Vector databases store high-dimensional embeddings and answer approximate nearest neighbor (ANN) queries — the core operation behind semantic search, RAG pipelines, recommendation systems, and image similarity. The market has 10+ options in 2026, but the real decision tree is simpler than the vendor landscape suggests.
Pinecone is the managed cloud-native choice. Setup takes 10 minutes, there's no infrastructure to operate, and it scales to billions of vectors without configuration. The free tier includes one index with 100K vectors. Paid plans start at $70/month. The downside: it's a black box (closed source), and costs climb fast at scale — 10M vectors with 1536 dimensions runs ~$300/month on the Standard plan. For teams that want to ship quickly and not operate infrastructure, Pinecone is the pragmatic choice.
Qdrant is the best open-source option. It's written in Rust, ships as a single binary or Docker image, and has competitive ANN benchmark performance — 1,400 queries per second at 99% recall on 1M 768-dim vectors on a 4-core machine, per the ann-benchmarks dataset. Qdrant Cloud starts at $25/month managed, or run it yourself on any VPS. Its filtering during search (payload indexing) is notably better than Pinecone for hybrid structured+vector queries.
Weaviate is strong when you want both vector and keyword search in one system, with built-in modules for OpenAI, Cohere, and Hugging Face embeddings. The module system adds operational weight — it's the right choice for teams who want an all-in-one search stack, not a focused vector store.
pgvector is the most underrated option. It's a PostgreSQL extension that adds a vector column type and HNSW/IVFFlat indexes. If you're already on PostgreSQL, adding pgvector means zero new infrastructure. At up to ~1-5M vectors with good indexing, query latency is competitive (2-20ms p99 on a standard instance). Above that, purpose-built vector databases pull ahead on throughput. Neon, Supabase, and AWS RDS all support pgvector.
Chroma is the developer-experience pick for local experimentation — in-process Python, easy LangChain integration, no server required. Not recommended for production workloads above 100K vectors.
The decision tree: if you're already on PostgreSQL and vectors are a secondary concern, start with pgvector. If you need a managed cloud solution fast, Pinecone. If you need open-source with strong filtering and production throughput, Qdrant. If you have a billion+ vectors and a dedicated team, Weaviate or a hosted Qdrant cluster.