Recall vs QPS (per config)
“Testing for regressions before public release”
cohere-100m | 2026-09-28 | r8i.8xlarge, Intel(R) Xeon(R) 6975P-C, 32 vCPU, 248 GB
PG 18.6 | shared_buffers=64 GB | work_mem=1 GB | max_parallel_maintenance_workers=16 | pgvector 0.8.6