ResearchBenchmarksMultiple-choice Question Reasoning on MMLU Pro (Acc, SL, SE)Follow78.57Accuracy (Acc)Cut204849.356456.940764.52572.1093Mar 23, 2026Evaluation ResultsMethodMethodLinksAccuracy (Acc)Score Level (SL)Score Error (SE)Cut2048max tokens=2048max tokens=20482026.0378.572,8392.77ROM_CSCCounterfactual Self-Co...Counterfactual Self-Correction=true2026.0377.11,8184.24EAT2026.0376.72,0523.74Vanilla2026.0376.672,8422.7ROM2026.0375.22,2993.27Cut1024max tokens=1024max tokens=10242026.0372.382,8392.55L12026.0371.41,4015.1Cut512max tokens=512max tokens=5122026.0350.482,8391.78