Eight questions about your stack, scale, hosting and team. The result is a ranked list with the reasons behind each score, not a sales pitch.
Answers are preset to a common mid-market profile. Change any of them; the ranking updates immediately.
Each answer adds or subtracts points for each of the five engines. The weights are opinionated and come from our implementation work, not from vendor material. The patterns the scoring encodes:
Qdrant and Weaviate are scored separately but share one product page, because the choice between them usually comes down to hybrid search style and the vectorise-on-ingest modules.
Qualitative profiles only. We publish no benchmark numbers because they depend on hardware, index settings and data that differ from yours; a vector database is chosen on fit, then proven on a sample of your own vectors.
An extension that adds vector columns and indexes to the database you already run. Transactions, joins and row-level security come with it. Best under a few million vectors; scaling beyond a single node is where it stops being simple.
pgvectorOpen-source, distributed, with storage and compute separated so it can grow to very large collections. Needs real operations skill to run as a cluster, or Zilliz Cloud to have it run for you. We run it in production.
MilvusFully managed, serverless, nothing to operate. The quickest path to a working index, in exchange for no self-hosting, region constraints and a usage-based bill that needs modelling at volume.
PineconeOpen-source engine where metadata filtering is part of the index design rather than an afterthought, with quantisation to cut memory and sparse vectors for hybrid search. Light to operate; managed cloud available.
Qdrant & WeaviateOpen-source engine with BM25 and hybrid search built in, modules that can vectorise content on ingest, and native multi-tenancy. Managed cloud available; comfortable on Kubernetes for self-hosting.
Qdrant & WeaviateEvery open-source engine here can be self-hosted or bought as a managed cloud. The chooser scores both; our guide on on-premises vs cloud covers the wider trade-off for AI systems.
Full comparison guideThe full qualitative comparison table and decision tree.
ReadSize the index and the monthly bill before you commit.
EstimateLong-form guide for decision makers.
ReadResidency, retention and access controls for AI systems.
ReadGood enough to narrow five options to two. The scoring encodes the trade-offs we apply in scoping calls, and the reasons are shown so you can disagree with any of them. A final choice still needs a short proof of concept on your data; see vector database selection for the longer comparison.
Because the cheapest database to run is the one you already operate. Under a few million vectors, PostgreSQL with pgvector gives you transactions, joins, backups and roles for free. The engines built purely for vectors earn their place at larger scale or with heavier filtering needs; the pgvector page explains its limits.
The weights follow the workload, not a vendor. We run Milvus and pgvector in our own systems and implement Pinecone, Qdrant and Weaviate for clients, so the scoring reflects what we have seen work in production rather than a commercial preference.
That usually means the decision will be made by factors the questions cannot capture: team familiarity, an existing cloud contract, or a specific feature such as vectorise-on-ingest. Read the two product pages, then ask us for a view; it is a short conversation.
Tell us about your systems, your data and your constraints. We reply within one business day with a straight answer and a suggested next step.