ResearchBenchmarksMathematical Reasoning on MGSM ChineseFollow90.8Pass@1Baseline86.2688.5390.893.07Jun 5, 2026Evaluation ResultsMethodMethodLinksPass@1Average Output TokensBaselineModel=Qwen3-4B-Thinkin...Model=Qwen3-4B-Thinking-25072026.0690.81,235DyConModel=Qwen3-4B-Thinkin...Model=Qwen3-4B-Thinking-25072026.0690.81,053