Benchmarks

Recall / QPS curves across 6 datasets.

Fashion-MNIST 784GIST 960GloVe 100GloVe 25NYTimes 256SIFT 128

Euclidean · k=10

Fashion-MNIST 784

Recall / Queries per second (1/s)

x-axis recall | y-axis qps log scale

KyroDB
Qdrant
Weaviate
Milvus
pgvector
Redis
Vearch
Vald
Vespa
OpenSearch
10301003001k3k10k27.5k0.00.20.40.60.81.0RecallQueries per second (1/s)

Euclidean · k=10

GIST 960

Recall / Queries per second (1/s)

x-axis recall | y-axis qps log scale

KyroDB
Weaviate
Milvus
pgvector
Vespa
4.610301003001k3k10k14.4k0.00.20.40.60.81.0RecallQueries per second (1/s)

Angular · k=10

GloVe 100

Recall / Queries per second (1/s)

x-axis recall | y-axis qps log scale

KyroDB
Qdrant
Weaviate
Milvus
pgvector
Redis
Vearch
Vald
Vespa
10301003001k3k10k30k43.1k0.00.20.40.60.81.0RecallQueries per second (1/s)

Angular · k=10

GloVe 25

Recall / Queries per second (1/s)

x-axis recall | y-axis qps log scale

KyroDB
Qdrant
Weaviate
Milvus
pgvector
Redis
Vearch
Vald
Vespa
19301003001k3k10k30k59.6k0.20.40.60.81.0RecallQueries per second (1/s)

Angular · k=10

NYTimes 256

Recall / Queries per second (1/s)

x-axis recall | y-axis qps log scale

KyroDB
Qdrant
Weaviate
Milvus
pgvector
Redis
Vearch
Vespa
OpenSearch
21301003001k3k10k30k0.00.20.40.60.81.0RecallQueries per second (1/s)

Euclidean · k=10

SIFT 128

Recall / Queries per second (1/s)

x-axis recall | y-axis qps log scale

KyroDB
Qdrant
Weaviate
Milvus
pgvector
Redis
Vearch
Vald
Vespa
19301003001k3k10k30k0.00.20.40.60.81.0RecallQueries per second (1/s)

Benchmark Follow-Up

Want the configs, workload notes, or a product walkthrough?

We can walk through the exact ANN configuration choices, what is measured here, and how those curves translate into production retrieval latency.

Book a DemoRead the Docs