pgvector

Prompt caching cuts a 400K-token long-context answer to $0.085 on Claude Sonnet 5.5, but RAG still answers for $0.017. The crossover formula, the accuracy-at-depth evidence and the latency users notice.

Re-embedding a 10M-chunk corpus costs about $520 at September 2026 rates, so the embedding model is not the one-way door teams fear. voyage-4-large is the default; OpenAI's model has had no successor in 32 months.

At 10,000 users, Postgres with pgvector costs $163-$332 a month against Zep's $375-$750 and Letta's $1,020 — and no independent party has confirmed that a commercial memory store retrieves better. Buy one when you can name the column you could not write yourself.

At 5M chunks and 200,000 queries a month, a managed retrieval API costs about $400 a month and a per-seat search product lists at about $80,000. Neither number decides it — who owns permission sync does.

Modelled at three workload sizes on 18 August 2026: managed pgvector runs about $1,910/month at 10M vectors against Pinecone's $19,716 on-demand. But the real problem is that Pinecone's cheaper Dedicated Read Node rate and Qdrant Cloud's per-unit rates are not published anywhere.

Self-hosting 50 million vectors costs about $1,533/month in reserved compute and roughly $1,800/month in the engineer who runs it. The right reasons to leave a managed vector database are residency, licence and re-pricing risk — not the invoice.

At 10 million vectors and 20 million queries a month, Pinecone bills $20,557 against one namespace and $157 against a thousand. The shape of your data, not its size, decides whether you need a dedicated vector store at all.