ResearchTasksInference Performance EvaluationFollowBenchmarksDataset NameSOTA MethodSortMost resultsRecently updatedMost papersApplyDataset NameSOTA methodMetricTrendResultsLast UpdatedGPU timing experiment 2K-8K context lengthsSpectrumKV116TTFT16Jun 9, 20261,424 simulated conversations on AWS EC2 P5 (test)π_θ^EAGLE+FP80.92Latency (s)4Jun 18, 2026