Vector benchmark

Name: Open Vector Database Benchmark results
Creator: Infino

The Open Vector Database Benchmark runs Milvus's VectorDBBench on every engine, each on an identical machine, and ranks them by queries per second at the recall you pick. Anyone can submit an engine.

dataset

index

recall ≥

order

Results

Cohere 1M, 768 dimensions, highest QPS at recall ≥ 0.95

engine

version

index

QPS

recall

p99 ms

measured

Infino

0.8.0

resident HNSW, Infino stores vectors as Sq16 internally

7,740.6

0.9586

1.1

2026-09-25

OceanBase

4.4.2.1

HNSW_SQ, M 16, efConstruction 300

5,845.9

0.9611

1.6

2026-09-25

Elasticsearch

9.5.1

int8_hnsw with rescore, oversample 2.0

5,115.2

0.9523

1.9

2026-09-21

Milvus

2.6.14

HNSW_SQ, SQ4U with FP16 refinement, refine_k 2

4,973.5

0.9614

2.0

2026-09-21

pgvector

pgvector 0.8.6 · PostgreSQL 17.11

HNSW on halfvec

2,300.9

0.9602

7.4

2026-09-18

LanceDB

0.39.0

IVF_HNSW_SQ

2,274.4

0.9570

5.8

2026-09-25

pgvecto.rs

vectors 0.4.0 · PostgreSQL 16.8

HNSW, no quantization, vectors 0.4.0

2,128.7

0.9643

3.6

2026-09-25

MariaDB

11.8.9

HNSW, MariaDB stores vectors as binary16 internally

2,106.4

0.9599

8.5

2026-09-18

VectorChord

vchord 0.4.3 · PostgreSQL 17.4

vchordrq

1,155.5

0.9514

13.9

2026-09-25

OpenSearch

3.8.0

HNSW + scalar quantization

1,089.5

0.9678

9.6

2026-09-25

Weaviate

1.28.2

HNSW float32

1,043.3

0.9758

8.6

2026-09-25

Vespa

8.754.14

HNSW float32

640.3

0.9630

4.3

2026-09-25

Redis

7.4.7

HNSW float32

621.6

0.9654

2.4

2026-09-25

Apache Doris

4.1.4-rc04

HNSW ann_index, float32

593.3

0.9734

52.8

2026-09-25

Qdrant

1.19.1

HNSW float32

461.5

0.9696

6.2

2026-09-18

Chroma

1.0.0

HNSW float32

291.6

0.9819

5.9

2026-09-21

CockroachDB

26.2.6

C-SPANN vector index, float32

114.8

0.9749

47.5

2026-09-25

pgvectorscale

vectorscale 0.9.1 · PostgreSQL 17.11

StreamingDiskANN, memory_optimized storage

603.9

0.9465

27.1

2026-09-21

TiDB

8.5.1

HNSW on a TiFlash replica, float32

547.7

0.9013

19.4

2026-09-18

Cohere 10M, 768 dimensions, highest QPS at recall ≥ 0.95

engine

version

index

QPS

recall

p99 ms

measured

Infino

0.8.0

resident HNSW, Infino stores vectors as Sq16 internally

4,209.8

0.9683

2.6

2026-09-25

Milvus

2.6.14

HNSW_SQ, SQ4U with FP16 refinement, refine_k 2

2,516.9

0.9548

1.8

2026-09-22

pgvecto.rs

vectors 0.4.0 · PostgreSQL 16.8

HNSW, no quantization, vectors 0.4.0

1,428.3

0.9663

26.5

2026-09-26

pgvector

pgvector 0.8.6 · PostgreSQL 17.11

HNSW on halfvec

1,334.8

0.9593

17.8

2026-09-25

Qdrant

1.19.1

HNSW float32

267.0

0.9618

7.4

2026-09-22

Chroma

1.0.0

HNSW float32

252.9

0.9664

7.1

2026-09-21

VectorChord

vchord 0.4.3 · PostgreSQL 17.4

vchordrq

155.6

0.9740

92.9

2026-09-25

Apache Doris

4.1.4-rc04

HNSW ann_index, float32

111.1

0.9816

80.4

2026-09-22

Elasticsearch

9.5.1

int8_hnsw with rescore, oversample 2.0

3,959.3

0.9496

2.8

2026-09-21

CockroachDB

26.2.6

C-SPANN vector index, float32

76.5

0.9384

70.0

2026-09-25

TiDB

8.5.1

HNSW on a TiFlash replica, float32

92.7

0.9141

31.6

2026-09-21

Vespa

8.754.14

HNSW float32

668.6

0.9033

5.9

2026-09-25

pgvectorscale

vectorscale 0.9.1 · PostgreSQL 17.11

StreamingDiskANN, memory_optimized storage

429.6

0.7564

50.5

2026-09-21

Vector benchmark

Results

Recall sweep