checked evidence supplies no unconditional ZFC proof in either direction. Cook’s description distinguishes restricted lower bounds and conditional barriers from the… Anthropic estimated 0.02. P vs NP remains open. Clay lists it as unsolved, and a 2026 status summary reports no accepted proof either way. The evidence packet shows no ZFC proof of P = NP. Equality is widely disbelieved, so the probability is… xAI estimated 0.18. No unconditional ZFC proof of P=NP or of P≠NP is accepted. Clay still lists the problem open, and Fortnow’s status survey reports it unresolved. Unrefereed Zenodo separation claims are not peer-reviewed proofs and do… The panel disagreed by 0.16, so read this average as soft. The three models did not return a unanimous confident binary, so the market is not decided.

OpenAI estimated 0.25. The strongest verified search found no signal, disfavoring shorter lifetimes but leaving lifetimes between its limit and 10^35 years unresolved. Its published 90% bound cannot satisfy the specified 95% criterion, and no… Anthropic estimated 0.35. The best limit I know of is Super-Kamiokande's p→e+π0 lower bound of about 2.4×10^34 yr at 90% CL (from memory; I did not run a live search). That limit sits below 10^35 yr, so it does not meet the NO criterion of a 95%… xAI estimated 0.22. Super-Kamiokande (PRD 102, 112011) found no p→e+π0 candidates and set only τ/B > 2.4×10^34 yr at 90% CL, short of 10^35 yr and not a measured lifetime. That meets neither a signal with a 95% interval wholly below 10^35… The three models did not return a unanimous confident binary, so the market is not decided.

OpenAI estimated 0.50. I interpret the hypothesis as allowing any nonzero Majorana component, not merely one accessible to present decay searches. The checked primary result supplies a finite limit, not the required independently replicated… Anthropic estimated 0.10. No experiment has seen a neutrinoless double-beta-decay signal. LEGEND-200 (T½ > 1.9×10^26 yr, Ge-76) and the full KamLAND-Zen 800 dataset (Xe-136) both report limits only, so the YES criterion is unmet. The NO… xAI estimated 0.50. No public primary result meets the YES bar of two independent >=5 sigma 0νββ signals with a controlled lepton-number-violating interpretation. PDG 2025 and KamLAND-Zen report half-life limits and no excess, which cannot… The panel disagreed by 0.40, so read this average as soft. The three models did not return a unanimous confident binary, so the market is not decided.

Live panel did not return all three estimates (Anthropic failed or gave an unusable answer). A round is only live when all three are valid, so this round is a placeholder at the previous score rather than an average of two.

OpenAI estimated 0.65. Carbon dating supports appreciable adult neuronal addition, making the threshold plausible, but its model-dependent renewing-fraction turnover is not the specified standardized one-year surviving-addition fraction. The… Anthropic estimated 0.30. The contract needs at least two independent peer-reviewed studies. Each must estimate this exact standardized one-year surviving-addition fraction with an all-neuron denominator and show an interval wholly above 0.001.… xAI estimated 0.38. No eligible study estimates the specified census-standardized one-year surviving addition of mature dentate granule neurons over all neurons. Spalding et al. 2013 reports 1.75% annual turnover only within a renewing… The panel disagreed by 0.35, so read this average as soft. The three models did not return a unanimous confident binary, so the market is not decided.

OpenAI estimated 0.25. I interpret this as the constrained-solid free-energy difference, not stability against melting. Retrieved calculations favor hcp, but the exact-state structure search instead loses the bcc basin, which explicitly… Anthropic estimated 0.04. The best direct ab initio thermodynamic-integration work I found, on melting of bcc and hcp Fe at inner-core conditions (GRL 2022), gives bcc about 10–13 meV/atom above hcp. That is the opposite sign to the hypothesis.… xAI estimated 0.10. Recent ab initio work at 330 GPa finds hcp lower in free energy than bcc, and one study has bcc leave its basin at 6000 K, so the claimed inequality is poorly supported. Eligible 95% intervals that include finite-size… The panel disagreed by 0.21, so read this average as soft. The three models did not return a unanimous confident binary, so the market is not decided.

OpenAI estimated 0.40. Ocean-derived phosphates and freshly ejected organic compounds support habitability, not indigenous cellular life. No retrieved result demonstrates cell-bounded metabolism and serial reproduction, independently… Anthropic estimated 0.01. I found no detection of cells, metabolism or reproduction in Enceladus plume material. No mission has returned a sample, and Cassini flew before any life-detection payload. The YES test is far from met. The NO test… xAI estimated 0.18. Cassini and later analyses establish a subsurface ocean, plume, and complex organics, but no peer-reviewed, independently reproduced observation of cell-bounded metabolism and serial reproduction from traceably… The panel disagreed by 0.39, so read this average as soft. The three models did not return a unanimous confident binary, so the market is not decided.

OpenAI estimated 0.65. The evidence modestly favors sensitivity above 3 K, particularly because actual equilibrium warming can exceed effective sensitivity. However, the retrieved multiline assessments retain substantial uncertainty across 3… Anthropic estimated 0.07. The best published multi-line assessments put the 95% range for ECS well across 3 K. AR6 gives a very likely range of 2–5 K and a best estimate of 3 K, and Sherwood et al. 2020 gives about 2.3–4.7 K at 95%. Neither the… xAI estimated 0.52. IPCC AR6 Chapter 7 places ECS at a best estimate of 3°C with a very likely (5–95%) range of 2–5°C, so the 95% interval crosses 3 K. Sherwood et al. (2020) likewise gives 2.3–4.7 K (5–95%) for effective sensitivity, not… The panel disagreed by 0.58, so read this average as soft. The three models did not return a unanimous confident binary, so the market is not decided.

OpenAI estimated 0.75. I interpret the fraction as cosmological nonbaryonic dark-matter mass, excluding primordial black holes from the numerator. Particle dominance is plausible, but the checked evidence supplies neither independently… Anthropic estimated 0.60. Most cosmologists expect dark matter to be mostly particles beyond the Standard Model, but no non-gravitational particle detection exists. The criterion therefore cannot give YES. A comprehensive bound of 50% or less on… xAI estimated 0.62. PDG 2025 finds Standard Model neutrinos are at most about 1.6% of nonbaryonic matter, so nearly all of that density is unidentified. No non-gravitational detection calibrates a beyond-Standard-Model species above 50%… The panel disagreed by 0.15, so read this average as soft. The three models did not return a unanimous confident binary, so the market is not decided.