ResearchBenchmarksModel Editing on ZsRE N=5000 edits (test)Follow99ReliabilityGRACE14.7636.6358.580.37May 2, 2026Evaluation ResultsMethodMethodLinksReliabilityGeneralizationLocalityOverall PerformanceGRACEModel=LLaMA-3.1-8BModel=LLaMA-3.1-8B2026.0599510037HoReNModel=LLaMA-3.1-8BModel=LLaMA-3.1-8B2026.0599929696UltraEditModel=LLaMA-3.1-8BModel=LLaMA-3.1-8B2026.0586854569WISEModel=LLaMA-3.1-8BModel=LLaMA-3.1-8B2026.0540379852AlphaEditModel=LLaMA-3.1-8BModel=LLaMA-3.1-8B2026.051815310