| Provider | Starting $/mo | ~1M vectors $/mo ▼ | Pricing model | Free tier | Open / self-host | Trustpilot | Notes |
|---|---|---|---|---|---|---|---|
| Loading… | |||||||
What to look at in vector database pricing?
Pricing model matters more than list price. Serverless (Pinecone, Weaviate, Chroma) is cheap at low usage but per-query fees add up with heavy search; pod/hour (Qdrant, MongoDB Atlas, Redis) is predictable but you pay for reserved capacity even when idle. 1M 768-dimensional vectors is only ~3 GB — that fits many free tiers or cheap Postgres plans. The ⚠︎ mark means an estimate that could not be verified from a fixed price list (calculator-based).
Open source is a strong card: Milvus, Qdrant, Weaviate, Chroma, and pgvector (Postgres extension) are self-hostable — if you already run Postgres, pgvector can be nearly free add-on without a separate service. See also managed databases (managed Postgres), LLM APIs (creating embeddings), and GPU cloud (your own models).
How to choose?
For beginners and small projects the cheapest options are pgvector (Supabase/Neon) and open serverless tiers (Zilliz/Milvus, Chroma) — a few dollars per month or free. If you need managed vector search without DevOps, Pinecone and Weaviate are mature but their minimum billing ($45–50/mo) is a commitment. MongoDB Atlas and Redis fit if you already use them for other workloads.
Scalability and hidden costs: check egress (Pinecone charges for outbound traffic), per-query fees, and RAM-based pricing (Redis keeps everything in memory → expensive at scale). EU data residency: most have EU regions; Qdrant and Weaviate are also Europe-focused and self-hostable, keeping data fully under your control. Quantization (e.g. int8) reduces memory and cost significantly at large scale.