ResearchTasksScaling behavior extrapolationFollowBenchmarksDataset NameSOTA MethodSortMost resultsRecently updatedMost papersApplyDataset NameSOTA methodMetricTrendResultsLast UpdatedCSR benchmarks Downstream & Upstream HellaSwag ARC PIQA WinoGrande OpenBookQA SIQA BoolQ (test)UNSL88.89Best Task Percentage6May 27, 2026Birds 200, Cars 196, ImageNet (test)UNSL60.87Percentage of Tasks Where Best6May 27, 2026
CSR benchmarks Downstream & Upstream HellaSwag ARC PIQA WinoGrande OpenBookQA SIQA BoolQ (test)UNSL88.89Best Task Percentage6May 27, 2026