ResearchTasksHarmful-request assistanceHarmful-request assistanceFollowBenchmarksDatasetSOTA methodSortMost resultsRecently updatedMost papersApplyDatasetSOTA methodMetricTrendResultsLast updatedSORRY-Bench linguistic mutationsQwen3.5-27B25.4Compliance Rate6Jun 9, 2026SORRY-Bench decontaminated base promptsQwen3.5-27B17.1Compliance Rate6Jun 9, 2026