ResearchBenchmarksImage-side Objective Verification on Expert-annotated prompt-image setFollow93.4AccuracyGPT-5.485.80887.77989.7591.721May 21, 2026Evaluation ResultsMethodMethodLinksAccuracyF1 ScoreGPT-5.4Model=GPT-5.4Model=GPT-5.42026.0593.490.6GPT-5.2Model=GPT-5.2Model=GPT-5.22026.0591.890.6Gem-3Model=Gem-3Model=Gem-32026.0590.789.6Cl-4.5Model=Cl-4.5Model=Cl-4.52026.0586.186.6