ResearchBenchmarksRegion-level faithfulness analysis on DeFacto 1.5K (test)Follow21.5Mislocalized & Wrong RateDeFacto20.6226.5632.538.44Sep 25, 2025Evaluation ResultsMethodMethodLinksMislocalized & Wrong RateSpurious Correct RateFaithful but Wrong RateMasked Evidence Abstention RateImage Replacement Abstention RateDeFacto2025.0921.55.61.364.161.4DeepEyes2025.0923.5111.833.140.3GRIT2025.0943.511.630.146.146.3