← all runs
Cohere-1M — 2026-08-26
m8g.large · peak recall 0.9972 @ 36 QPS
“cohere-1m warm reference, builds the indexes”
QPS at recall
| Recall | Meerkat | pgvector HNSW (halfvec) | Δ vs best competitor |
|---|
| 0.90 | 2,009 | 928 | +116% |
| 0.95 | 1,080 | 757 | +43% |
| 0.99 | 340 | 280 | +21% |
Latency at recall
| Meerkat | pgvector HNSW (halfvec) | Δ vs best competitor |
|---|
| @ recall 0.90 | | | |
| p50 | 0.49 ms | 0.96 ms | -49% |
| p95 | 0.63 ms | 1.86 ms | -66% |
| p99 | 0.78 ms | 2.60 ms | -70% |
| @ recall 0.95 | | | |
| p50 | 0.92 ms | 1.30 ms | -29% |
| p95 | 1.10 ms | 1.87 ms | -41% |
| p99 | 1.17 ms | 2.21 ms | -47% |
| @ recall 0.99 | | | |
| p50 | 2.94 ms | 3.59 ms | -18% |
| p95 | 3.41 ms | 4.83 ms | -29% |
| p99 | 3.54 ms | 5.09 ms | -30% |
All values are taken at exactly the stated recall: each configuration's own sweep is log-interpolated between its two surrounding measured points (measured values are used verbatim on an exact hit) and the best configuration at that recall is shown — values are never interpolated across different configurations. “—” means no single sweep spans that recall; nothing is extrapolated. Recall is distance-based (ann-benchmarks, tie-aware). Insert/update/delete workloads are not measured.
Interpolated on config — Meerkat @0.90: mkt-defaults-rerank-sweep · pgvector HNSW (halfvec) @0.90: hnsw-m24-halfvec · Meerkat @0.95: mkt-defaults-rerank-sweep · pgvector HNSW (halfvec) @0.95: hnsw-m24-halfvec · Meerkat @0.99: mkt-defaults-rerank-sweep · pgvector HNSW (halfvec) @0.99: hnsw-m24-halfvec
Build comparison
| Meerkat | pgvector HNSW (halfvec) | Δ vs best competitor |
|---|
| Build time | 244 s | 1367 s | -82% |
| Index size | 266.4 MB | 1.9 GB | -86% |
Build stats are for the configuration backing each column's best operating point: Meerkat = mkt-defaults-rerank-sweep, prewarmed · pgvector HNSW (halfvec) = hnsw-m24-halfvec, prewarmed.
Plots
Pareto svg ·
Distance-recall svg ·
Per-config svg ·
Build times svg ·
Index sizes svgHost
| Instance type | m8g.large |
| Hostname | ip-172-31-0-116 |
| CPU | unknown (2 vCPU) |
| Memory | 8 GB |
| OS | Ubuntu 26.04 LTS |
| Kernel | 7.0.0-1011-aws |
Versions
| rekall | 0.1.0 (739ae7a / main) |
| PostgreSQL | PostgreSQL 18.6 (Ubuntu 18.6-1.pgdg26.04+2) on aarch64-unknown-linux-gnu, compiled by gcc (Ubuntu 15.2.0-16ubuntu1) 15.2.0, 64-bit |
| meerkat | 0.1.0 (12d8109) |
| pg_prewarm | 1.2 |
| vector | 0.8.6 |
Search
Index builds
| Index | Type | Build time | Size | Workers |
|---|
| hnsw-halfvec | pgvector-hnsw | 988.4s | 1.9 GB | 1 |
| hnsw-m24-halfvec | pgvector-hnsw | 1367.5s | 1.9 GB | 1 |
| mkt-defaults-rerank-sweep | meerkat | 243.7s | 266.4 MB | 1 |
Index parameters
hnsw-halfvec: {"ef_construction":200,"encoding":"halfvec","m":16}
hnsw-m24-halfvec: {"ef_construction":200,"encoding":"halfvec","m":24}
mkt-defaults-rerank-sweep: {"boundary_epsilon":null,"centroid_compression":null,"centroid_fastscan":null,"fan_out":null,"fastscan":null,"fastscan_bits":null,"nlist":null,"rerank_vectors":null,"soar_lambda":null}
PostgreSQL settings
effective_cache_size = 6 GB
effective_io_concurrency = 200
huge_pages = try
maintenance_work_mem = 3 GB
max_connections = 100
max_parallel_maintenance_workers = 1
max_parallel_workers = 2
random_page_cost = 1.1
shared_buffers = 2 GB
wal_level = replica
work_mem = 64 MB
