Where embeddings live, and the layer every retrieval system is built on. · ranked by adoption, then measured latency · verified 13 Aug 2026

#	Name	Index	Hybrid search	Hosting	Licence	p50 latency	Status
1	pgvector PostgreSQL	HNSW, IVFFlat	Yes, with SQL	Self-host or any managed Postgres	PostgreSQL	~15ms	current
2	Qdrant	HNSW	Yes	Self-host and managed	Apache 2.0	4ms	current
3	Pinecone	Proprietary	Yes	Managed only	Proprietary	<10ms	current
4	Milvus Zilliz	HNSW, IVF, DiskANN, GPU	Yes	Self-host and managed	Apache 2.0	~10ms	current
5	Weaviate	HNSW	Yes, native	Self-host and managed	BSD-3	~12ms	current
6	Chroma	HNSW	Limited	Embedded, self-host, managed	Apache 2.0	Varies	current
7	turbopuffer	Proprietary, object storage	Yes	Managed only	Proprietary	~20ms	current
8	LanceDB	IVF-PQ, HNSW	Yes	Embedded and managed	Apache 2.0	Varies	current
9	Vespa Vespa.ai	HNSW plus inverted	Yes, native	Self-host and managed	Apache 2.0	~10ms	current
10	Redis	HNSW, FLAT	Yes	Self-host and managed	RSALv2 / SSPL	5ms	current
11	Atlas Vector Search MongoDB	HNSW	Yes	Managed	Proprietary	~20ms	current
12	Vertex AI Vector Search Google	ScaNN	Yes	Managed	Proprietary	~15ms	current