ResearchTasksCompositional SteeringFollowBenchmarksDataset NameSOTA MethodSortMost resultsRecently updatedMost papersApplyDataset NameSOTA methodMetricTrendResultsLast Updated3-behavior combinations (unseen split)Hybrid: Composition (<and>) + Instr.62.9Accuracy6Feb 26, 20263-behavior combinations (seen)Hybrid: Composition (<and>) + Instr.87.9Accuracy6Feb 26, 20262-behavior combinations (unseen split)Composition (<and>)76.9Accuracy6Feb 26, 20262-behavior combinations (seen)Hybrid: Composition (<and>) + Instr.0.922Accuracy6Feb 26, 2026