What vector database has the best recall at low latency?

There isn’t a single universally “best” vector database for both highest recall and lowest latency—it depends on your data size, dimensionality, filter complexity, update rate, and hardware.

That said, if you want a practical short answer:

• For top low-latency performance with very strong recall:
  Pinecone, Milvus, and Qdrant are commonly strong choices.
• For best raw search performance at scale with tight tuning:
  FAISS (not a database by itself, more a library) is often the benchmark.
• For excellent recall/latency tradeoffs in production with filtering:
  Qdrant is frequently praised.
• For very large-scale, managed, operationally simple setups:
  Pinecone is often favored.

Rule of thumb

• If you want managed + easy + fast: Pinecone
• If you want open source + strong performance + filtering: Qdrant
• If you want open source + huge scale + flexibility: Milvus
• If you want maximum tuning/control and don’t need a DB: FAISS

Important caveat

“Best recall at low latency” is usually a tradeoff between:

• index type (HNSW, IVF, PQ, etc.)
• exact vs approximate search
• filtering
• memory availability
• vector count and dimensionality

If you tell me:

1. how many vectors you have,
2. dimensions,
3. whether you need metadata filtering,
4. read/write ratio,
5. cloud or self-hosted,

…I can recommend the best option for your specific case.