ResearchTasksEnd-to-end decode throughputFollowBenchmarksDataset NameSOTA MethodSortMost resultsRecently updatedMost papersApplyDataset NameSOTA methodMetricTrendResultsLast Updatedcontext length 128KLlama-3.1-8B108Throughput (tok/s)6May 20, 2026Context Length 32KLlama-3.1-8B183Decode Throughput (tok/s)6May 20, 20268K Context LengthLlama-3.1-8B268.5Throughput (tok/s)6May 20, 2026