ResearchTasksMaxSim ScoringFollowBenchmarksDataset NameSOTA MethodSortMost resultsRecently updatedMost papersApplyDataset NameSOTA methodMetricTrendResultsLast UpdatedMaxSim Synthetic Workload (Nq=32, d=128)PT Loop867Throughput (docs/s)33Jun 26, 2026L2-normalized FP16 embeddings Nq=32, d=128PLAID GPU0.049Latency (ms)15Jun 26, 2026