Best Free Vector Databases in 2026: Free Tier Limits Compared

Free vector database tiers break at different limits: the storage cap that ends a prototype, the query unit that a busy day burns through, and the idle rule that pauses a demo database. Toolradar data: we track 18 vector databases as of October 10, 2026, and 94% of their pricing was re-checked in the last 90 days, while Milvus drew the most side-by-side comparisons on Toolradar over the same window, 285 of them (Toolradar's vector database ranking).

Methodology: we read each free plan and its first paid step on the vendor's own pricing page, checked in October 2026, and used the vendor's documentation where the pricing page is silent. MongoDB's free-cluster limits and Supabase's pgvector guide are the two cases. Storage printed in gigabytes is converted to vectors with one stated assumption: 1,536 dimensions stored as 32-bit floats, or 6,144 bytes per vector, before any index, metadata or text. Items run from the most vectors a free tier holds down to the fewest. Qdrant and Chroma come last, because Qdrant prints RAM and disk per node rather than a vector count, and Chroma prints no storage cap.

A free vector database is a managed store for embedding vectors, the numeric arrays behind semantic search and retrieval in AI apps, with a no-cost tier capped on storage, queries or projects. Some free tiers are dedicated engines such as Pinecone, Qdrant, Weaviate, Milvus through Zilliz and Chroma. Others are Postgres extensions: pgvector runs on the free plans of Neon and Supabase, which also hold the rest of your app's data. If you want the wider Postgres picture first, see our Postgres hosting guide. A free plan is where a prototype lives before real traffic, so the cap you hit first matters more than the price. Only Chroma's Starter bills usage on top of its free tier.

The free tier decides how far a prototype gets before it costs money, and the storage cap usually ends it first. A 1,536-dimension embedding stored as 32-bit floats takes 6,144 bytes, so 100,000 chunks need about 614 MB before any index. That overflows the 500 MB database on Supabase's free plan, fits inside Neon's 1 GB and Pinecone's 2 GB, and sits exactly at Weaviate's 100,000-object cap. Cloudflare Vectorize caps stored vector dimensions instead, at 5 million, which holds about 3,250 vectors at that size. It suits an experiment, not a corpus. The embedding calls themselves have their own free allowances, covered in our free LLM APIs guide and the LLM API price comparison.

Query units differ too, so compare them with the same care. Upstash allows 10,000 queries or updates a day, roughly 300,000 a month if you use the full allowance every day. Cloudflare counts queried vector dimensions, 30 million a month on its free plan. Qdrant's pricing page prints no query cap at all, so check its limits in the console before you build on it.

Every price here was read on the vendor's own pricing page in October 2026, with the Cloudflare minimum taken from its Workers pricing page. Qdrant prints no unit price for its Standard tier, so its paid step stays usage-based. Zilliz's Serverless plan starts at a $0 monthly fee and bills usage on top, so its flat comparison uses the Enterprise Dedicated floor.

Cost query: a full year of the flat monthly fee alone costs $60 on Cloudflare Workers Paid, $240 on Pinecone Builder, $300 on Supabase Pro, up to $360 on MongoDB Flex, $540 on Weaviate Flex, $2,364 on Zilliz Enterprise Dedicated and $3,000 on Chroma Team, which also bills usage on top.

The arithmetic is worth doing before you commit, because a free plan with a hard cap has no bill to compare at all. For the wider cloud allowances behind Cloudflare and the other hosts, see our cloud free tiers comparison, and for what Supabase costs once its free project fills up, read the Supabase pricing breakdown.