ResearchBenchmarksReasoning on Minerva (Total, Accuracy, Spikes, Repairs)Follow272Total ScoreBaseline258.4265.2272278.8Apr 20, 2026Evaluation ResultsMethodMethodLinksTotal ScoreAccuracyAccuracy (%)ImprovementSpikes CountRepairs CountBaselineBase Model=Qwen3-8BBase Model=Qwen3-8B2026.042727929———SPREGBase Model=Qwen3-8BBase Model=Qwen3-8B2026.04272792905,7875,787