ResearchTasksRefusal behavior analysisFollowBenchmarksDataset NameSOTA MethodSortMost resultsRecently updatedMost papersApplyDataset NameSOTA methodMetricTrendResultsLast UpdatedPaired safety frame contrast items n=20 per model (test)OLMo-3-Instruct (7B)0.043Refusal Delta (eval vs dep)5May 8, 2026
Paired safety frame contrast items n=20 per model (test)OLMo-3-Instruct (7B)0.043Refusal Delta (eval vs dep)5May 8, 2026