Vector Databases Compared: Pricing, Limits and Lock-In

The problem with every vector database comparison

Including, until you reach the tables below, this one.

Four managed vector databases publish four entirely different billing units. Not four different prices for the same thing, which would be a comparison. Four different things.

We last looked at this category in our vector database showdown, which ranked on features. Five months on, the features have converged and the billing models have not, so this time we price instead.

A table that puts "$20 a month" next to "from $45 a month" next to "$250 a month" is not telling you which is cheaper. It is telling you that three companies picked three different numbers to put on a webpage.

So the useful work here is not ranking. It is showing you what each vendor sells, so you can price your own workload in their units.

What each one actually charges

Every figure below was read from the vendor's own pricing page on 18 October 2026, linked once from each product's heading.

Pinecone

The most completely published pricing of the four, and the only one with per-unit rates visible on every paid tier.

Two details matter. The per-unit rates vary by cloud and region, which is why they appear as ranges rather than figures. And the Standard and Enterprise numbers are minimums, not allowances: you are charged at least $50 or $500, and usage above that is billed on top.

Best for: teams that want to model costs precisely before committing, because Pinecone is the only one here where you can.

Limitations: the unit abstraction cuts both ways. A read unit is not a query; the number a query consumes depends on the query. Precise modelling still requires measuring your own workload first.

Weaviate

Bills on the width of your embeddings rather than on query volume, which suits a very different shape of workload.

The free tier is specific: 1 cluster, 100,000 objects, 1 GB memory and 10 GB disk, 1 collection with up to 3 tenants, 2,000 embedding requests a day and 1,000 Query Agent requests a month. No backup retention, no replication, no SSO.

The page contradicts itself on Premium. The pricing table says contact sales; the FAQ says Premium starts "from $400/month." We report both because the vendor does.

Best for: workloads with many small vectors, where dimension-based billing is cheaper than unit-based billing.

Limitations: every rate is prefixed "from", and the page says they vary by index type, compression method and region. The published figures are floors, not quotes.

Chroma

The most legible model of the four, because it charges for the physical things rather than an abstraction.

Usage rates are the same on every plan:

• Writes: $2.50 per GiB written
• Storage: $0.33 per GiB per month
• Queries: $0.0075 per TiB queried
• Network: $0.09 per GiB returned

The $250 Team plan is a platform fee, not a bundle. It includes $100 of credit and the FAQ states that credit does not roll over, so a team using less than $100 of resources is paying $250 for seats, database count and SOC II rather than for capacity.

Best for: anyone who wants to estimate a bill from first principles. Bytes written, bytes stored and bytes returned are quantities you can measure without a calculator.

Limitations: the jump from Starter to Team is $250 a month for what is largely a collaboration and compliance upgrade. There is nothing in between.

Qdrant

We can describe the model. We cannot tell you what it costs.

Qdrant's pricing page names the billable components clearly: compute (vCPU), memory (GB), storage (GB), backup storage (GB) and tokens for paid inference models, billed hourly on actual usage. The free tier is a single node with 0.5 vCPU, 1 GB RAM and 4 GB disk, with no high availability.

It publishes no rate for any of those components. There is a pricing calculator and there is a sales team.

We are not publishing an estimate. Importing a figure from a third-party comparison and presenting it as Qdrant's price is how most of the inaccurate numbers in this category get into circulation. If the vendor has not stated a rate, we do not have one.

Best for: teams willing to run the calculator or talk to sales, and teams self-hosting the open-source version, where this pricing question does not arise.

Why the free tiers are not comparable either

These are four different kinds of allowance. A $5 credit is not a storage limit. 100,000 objects is not 2 GB, unless you know your dimension width, and 100,000 objects of 1,536 dimensions is a very different thing from 100,000 objects of 384.

Which free tier is most generous depends entirely on the shape of your data, which is the honest answer and also the one nobody wants.

How should you actually choose?

Not on the headline price. Three questions settle it faster:

• What grows fastest in your workload? Query volume, vector count, dimension width and stored bytes do not scale together. Pick the vendor whose billing unit grows slowest for you.
• Do you need a number before you commit? If budget sign-off requires a figure, Pinecone and Chroma let you produce one from published rates. Weaviate gives you a floor. Qdrant does not.
• Do you need a managed service at all? If you already run Postgres and your vector count is in the low millions, pgvector and its relatives remove the vendor question entirely, at some cost in index options and query performance.

If you are still working out what you are storing and why, start with what vector embeddings actually are, and our RAG pipeline tutorial covers where the database sits in a working system.

What about switching later?

Cheaper than people assume on the data, more expensive than people assume on everything else.

The vectors are just numbers. Exporting and re-importing them is straightforward and re-embedding is usually unnecessary, which is the part people worry about and should not.

What does not travel is the layer around the index: filter syntax, hybrid search configuration, metadata schemas, index tuning and every line of client code that touches them. Budget the migration as an integration rewrite, not a data transfer, and the estimate will be closer.

Conclusion

Four vendors, four billing units, and one that does not publish a rate at all. That is the actual state of managed vector database pricing in October 2026, and it is why most comparisons of this category are not comparisons.

Price your own workload in each vendor's units. It takes an afternoon and it is the only method that produces a real answer. Anything faster is a guess with a table around it.

Every figure above was read from the vendor's own pricing page on 18 October 2026, each linked once from its product heading: Pinecone, Weaviate, Chroma and Qdrant. Qdrant publishes no rates, so none are given for it here. Weaviate's Premium minimum appears as "contact sales" in its pricing table and as "from $400/month" in its FAQ; both are reported. Rates described as "from" are vendor floors that vary by region, index type and compression. Confirm current pricing before committing, as this category changes frequently.