ResearchBenchmarksMathematical Reasoning on AMC23 (Avg@30 and Token Statistics)Follow99.9Avg@30 AccuracyDyCon89.70892.3549597.646Jun 5, 2026Evaluation ResultsMethodMethodLinksAvg@30 AccuracyStd Dev (Avg@30 Accuracy)Token CountStd Dev (Token Count)DyConBackbone=Qwen3-4B-Thin...Backbone=Qwen3-4B-Thinking-25072026.0699.90.469,034381BaselineBackbone=Qwen3-4B-Thin...Backbone=Qwen3-4B-Thinking-25072026.0699.80.6211,145391DyConBackbone=DeepSeek-R1-D...Backbone=DeepSeek-R1-Distill-Qwen-7B2026.0691.22.943,741480BaselineBackbone=DeepSeek-R1-D...Backbone=DeepSeek-R1-Distill-Qwen-7B2026.0690.13.136,243646