ResearchBenchmarksMathematical Reasoning on DAPO-Math-17k (test)Follow0.627Final Eval RewardRecompute0.435640.485320.5350.58468Dec 6, 2025Evaluation ResultsMethodMethodLinksFinal Eval RewardTraining Time (hours)RecomputeBackbone=Qwen3-8BBackbone=Qwen3-8B2025.120.62716.1LoglinearBackbone=Qwen3-8BBackbone=Qwen3-8B2025.120.62314.54SyncBackbone=Qwen3-8BBackbone=Qwen3-8B2025.120.44326.15