ResearchTasksTrajectory-level safety evaluationTrajectory-level safety evaluationFollowBenchmarksDatasetSOTA methodSortMost resultsRecently updatedMost papersApplyDatasetSOTA methodMetricTrendResultsLast updatedR-judge (test)Gemini-3-Flash95.2Accuracy32Jun 2, 2026ATBench (test)AgentDoG-Qwen3-4B0.928Accuracy20Feb 26, 2026ASSE-Safety (test)AgentDoG-Llama3.1-8B81.1Accuracy20Feb 26, 2026