ResearchBenchmarksMathematical Reasoning on GSM8K (Spearman Rho Correlation & Resolution Gain)Follow1Spearman Rho (Agreement)FullPair-0.6328-0.20890.2150.6389May 28, 2026Evaluation ResultsMethodMethodLinksSpearman Rho (Agreement)Resolution GainFullPairProtocol=FullPairProtocol=FullPair2026.05130.07SEALProtocol=SEALProtocol=SEAL2026.050.8310.5PointwiseProtocol=PointwiseProtocol=Pointwise2026.050.815.29ListwiseProtocol=ListwiseProtocol=Listwise2026.050.8128.64FlatBrkProtocol=FlatBrkProtocol=FlatBrk2026.050.579.87FixedRubProtocol=FixedRubProtocol=FixedRub2026.050.265.72OriginalProtocol=OriginalProtocol=Original2026.05-0.571