ResearchBenchmarksMathematical Reasoning on GSM8K (Detailed Accuracy & Token Costs)Follow93.83Mean AccuracyACTIVE90.855691.627892.493.1722May 7, 2026Evaluation ResultsMethodMethodLinksMean AccuracyAccuracy Standard Deviation (STD)Accuracy Q1 (25th Percentile)Accuracy Worst 25%Token Cost (Selection Phase)Token Cost (Training Phase)Token Cost (Testing Phase)ACTIVESetup=ACTIVESetup=ACTIVE2026.0593.830.7693.1592.931.830.132.36RANDOMSetup=RANDOMSetup=RANDOM2026.0592.9192.1791.69—0.132.47NO TRAINSetup=NO TRAINSetup=NO TRAIN2026.0590.971.3289.8389.56——3.91