ResearchBenchmarksImage-Text Evaluator Robustness (Economic Sensitivity) on COMPOSITEFollow-0.6Median % Change vs NeutralJudge-2.264-1.832-1.4-0.968May 23, 2026Evaluation ResultsMethodMethodLinksMedian % Change vs NeutralJudge2026.05-0.6UMIC2026.05-1.1FLEUR2026.05-1.3CLIPScore2026.05-1.9PAC-S2026.05-2.2