What a Vector Database Costs After the Free Tier

Pinecone Standard has a $50 monthly usage minimum, Weaviate Flex starts at $45 a month, and turbopuffer launch has a $16 monthly usage minimum. Those numbers do not price the same workload: other services charge for server resources, stored bytes, query processing, or database compute. The useful comparison is a monthly bill for your index and query pattern, followed by the extra cost of keeping both old and new indexes live during a move. Pinecone, Weaviate, and turbopuffer publish those entry terms.

This guide starts with vectors you already have. Producing embeddings and generating an answer are separate costs. If the database serves a model, our API price comparison helps price the model side; it does not include vector storage.

Compare eight ways to store and search vectors

The table identifies the meter you must estimate before comparing bills. A starting price is a plan condition, not a quote for a working production index. Check the region, plan, backup and transfer terms for your account.

Route	Published starting term	Main billable units and constraint
Pinecone serverless	Starter free; Builder $20/month flat; Standard $50/month minimum usage	Storage, read and write units, plus other metered services. Builder blocks use beyond its included limits instead of billing overage. Standard charges the greater of eligible usage or its minimum. How Pinecone measures cost.
Weaviate Cloud	Free tier includes 100,000 objects; Flex starts at $45/month	Flex lists vector dimensions from $0.00465 per million and storage from $0.12/GiB, with backup another meter. The listed rates say “from”; your deployment and calculator result matter.
Qdrant Cloud	Free tier includes 1 GB RAM and 4 GB disk; Standard is usage based	Dedicated cluster resources determine the bill. Its public page does not give one universal Standard dollar rate; obtain a quote for the selected resources and cloud region.
Zilliz Cloud	Free includes 5 GB storage and 2.5 million vCUs/month; Standard serverless starts from $0/month, Dedicated from $126/month	Serving compute and storage depend on deployment. Its list price guide shows dedicated compute at $0.273 per CU-hour and dedicated storage at $0.025/GB-month for the displayed selection; confirm the plan, cloud and region before using either rate.
turbopuffer	launch has a $16/month minimum usage	Logical storage bytes, write bytes and query processing; scale has a $256/month minimum. The calculator’s small example can show $7 of usage while the launch minimum still makes the invoice $16.
Chroma Cloud	Starter $0/month plus usage and $5 in free credits; Team $250/month plus usage	$2.50/GiB written, $0.33/GiB-month stored, $0.0075/TiB queried and $0.09/GiB returned on the displayed usage rate card. Team’s fee sits alongside usage, subject to its credit terms.
Supabase Postgres with pgvector	Free database has a 500 MB limit; Pro starts at $25/month	pgvector runs in Postgres. Supabase pricing includes 8 GB disk on Pro, then lists $0.125/GB, while project compute, egress and other services can change the bill. There is no separate per-vector price in that rate card.
Amazon S3 Vectors	No plan floor stated on the public S3 Vectors rate section	Storage, uploaded logical GB, query requests, data processed and returned data. AWS’s US East example uses $0.06/GB-month storage, $0.20/GB uploaded and $2.50 per million query requests; query processing and data returned are additional. Region matters.

These are eight ways to buy retrieval, not eight interchangeable performance profiles. Supabase is attractive when a suitable Postgres project already exists; its incremental cost can differ from opening a new Pro project. S3 Vectors is an object storage service with its own query and result-size charges. Qdrant and Zilliz dedicated deployments require you to size resources. Do not compare the table’s free tiers with a production requirement for availability, filters, backups and response time.

The comparison also needs the plan you would actually qualify for. Pinecone’s free and Builder plans have limits, while Standard’s minimum is a usage commitment rather than a $50 fee added to every metered line. Chroma Team explicitly says $250 plus usage and includes credits under its terms. Treating these as the same kind of “base price” will distort a spreadsheet before the first query is counted.

Turn your workload into billable units

Record a representative month: active vectors, dimensions, precision, metadata and key bytes; new and updated vectors; query count; the number of results and bytes returned; backup retention; and region. Split normal traffic from peak traffic. Check whether the provider meters the logical bytes you send, the bytes it stores, index resources, or work done per query.

For a transparent illustrative raw-data calculation, one million 1,024-dimensional float32 vectors contain 1,000,000 × 1,024 × 4 = 4,096,000,000 vector bytes, or 4.096 decimal GB. That is roughly 3.81 GiB before keys, metadata, indexes, replicas and backups. This arithmetic is not a prediction of any provider’s storage bill or RAM requirement. AWS says S3 Vectors uses four bytes per dimension plus key and metadata in its logical size calculation; turbopuffer describes its own logical-byte and precision rules. Other providers use different accounting.

In decimal notation, 1 GB is 1,000,000,000 bytes; 1 GiB is 1,073,741,824 bytes. Billing labels need a separate check: AWS’s S3 pricing details define their GB as 2^30 bytes, equivalent to GiB. Chroma and Weaviate explicitly label storage in GiB. Use each provider’s definition when converting bytes into billable units.

Queries need more than a count. S3 Vectors bills query requests, processed index data and returned data. AWS’s pricing page states a 512KB free returned-data allowance per query in its overview and footnote, but uses 500KB in its worked examples. Confirm the applicable allowance before relying on a boundary-case estimate. Pinecone meters read units; its cost guide also lists egress separately. A month with many small searches can have a different cost shape from one with fewer searches over a large index. Use provider billing metrics or a trial workload to fill unknown quantities; raw vector bytes alone cannot supply them.

Calculate a normal month

For a plan with a minimum usage commitment, first total only usage that the provider says counts toward that commitment. Then use monthly invoice for those items = max(monthly minimum, eligible usage), and add any items outside that commitment. Pinecone’s billing example says $20 of Standard usage produces a $50 invoice; $100 of usage produces a $100 invoice. The same pattern appears in turbopuffer’s minimum FAQ.

For a fixed plan fee plus metered usage, keep the fee separate. Chroma’s Team page says $250/month plus usage and lists $100 in credits. For an illustrative month with $70 of eligible usage, the calculation is 250 + max(0, 70 − 100) = $250, if those credits apply to all $70 and have not been spent elsewhere. With $140 of eligible usage on the same assumptions, it is 250 + (140 − 100) = $290. Confirm how credits apply on the actual invoice; do not carry unused Team credit into another month, since Chroma says that included usage does not roll over.

For a resource-priced cluster, use the chosen resource size multiplied by billable hours, then add storage, backup and transfer as the provider specifies. For Supabase, compare the whole project bill rather than a hypothetical “pgvector fee.” For S3 Vectors, calculate each of its five meters. Our document parsing cost worksheet uses the same discipline of counting the unit behind an operation, but its API prices do not apply to vector indexes.

Budget the month when two indexes run

An old index normally remains billable while you import vectors into a new one and test reads. During the overlap, count both providers, not just the destination’s steady-state quote. Use a ledger with four lines: old provider’s charge for the billing period, new provider’s charge for that period, one-time import/export and data transfer, and additional writes caused by feeding both indexes. Include any new backup or test environment that the migration actually requires. Do not insert an estimated fee where the provider or your account has not supplied one.

Here is a wholly illustrative, provider-neutral decision worksheet, not a claim about a vendor’s invoice:

Item	Example amount	Why it appears
Existing service in the migration month	$80	The old index remains available until cutover.
New service in the migration month	$50	Its full assumed monthly minimum binds.
Import, export, transfer and duplicate writes	$30	Hypothetical measured or quoted transition items.
Migration-month total	$160	$80 + $50 + $30.
Extra over staying on the old service that month	$80	$160 − $80.

The overlap might last seven days, yet a monthly minimum might still apply. Do not multiply a $50 monthly commitment by 7/30 without a provider term or an invoice showing proration. Conversely, hourly resources may be billed only for the hours they exist; read the selected provider’s terms. The point of the worksheet is to expose the extra bill, not to pretend that all providers prorate alike.

Before switching reads, compare representative queries for correct filters, result relevance, latency at peak load, availability and recovery. Those are acceptance criteria to measure in your own environment, not performance claims made here. Keep the old index until the new one meets them. If you need model generation in the same workflow, our web-search API cost example shows how a second service can enter an agent’s bill; that bill is separate from vector retrieval.

Decide whether moving pays back

Subtract the new steady-state monthly total from the old one. If old is an illustrative $80 and new is $50 under the same workload and service requirements, monthly savings are $30. Divide the extra transition cost of $80 above by $30: 80 / 30 = 2.67 months. That means roughly three full months of those assumed savings are needed after cutover to recover the extra migration spend. If the new service has another $20 of ongoing backup or transfer expense, savings fall to $10 and payback becomes eight months.

If the calculated recurring saving is zero or negative, there is no finite cost payback. A move might still be justified by a feature or reliability requirement, but label that separately. Add your team’s migration labor, any commitment termination cost, and the cost of retaining rollback data using your figures. No public rate card can price your engineering time.

Recalculate after a representative billing period. Invoice categories, especially query processing, backups and data transfer, can differ from a pre-migration estimate. The AI API cost calculator is useful for downstream model tokens, but it does not calculate any of the vector database invoices in this guide.

Sources

• Pinecone pricing and cost guide: plans, minimum commitments and meters.
• Weaviate pricing: free tier, Flex floor and listed units.
• Qdrant pricing: free tier and Standard resource model.
• Zilliz pricing and list price guide: plan entry and selected CU/storage rates.
• turbopuffer pricing: plan minimums, usage meters and logical bytes.
• Chroma pricing: usage rates, Team fee and credits.
• Supabase pricing and Vector: project charges and pgvector route.
• Amazon S3 pricing, Vectors section: logical size, regional example and query meters.

All pages read on November 7, 2026.

Free toolWhat will 10,000 chats a month cost?Estimate my bill