ResearchBenchmarksReasoning on GPQA (Accuracy, Avg Entropy)Follow23.89Accuracy (GPQA)Epsilon20.87421.65722.4423.223Mar 15, 2026Evaluation ResultsMethodMethodLinksAccuracy (GPQA)Avg Entropy (GPQA)Epsilon2026.0323.890.1712Top-k2026.0323.330.1815Top-b2026.0322.850.1579Min-p2026.0321.990.1714Top-p2026.0321.320.1764Temperature2026.0321.210.1774Eta2026.0320.990.1592