ResearchTasksSkill SelectionFollowBenchmarksDataset NameSOTA MethodSortMost resultsRecently updatedMost papersApplyDataset NameSOTA methodMetricTrendResultsLast UpdatedProduction AgentIter81F1 Score40Jul 1, 2026ToolBench-60 one-hot preference regime (held-out pool)Freq-as-Override103.3Regret27Jun 5, 2026ToolBench-60 soft-0.3 preference regime (held-out pool)Bandit-as-Override255.9Regret18Jun 5, 2026TOOLBENCH-60 soft-0.3 preference regimeBandit-as-Override264.8Regret9Jun 5, 2026SkillsBench 1000-skill scaleSkillDAG36.8Reward (%)8Jun 4, 2026