About this tool
What this vector database cost calculator answers
The Vector Database Cost Calculator estimates raw and indexed vector storage size, then the monthly bill on Pinecone, Weaviate Cloud, Zilliz Cloud, turbopuffer and pgvector on Supabase, with each plan minimum applied. Use it to run a vector database pricing comparison on your own corpus and traffic before you pick a vendor or sign a commitment.
Feature matrices tell you which database can do the job. This tool tells you what the job costs. It complements the vector database comparison matrix, which covers hosting, hybrid search and open source status, by modelling the money side: storage, reads, writes and the minimum each vendor charges before usage counts at all.
How to use it
- Enter how many vectors you will store and the embedding dimensions your model produces.
- Choose float32, float16, int8 or binary quantization to set bytes per dimension.
- Add metadata bytes per vector, an index overhead factor and the replica count you plan to run.
- Enter queries and upserts per month, including retries and scheduled reindexing.
- Read the ranked costs, open the formulas panel to check each unit price, then copy the scenario link.
Take vector count from your chunking plan, not your document count: one PDF often becomes dozens of chunks. Dimensions come from the embedding model card. Metadata bytes cover the stored text snippet plus filter fields; 300 to 1,000 bytes is common when you keep the chunk text beside the vector. Count queries from logs or forecasts, and remember that agents and rerankers often run several searches per user request.
What goes into the math
Vector storage size equals vectors times the sum of dimensions times bytes per dimension and metadata bytes. Indexed storage multiplies that by the overhead factor and the replica count. Each vendor then applies its own rules. Pinecone charges storage plus read units, where one query costs one read unit per GB of namespace, plus write units at one per KB upserted. Weaviate Flex charges per million stored vector dimensions plus storage. Zilliz Serverless charges vCUs: 0.25 per KB inserted, and a read cost that grows with the data scanned, with a floor of 6 per query. The turbopuffer bill counts logical GB stored, GB written and data queried, with a 1.28 GB floor per query. The pgvector figure is an approximation of instance plus storage: the smallest Supabase compute size whose memory holds the index, plus disk.
Where the answer usually breaks down
Query cost is the usual surprise. On a metered vendor every search reads the whole namespace, so a single large namespace with heavy traffic costs far more than the same data split by tenant. Retries, agent loops and evaluation runs inflate query counts that teams forget to forecast.
Quantization cuts memory and disk for self hosted engines, but Pinecone and Weaviate still bill dimensions at full float32 size, so the savings do not always reach the invoice. Backups, egress, support plans and region premiums sit outside these estimates.
When the answer is real and when it is not
The estimate is strongest for a single index with steady traffic, the shape of most retrieval augmented generation backends in their first year. It is a fair first pass for pinecone vs qdrant cost decisions once you price the indexed size in the Qdrant calculator.
It is weaker for multitenant systems with thousands of namespaces, workloads with bursty traffic that need dedicated read capacity, and pgvector deployments where query concurrency, not memory, decides the instance size.
How this tool differs from vendor calculators
Vendor calculators price one product with that vendor's defaults. This one runs the same workload through every vendor, keeps each formula visible with the unit price plugged in, and stores the dated price book in a repository data file instead of hiding it in a component. Where a vendor publishes no unit price, the tool says so rather than inventing one.
Sources and methodology
Pinecone rates come from the Pinecone pricing page and the read and write unit rules from Pinecone's cost documentation. Weaviate rates come from the Weaviate pricing page. Zilliz rates come from the Zilliz pricing page and the vCU rules from the Zilliz serverless cost guide. Rates for turbopuffer come from the cost calculator on the turbopuffer pricing page. The pgvector cost model uses Supabase compute and disk pricing. Qdrant Cloud is priced by cluster size on the Qdrant pricing page, so it links to the vendor calculator instead of showing a number. Where a vendor lists a range, the calculator uses the lowest listed rate and says so.
Pressure test your vector database budget
Namespace layout, quantization and retrieval design move the bill more than the vendor choice. Bring the numbers to an architecture review before you commit.
Book an architecture reviewFrequently asked questions
- How much does a vector database cost per month?
- A small workload usually costs the plan minimum, because usage rarely reaches it: $16 per month on turbopuffer Launch, $25 on Supabase Pro for pgvector, $45 on Weaviate Flex and $50 on Pinecone Standard, while Zilliz Serverless has no minimum. Past a few million vectors, query volume usually dominates, so heavy traffic can push the same index into hundreds or thousands of dollars per month.
- How much does Pinecone cost?
- Pinecone Standard bills $0.33 per GB per month for storage, $16 to $18 per million read units and $4 to $4.50 per million write units, with a $50 monthly minimum. A query uses one read unit per GB of namespace, so read cost grows with index size. Starter is free with limits, and Builder is a flat $20 per month.
- How do I calculate vector storage size?
- Multiply the number of vectors by dimensions times bytes per dimension, then add metadata bytes per vector. One million 1536 dimension float32 vectors take 6.14 GB before metadata. Multiply by an index overhead factor, often about 1.5 for HNSW graphs, and by your replica count to get the indexed storage you must provision or pay for.
- Is pgvector cheaper than Pinecone?
- pgvector is free software, so its cost is the Postgres instance and disk that hold the index. At small scale a managed vector database minimum is often cheaper than a Postgres instance with enough memory for the index. At high query volume pgvector usually wins, because queries are not metered and the price stays flat until you need a larger instance.
- Pinecone vs Qdrant cost, which is cheaper?
- It depends on traffic. Pinecone meters every query against namespace size, while Qdrant Cloud charges for cluster resources by the hour and publishes no per unit list price. A small index with heavy query traffic often favours a fixed size Qdrant cluster. A large index with light traffic often favours Pinecone. Price both with the indexed storage size from this calculator.
- Which vector databases have a free tier?
- Pinecone Starter, Weaviate Cloud Free, Zilliz Cloud Free, Qdrant Cloud Free and the Supabase free plan all have free tiers. Limits differ: Pinecone allows 2 GB of storage, Zilliz 5 GB and 2.5 million vCUs per month, Weaviate 100,000 objects, Qdrant a single node with 1 GB of memory and 4 GB of disk. turbopuffer has no free tier.
- How accurate is this vector database cost calculator?
- It applies each vendor's published unit prices and billing rules exactly, so it is accurate for the inputs and assumptions shown in the formulas panel. Real bills move with region, compression profile, backups, egress, namespace layout and query patterns. Treat the output as a pricing comparison for planning, then confirm the leading option in the vendor's own calculator.
- Can I share my scenario with my team?
- Yes. The share button copies a link with every input encoded in the URL hash. Anyone who opens it sees the same vector count, dimensions, precision, traffic and results. The hash never reaches a server, and nothing you enter is stored, so you can share a scenario without creating an account or exposing workload details to a vendor.
Related services and reading
From cost model to retrieval design.