ResearchDatasetsUltra-RealFollowBenchmarksTask NameDataset NameSOTA MethodSortMost resultsRecently updatedMost papersApplyTask NameDataset NameSOTA ResultTrendResultsHelpfulness preference labeling accuracyUltra-Real74.8Benchmark Score15Preference evaluationUltra-Real (target)43Win Rate2Preference evaluationUltra-Real benchmark43.1Win Rate2