Quick answer
Chroma is the simplest, ideal for a first RAG app and small collections. Qdrant is a fast Rust engine with strong filtering and quantisation, with a free cloud cluster. Pinecone is the fully managed option with the least operations and the most predictable enterprise story. Weaviate is the most feature-rich open-source choice, with built-in hybrid search and modules for embedding. All four have free tiers.
A vector database stores embeddings and finds the ones closest to a query. That is a narrow job, which is why the differences between products are less about search quality and more about how easy they are to run, how well they filter, and what they cost once your collection has a few million entries.
Chroma
Chroma runs inside your Python or JavaScript process with a handful of lines, or as a server, and Chroma Cloud offers a serverless managed version billed on usage. It is what most tutorials use, and for collections up to the low millions it is perfectly adequate. It has metadata filtering and full-text search, but its hybrid search and enterprise features trail the others.
Qdrant
Qdrant is written in Rust and built for throughput. Its payload filtering is expressive, it supports sparse vectors for hybrid search, and its quantisation options can cut memory by a large factor, which translates directly into lower hosting costs. Qdrant Cloud includes a permanent free 1 GB cluster, which is enough for a real small application, not just a demo.
Pinecone
Pinecone is managed only; there is nothing to install or scale. Serverless indexes bill on storage and reads, and the platform handles sharding and replication invisibly. It has sparse-dense indexes for hybrid search, integrated embedding and reranking, and the compliance certifications enterprises ask for. The trade-offs are cost at high query volumes and having no self-hosted option.
Weaviate
Weaviate is open source with native hybrid search, a schema that supports multiple vectors per object, and modules that call embedding models for you on insert. Weaviate Cloud offers managed clusters, and there are serverless and enterprise tiers. It is the most complete open option and the one with the steepest learning curve.
How to choose
- Prototype or small app: Chroma, embedded, today
- Production with complex filters and cost pressure: Qdrant, self-hosted or cloud
- No ops team and enterprise procurement: Pinecone
- Hybrid search and rich schemas in open source: Weaviate
- Already on Postgres with modest scale: pgvector before any of the above
Bottom line
Start with Chroma, move to Qdrant or Weaviate when you need scale and filters in open source, or to Pinecone when you want to pay for zero operations. Retrieval quality depends far more on your chunking, embedding model, and reranker than on which of these four holds the vectors.
