ResearchTasksServing LatencyServing LatencyFollowBenchmarksDatasetSOTA methodSortMost resultsRecently updatedMost papersApplyDatasetSOTA methodMetricTrendResultsLast updatedMATH500, GSM8K, SharedGPT, and HumanEval (test)MarginGate9.3Overhead (%)20May 29, 2026