ResearchBenchmarksReasoning on MMLU-Pro (600 sampled questions)Follow72.27Average @5DelTA63.63865.87968.1270.361May 20, 2026Evaluation ResultsMethodMethodLinksAverage @5Weighted AverageDelTABackbone=Qwen3-14B-BaseBackbone=Qwen3-14B-Base2026.0572.2768.4DAPOBackbone=Qwen3-14B-BaseBackbone=Qwen3-14B-Base2026.0570.866.77DelTABackbone=Qwen3-8B-BaseBackbone=Qwen3-8B-Base2026.0566.4762.38DAPOBackbone=Qwen3-8B-BaseBackbone=Qwen3-8B-Base2026.0563.9758.87