Question 1

What is the cheapest hosted vector database in 2026?

Accepted Answer

Turbopuffer is the cheapest at-scale managed vector DB by storing vectors on object storage (S3/GCS) and caching hot data in memory. Pricing is pure usage-based: $0.30 per 1M reads, $0.10 per 1M writes, $0.20 per GB-month. For small workloads pgvector on Supabase or Neon free tiers is effectively free.

Question 2

Which vector database is best for hybrid search (BM25 + vector)?

Accepted Answer

Vespa is the strongest hybrid retrieval engine in the catalog (steepest learning curve too). Weaviate and Elasticsearch are close runners-up with first-class hybrid scoring (RRF). Qdrant has native sparse-vector support which covers most hybrid use cases. Chroma and most embedded options do not support hybrid search natively.

Question 3

What is the difference between managed, oss, and hybrid in this catalog?

Accepted Answer

Managed = SaaS only (Pinecone, Turbopuffer). OSS = open source self-host only (pgvector, OpenSearch). Hybrid = both, with feature parity between cloud and self-host (Qdrant, Weaviate, Milvus, Chroma, LanceDB, Vespa). The hybrid category is usually the safest pick if you might need to migrate or run an air-gapped instance.

Question 4

Is pgvector good enough for production RAG?

Accepted Answer

Yes for small-to-medium workloads (under 10M vectors per index, sub-100ms p99 latency target). Beyond that you start hitting Postgres scaling ceilings: ingestion throughput on large indexes, query latency under high write load, and connection pool contention. Switch to a dedicated vector DB when those become real constraints, not before.

Vector Databases