ResearchBenchmarksScientific Reasoning on GPQA (Total, Accuracy, Improvement, Spikes, Repairs)Follow198Total ScoreBaseline188.1193.05198202.95Apr 20, 2026Evaluation ResultsMethodMethodLinksTotal ScoreAccuracy (0-100 Scale)Accuracy (0-1 Scale)Improvement (%)Spikes CountRepairs CountBaselineBase Model=Qwen3-8BBase Model=Qwen3-8B2026.04198980.495———SPREGBase Model=Qwen3-8BBase Model=Qwen3-8B2026.04198960.485-19,4569,456