Vector Database Comparison: Pinecone vs Weaviate vs Qdrant vs pgvector

This is a curated reference tool, not a benchmark or a cost calculator. There is no formula to run — the value is in the data provenance and how each cell is normalized so that columns are genuinely comparable across vendors.

1. Cell sourcing.Every value is taken from the vendor's own documentation or pricing page — never a third-party blog — and each row links back to that source so you can verify it.
2. License.We record the exact OSI identifier (Apache-2.0, BSD-3-Clause, PostgreSQL) where one applies, or “Proprietary” / “source-available” with a note. Redis and MongoDB carry licensing caveats because their terms changed over time, so those are flagged rather than simplified.
3. Free tier.We list the concrete published limit — “1GB managed cluster”, “M0 shared 512MB”, “Starter serverless” — not just “yes”, because the size of the free tier is what actually decides whether you can ship on it.
4. Max dimensions. The largest indexable vector dimension the docs state. Where indexed and stored limits differ (pgvector caps indexes at 2,000 dimensions but stores up to 16,000), we record both. Where the limit is memory- or config-bound rather than a published hard number, we say so instead of inventing a figure.
5. Hybrid search.Marked yes only when the engine ships first-party keyword + vector fusion (BM25 or sparse vectors). pgvector and Chroma are marked no because you must combine full-text and vector results yourself — pgvector via Postgres' tsvector, Chroma via your own re-ranking.
6. Starting paid price.The first paid step above the free tier, with its unit. SaaS pricing drifts, so this is a dated snapshot used mainly for the “starting price” sort — confirm the live figure at the source link before you commit.

The “Quick verdict” panel is a fixed decision map: each recommendation is keyed to the normalized cells above (the open-source flag, free-tier presence, hosting model and index families), not an opinion. As a self-consistency check, the set of databases that are open-source and offer a managed cloud andhave a free tier is computed directly from the table — and it contains Qdrant and Weaviate, exactly the “open-source + managed cloud” recommendation. Because pricing and limits change, every figure is dated with 2026-06-21 and linked to its source so you can re-check it yourself.

• Pinecone — pricing & limits
• Weaviate — pricing & docs
• Qdrant — pricing & documentation
• Milvus / Zilliz Cloud — pricing & docs
• Chroma — docs & Chroma Cloud
• pgvector — Postgres extension (README & limits)
• Redis — vector search documentation
• MongoDB Atlas Vector Search — documentation
• ANN-Benchmarks — independent recall/latency benchmarks

All cells were last cross-checked against these vendor sources on 2026-06-21. Pricing, free-tier sizes and starting prices change frequently — click through to confirm the current figure before you decide.