Appendix D — Vector Database Comparison Matrix

Last updated: February 2025

---

D.1 Architecture Comparison

System	Language	License	Architecture	Deployment
Pinecone	Unknown	Proprietary	Managed cloud	SaaS only
Weaviate	Go	BSD-3	Monolithic binary	Self-host / Cloud
Qdrant	Rust	Apache-2.0	Segments + Raft	Self-host / Cloud
Milvus	Go/C++	Apache-2.0	Microservices	Self-host / Zilliz Cloud
ChromaDB	Python	Apache-2.0	Embedded / Client-server	Self-host
pgvector	C	PostgreSQL	PostgreSQL extension	Anywhere PG runs
LanceDB	Rust	Apache-2.0	Embedded, columnar	Self-host
Vespa	Java/C++	Apache-2.0	Distributed platform	Self-host / Cloud

---

D.2 Index and Search Features

System	HNSW	IVF	PQ/SQ	Disk Index	Hybrid Search	Filtering
Pinecone	✅	✅	✅	✅	✅ (sparse-dense)	✅
Weaviate	✅	❌	✅ (SQ)	❌	✅ (BM25 + vector)	✅
Qdrant	✅	❌	✅ (SQ + PQ)	✅ (mmap)	✅ (sparse + dense)	✅
Milvus	✅	✅	✅	✅ (DiskANN)	✅	✅
ChromaDB	✅	❌	❌	❌	❌	✅ (basic)
pgvector	✅	✅	❌	❌	✅ (via PG FTS)	✅ (SQL)
LanceDB	✅	✅	✅	✅	✅	✅ (SQL)
Vespa	✅	❌	❌	❌	✅ (native)	✅

---

D.3 Scale and Performance

System	Max Vectors (Tested)	Dimensions	Replicas	Sharding
Pinecone	Billions (managed)	Up to 20K	Automatic	Automatic
Weaviate	~100M per node	Up to 65K	✅	✅ (hash)
Qdrant	~100M per node	Up to 65K	✅ (Raft)	✅
Milvus	Billions (cluster)	Up to 32K	✅	✅ (time-based)
ChromaDB	~1M (embedded)	No limit	❌	❌
pgvector	~10M (practical)	Up to 2000	PG replicas	PG partitioning
LanceDB	~100M	Up to 2048	❌	❌
Vespa	Billions (cluster)	Up to 65K	✅	✅

---

D.4 API and Ecosystem

System	API	SDKs	LangChain	LlamaIndex
Pinecone	REST, gRPC	Python, Node, Go, Java	✅	✅
Weaviate	REST, GraphQL, gRPC	Python, JS, Go, Java	✅	✅
Qdrant	REST, gRPC	Python, JS, Rust, Go	✅	✅
Milvus	gRPC, REST	Python, Java, Go, Node	✅	✅
ChromaDB	Python, REST	Python, JS	✅	✅
pgvector	SQL	Any PG driver	✅	✅
LanceDB	Python, REST	Python, JS	✅	✅
Vespa	REST, Java	Python, Java	✅	✅

---

D.5 Decision Guide

flowchart TD
    A{Already using PostgreSQL?}
    A -->|Yes| B[pgvector]
    A -->|No| C{Scale?}
    C -->|"Under 1M, prototype"| D[ChromaDB or LanceDB]
    C -->|"1-100M, production"| E{Self-host?}
    C -->|"Over 100M, enterprise"| F{Budget?}
    E -->|Yes| G[Qdrant or Weaviate]
    E -->|No| H[Pinecone]
    F -->|Managed| H
    F -->|Self-host| I[Milvus or Vespa]

---

D.6 Pricing Comparison (as of Feb 2025)

Prices are approximate

Managed service pricing changes frequently. Self-hosted costs depend on instance type and provider.

System	Free Tier	Starting Price	For 10M vectors (768-dim)
Pinecone	100K vectors	~$70/month (s1.x1)	~$250/month
Weaviate Cloud	1M vectors (sandbox)	~$25/month	~$300/month
Qdrant Cloud	1M vectors (free)	~$9/month	~$150/month
Zilliz Cloud (Milvus)	100K vectors	~$65/month	~$400/month
pgvector (self-host)	N/A	EC2/RDS cost	~$200/month (r6g.xlarge)