Recall vs QPS (Pareto)
“Testing concurrent clients”
cohere-1m | 2026-10-07 | c8i.xlarge, Intel(R) Xeon(R) 6975P-C, 4 vCPU, 8 GB
PG 18.6 | shared_buffers=2 GB | work_mem=64 MB | max_parallel_maintenance_workers=2 | pgvector 0.8.7