ResearchBenchmarksModel Editing on ZsRE N=2000 edits (test)Follow99ReliabilityGRACE4760.57487.5May 2, 2026Evaluation ResultsMethodMethodLinksReliabilityGeneralizationLocalityOverall PerformanceGRACEModel=LLaMA-3.1-8BModel=LLaMA-3.1-8B2026.0599510037HoReNModel=LLaMA-3.1-8BModel=LLaMA-3.1-8B2026.0599949897AlphaEditModel=LLaMA-3.1-8BModel=LLaMA-3.1-8B2026.0595886582UltraEditModel=LLaMA-3.1-8BModel=LLaMA-3.1-8B2026.0587855373WISEModel=LLaMA-3.1-8BModel=LLaMA-3.1-8B2026.0549479961