ResearchBenchmarksMathematical Reasoning on OpenMath-InstructFollow81.022Accuracy (Easy)KRPO10.3373628.6881847.03965.38982May 12, 2025Evaluation ResultsMethodMethodLinksAccuracy (Easy)Accuracy (Normal)Accuracy (Hard)P-ValueKRPO2025.0581.02266.51751.859—GRPO2025.0576.39954.14433.9830.047PPO2025.0566.87139.78327.9830.015Llama3.22025.0513.05612.4239.7170.009