ResearchBenchmarksScience Evaluation on GPQA-D v1 (test)Follow49.9AccuracyPACE35.75639.42843.146.772Feb 12, 2026Evaluation ResultsMethodMethodLinksAccuracyLengthTRPACEModel=DeepSeek-R1-Dist...Model=DeepSeek-R1-Distill-Qwen-7B2026.0249.95,34633.3VanillaModel=DeepSeek-R1-Dist...Model=DeepSeek-R1-Distill-Qwen-7B2026.0249.38,011—PACEModel=DeepSeek-R1-Dist...Model=DeepSeek-R1-Distill-Qwen-1.5B2026.0239.45,88937VanillaModel=DeepSeek-R1-Dist...Model=DeepSeek-R1-Distill-Qwen-1.5B2026.0236.39,342—