ResearchBenchmarksHierarchical Instruction Following on IHEval Safety Defense Reference settingFollow78.9Extract AccuracyVanilla45.20453.95262.771.448Jun 29, 2026Evaluation ResultsMethodMethodLinksExtract AccuracyHijack AccuracyAverage AccuracyVanillaBackbone=Llama-3.1-8B-...Backbone=Llama-3.1-8B-Instruct2026.0678.979.179IHDecBackbone=Llama-3.1-8B-...Backbone=Llama-3.1-8B-Instruct2026.0666.778.172.4ISEBackbone=Llama-3.1-8B-...Backbone=Llama-3.1-8B-Instruct2026.0646.544.945.7