ResearchBenchmarksMultimodal Safety Evaluation on GOAT (test)Multimodal Safety Evaluation on GOAT (test)Follow56.9Misogyny AccuracyOSGA49.151.12553.1555.175Jan 30, 2026Evaluation resultsMethodMethodLinksMisogyny AccuracyMisogyny F1 ScoreOffensiveness AccuracyOffensiveness F1 ScoreSarcasm AccuracySarcasm F1 ScoreHarmfulness AccuracyHarmfulness F1 ScoreAverage AccuracyAverage F1 ScoreOSGABase Model=LLaVA-v1.5Base Model=LLaVA-v1.52026.0156.936.7147.5157.6150.3838.368.0247.1255.744.94BaselineBase Model=LLaVA-v1.5Base Model=LLaVA-v1.52026.0149.432.7641.5930.4850.1133.584434.5246.2832.84