ResearchBenchmarksMath Reasoning on AMC (Token Efficiency Metrics)Math Reasoning on AMC (Token Efficiency Metrics)Follow3.81TauLongSpec0.88761.64632.4053.1637Feb 24, 2025Evaluation resultsMethodMethodLinksTauTokens/sImprovement FactorLongSpecModel=QwQ-32B, Maximum...Model=QwQ-32B, Maximum output length=32k tokens2025.023.8145.162.33VanillaModel=QwQ-32B, Maximum...Model=QwQ-32B, Maximum output length=32k tokens2025.02119.41—