ResearchBenchmarksMathematical Reasoning on AIME24, MATH500, and Minerva AggregateMathematical Reasoning on AIME24, MATH500, and Minerva AggregateFollow39.8Macro AccuracyVeriGate37.09637.79838.539.202May 28, 2026Evaluation resultsMethodMethodLinksMacro AccuracyVeriGateBase Model=Qwen2.5-Mat...Base Model=Qwen2.5-Math-1.5B2026.0539.8TreeRPOBase Model=Qwen2.5-Mat...Base Model=Qwen2.5-Math-1.5B2026.0537.2