ResearchBenchmarksLarge Model Performance Prediction on Paradigm RLHF pattern shiftFollow9.55RMSESTAR9.40610.37811.3512.322Feb 12, 2026Evaluation ResultsMethodMethodLinksRMSEMAEAverage ScoreSRCCKRCCMAE@3%Average RankTotal ScoreSTARShift Type=Model-side,...Shift Type=Model-side, Scenario=Paradigm (RLHF)2026.029.557.338.4492.4884.0739.3771.97—CPMFShift Type=Model-side,...Shift Type=Model-side, Scenario=Paradigm (RLHF)2026.0210.147.949.0491.8883.2636.3570.5—PMFShift Type=Model-side,...Shift Type=Model-side, Scenario=Paradigm (RLHF)2026.0213.1510.3211.7488.5179.5424.0264.02—