Most "vector database comparison" posts you'll find online were written in 2023, when pgvector was a research curiosity and Pinecone was the default. Two years later, pgvector handles workloads that we'd assumed only Pinecone could touch. We benchmarked both in production conditions on real client data, and the results changed how we recommend vector infrastructure. This isn't a marketing piece. We've shipped both. We have opinions. We'll show you the numbers. TL;DR For workloads under ~50M vectors with HNSW indexing, Postgres + pgvector beats Pinecone on cost, ops simplicity, and query flexibility, while matching it on latency. Pinecone wins clearly past 100M vectors with sustained high QPS, multi-region replication, and zero-ops requirements. If you're starting a new RAG project and your team already operates Postgres, the default should be pgvector unless you have specific evidence otherwise. We'll explain how we got here.…