ResearchBenchmarksMathematical Reasoning on 8 Reasoning Benchmarks (GSM8K, MATH500, AIME 24, AIME 25, OlympiadBench) (test)Follow53.3Token SavingsDEER7.95619.72831.543.272Oct 22, 2025Evaluation ResultsMethodMethodLinksToken SavingsDEERModel=DS-R1-Qwen-7BModel=DS-R1-Qwen-7B2025.1053.3DEERModel=Qwen3-4BModel=Qwen3-4B2025.1027.5DiffAdaptModel=Qwen3-4BModel=Qwen3-4B2025.1022.4DiffAdaptModel=ThinkPrune-7BModel=ThinkPrune-7B2025.1010.1DiffAdaptModel=DS-R1-Qwen-7BModel=DS-R1-Qwen-7B2025.109.7