ResearchDatasetsNeedleBenchNeedleBenchFollowBenchmarksTaskDatasetSOTA methodSortMost resultsRecently updatedMost papersApplyTaskDatasetSOTA resultTrendResultsMulti-objective retrievalNeedleBench M-RT (32k context stress slice)100Slot-set Exactness20Long-context retrieval and reasoningNeedleBench91.9Accuracy @ 8k Context5