Vector databases store embeddings and find similar ones fast. They’re the retrieval layer behind every RAG system, AI search engine, and recommendation system.

Here’s how the top four compare in production.

Quick comparison

	Pinecone	Qdrant	Weaviate	Chroma
Type	Fully managed	Open source + cloud	Open source + cloud	Open source
p50 latency (1M vectors)	8ms	6ms	12ms	18ms
Max scale	Billions	Billions	Billions	~5M practical
Hybrid search	✅	✅	✅ Native	❌
Self-host	❌	✅ (Apache 2.0)	✅ (BSD-3)	✅ (Apache 2.0)
Free tier	✅ Serverless	✅ 1GB cloud	✅ Sandbox	✅ (local only)
Best for	Zero-ops production	Performance + cost control	Complex queries	Prototyping

Pinecone — best for zero-ops

You don’t manage servers, indexes, or scaling. Send vectors, query vectors, done. The serverless tier handles burst traffic without pre-provisioning.

Pricing: Serverless starts free (2GB storage). Pay-as-you-go after that. Roughly $0.33/1M reads at scale.

Pick Pinecone when: You want production-ready vector search without any infrastructure work. Your team doesn’t have (or want) a dedicated ops person.

Qdrant — best performance per dollar

Fastest p50 latency at 6ms. Written in Rust with HNSW indexing and product quantization. Apache 2.0 means zero licensing costs for self-hosted.

Pricing: Self-hosted is free. Cloud starts at $0.05/hour (~$36/month).

Pick Qdrant when: You want the best raw performance, you’re comfortable self-hosting, or you need to keep data on your own infrastructure for privacy.

Weaviate — best for hybrid search

The only database with native BM25 + vector hybrid search built into the query engine (not bolted on). Also supports GraphQL API and knowledge graph features.

Pricing: Self-hosted is free. Cloud sandbox is free. Production cloud starts at ~$25/month.

Pick Weaviate when: You need hybrid search (keyword + semantic), your queries are complex, or you want GraphQL.

Chroma — best for prototyping

Install as a Python package, embed documents with three lines of code, query immediately. The fastest path from zero to working vector search.

Pricing: Free (open source, runs locally).

Pick Chroma when: You’re prototyping, learning, or building something with <1M vectors. Migrate to Pinecone or Qdrant when you outgrow it.

# Chroma: 3 lines to working vector search
import chromadb
collection = chromadb.Client().create_collection("docs")
collection.add(documents=["your text here"], ids=["1"])
results = collection.query(query_texts=["search query"], n_results=5)

pgvector — the “just use Postgres” option

If you already run PostgreSQL, pgvector adds vector search without a new database. Performance is good enough for most apps under 10M vectors.

Pick pgvector when: You already use Postgres and don’t want another database to manage. Your vector count is under 10M.

Decision flowchart

• Prototyping? → Chroma
• Already on Postgres? → pgvector
• Want zero ops? → Pinecone
• Want best performance? → Qdrant
• Need hybrid search? → Weaviate
• Need to self-host? → Qdrant or Weaviate

What about scale?

At 1M vectors, all four are fast enough. The differences matter at 10M+:

Vectors	Chroma	pgvector	Weaviate	Qdrant	Pinecone
100K	✅ Fast	✅ Fast	✅ Fast	✅ Fast	✅ Fast
1M	✅ OK	✅ Good	✅ Good	✅ Best	✅ Good
10M	⚠️ Slow	⚠️ Needs tuning	✅ Good	✅ Good	✅ Good
100M+	❌	❌	✅	✅	✅

Related: Embeddings Explained · How to Build an AI Search Engine · RAG vs Fine-Tuning