ResearchBenchmarksMathematical Reasoning on AMC (Acc, Tok)Follow97.5AccuracyVanilla71.578.258591.75Apr 8, 2026Evaluation ResultsMethodMethodLinksAccuracyToken CountVanillaBackbone=Qwen3-14BBackbone=Qwen3-14B2026.0497.56,562DTSRBackbone=Qwen3-14BBackbone=Qwen3-14B2026.0497.53,975TALEBackbone=Qwen3-14BBackbone=Qwen3-14B2026.04956,623CoDBackbone=Qwen3-14BBackbone=Qwen3-14B2026.0472.54,122