WorkflowStack AI
WorkflowsIndustriesToolsGuidesAI QuizBlogEnterprise
Get Free Workflows
WorkflowStack AI

Practical AI workflows for SMB operators and enterprise teams. No fluff. No hype. Just what ships.

Library

  • All Workflows
  • Industries
  • Enterprise
  • Tools
  • Guides

Company

  • About
  • Blog
  • Newsletter
  • Contact

Stay Updated

Weekly workflow ideas for operators and enterprise teams.

Get Free Workflows

© 2026 Blueteem LLC. All rights reserved.

Privacy PolicyTerms of Service
HomeBlogPinecone vs Weaviate vs Qdrant vs Milvus (and pgvector): Vector Database Pricing 2026
November 10, 2026

Pinecone vs Weaviate vs Qdrant vs Milvus (and pgvector): Vector Database Pricing 2026

Seven vector database options priced from each vendor's own pricing page and docs, with a worked monthly bill for one million 1,536-dimension vectors and when plain pgvector is the right answer.

The short answer

If you already run Postgres and have fewer than a few million vectors, use pgvector: the extension is free. Buy a managed vector database when search load, filtering or tenant count outgrows one Postgres server. For 1 million vectors and 1 million queries a month in one namespace, turbopuffer's $16 minimum is cheapest, then Weaviate Flex ($45), Chroma ($53.50), Zilliz-hosted Milvus (about $101 plus storage) and Pinecone Standard (about $123). Qdrant Cloud publishes no unit prices, only a calculator.

What you are actually paying for

A vector database stores embeddings and returns the nearest ones to a query vector. It is the retrieval half of RAG. Each vendor here bills a different part of that work:

  • Pinecone bills read units and write units, and a query's read units grow with the size of the namespace it searches.
  • turbopuffer and Chroma bill the bytes each query scans plus the bytes it returns.
  • Weaviate bills the total number of vector dimensions stored.
  • Zilliz Serverless bills vCUs, and the read vCUs per query grow with the data scanned.
  • Qdrant and Zilliz Dedicated bill provisioned compute, so you pay for a cluster whether or not it is busy.

The billing unit matters more than the list price. Per-customer namespaces can cut a scan-billed bill several times over and do nothing to a cluster-billed one.

Pricing at a glance

Prices and plan names are as published on each vendor's own pricing page; last verified 10 November 2026.

Pricing at a glance
VendorBilling unitPublished entry priceWhat it's for
pgvectorThe Postgres instance you already pay forFree extensionSmall to mid corpora next to relational data
PineconeRead units, write units, GB storedStarter free (2 GB, 1M RUs, 2M WUs); Builder $20/month flat; Standard $50/month minimumFully managed serverless, no capacity planning
Weaviate CloudVector dimensions, storage GiB, backup GiBFree (100,000 objects); Flex from $45/monthHybrid search with built-in embeddings
Qdrant CloudvCPU, memory and disk, billed hourlyFree cluster (0.5 vCPU, 1 GB RAM, 4 GB disk); Standard usage-based via calculatorDedicated clusters on Apache-2.0 open source
Zilliz Cloud (Milvus)vCUs (serverless) or CUs (dedicated)Free (5 GB, 2.5M vCUs/month); Serverless $4 per million vCUs; Dedicated from $126/monthVery large corpora on Milvus
turbopufferGB stored, GB written, PB queried, GB returnedLaunch: $16/month minimumMany namespaces on object storage, low cost
Chroma CloudGiB written and stored, TiB queried, GiB returnedStarter $0 plus usage, $5 credits for new users; Team $250/monthServerless Chroma, same API as the open-source library

Pinecone

Pinecone Standard charges $0.33/GB-month for storage, $4–$4.50 per million write units and $16–$18 per million read units, depending on cloud and region. The $50/month minimum is credited against that usage. Enterprise has a $500/month minimum, higher unit rates, a 99.95% uptime SLA, private endpoints and customer-managed keys. Pinecone's cost docs set the meters. "A query uses 1 RU for every 1 GB of namespace size, with a minimum of 0.25 RUs per query." "An upsert request uses 1 WU for each 1 KB of the request, with a minimum of 5 WUs per request," and an overwrite is also charged for the size of the record it replaces. Builder ($20/month flat) caps you at 10 GB, 5M WUs and 2M RUs a month.

Pinecone has also released pg_pinecone, an Apache-2.0 Postgres extension. Its "native" engine applies filters during the search inside Postgres, and its "remote" engine moves a table's search to a Pinecone serverless index while Postgres keeps every row. It requires PostgreSQL 17 or 18 and superuser access, which managed services such as Supabase don't grant.

When not to use Pinecone: when you have one large namespace with heavy query traffic. 1M queries against a 7 GB namespace cost over $100 in reads alone. It's also the wrong pick if you need to self-host outside the Enterprise BYOC option.

Weaviate Cloud

Weaviate's free tier is "1 cluster per user" with 100,000 objects, 1 GB memory and 10 GB disk. Flex is pay-as-you-go from $45/month on a shared cluster with 99.5% uptime. Premium is a prepaid contract starting at $400/month. Pricing is per vector dimension: "from $0.00465 / 1M" on Flex, plus storage from $0.12/GiB and backups from $0.03036/GiB. Weaviate says "Vector dimension pricing varies by index type, compression method, and region." The open-source repository is BSD-3-Clause outside a `wl` directory, which holds enterprise features under a separate license.

When not to use Weaviate: when your corpus is small enough that the $45 Flex minimum is most of the bill, or when you need SSO or HIPAA, which start at Premium and Premium Dedicated respectively.

Get one new AI workflow every week

Practical playbooks, prompts, and tool stacks — for the SMB operators and enterprise teams actually shipping AI.

Qdrant Cloud

Qdrant publishes no per-unit rates. It bills vCPU, memory, disk, backup storage and paid inference tokens by the hour, and you size a cluster with its calculator. The free tier is a single node with 0.5 vCPU, 1 GB RAM and 4 GB disk and no high availability. Premium has a minimum spend, which isn't published, and adds SSO and private VPC links. A serverless tier is on a waitlist. The engine is Apache-2.0.

When not to use Qdrant Cloud: when traffic is spiky or near zero. You pay for provisioned resources by the hour either way.

Milvus and Zilliz Cloud

Milvus is Apache-2.0. Zilliz, the company behind Milvus, runs it as Zilliz Cloud. Serverless is "$4 per million vCUs" for reads and writes. Read vCUs scale with the data scanned. Zilliz's cost docs list 1 million reads at 25 million vCUs ($100) over 1M 1,536-dim vectors and $300 over 10M, excluding scalar fields. Writes cost 0.25 vCU per KB (1 KB = 1,024 bytes). Dedicated clusters start "From $126 /month" on Standard and "From $197 /month" on Enterprise.

When not to use Milvus or Zilliz: self-hosting Milvus under a few million vectors adds distributed-system operations you don't need. On Zilliz Serverless, model reads and storage in Zilliz's calculator first. Neither the pricing page nor the serverless cost docs state a Serverless storage rate.

turbopuffer and Chroma Cloud

turbopuffer has no free tier ("a commercial-only model"). Launch has a $16/month minimum, Scale $256 and Enterprise $4,096 or more. Indexed storage is $0.33/GB-month, writes $2/GB, data queried $1/PB and data returned $0.05/GB. Each query is billed for at least 1.28 GB scanned, every write request for at least 10 KB, and unused minimums don't roll over.

Chroma Cloud bills $2.50 per GiB written, $0.33 per GiB-month stored, $0.0075 per TiB queried and $0.09 per GiB returned. If you hit a usage limit, service pauses until the limit is raised.

When not to use them: for one huge namespace queried constantly. Per-scan billing favours many small namespaces.

Worked example: 1 million vectors, 1 million queries a month

Assumptions: 1,000,000 records with 1,536-dimension float32 embeddings (6,144 bytes), 1,000 bytes of metadata and an 8-byte ID. That is 7,152 bytes per record, or 7.152 GB (6.661 GiB), in one namespace. Traffic is 1,000,000 top-10 queries a month, each returning about 10 KB of metadata (10 GB a month), plus 100,000 record updates a month. Initial load and embedding costs are excluded; lowest published regional rates are used.

Worked example: 1 million vectors, 1 million queries a month
OptionArithmeticMonthly total
Pinecone StandardStorage 7.152 × $0.33 = $2.36. Reads 1M × ~7.15 RU = 7.15M × $16/M = $114.43. Writes 100k × ~14.3 WU (new plus overwritten record) = 1.43M × $4/M = $5.72≈ $122.51
turbopuffer LaunchStorage $2.36. Queried 1M × 7.152 GB = 7.152 PB × $1 = $7.15. Returned 10 GB × $0.05 = $0.50. Writes 0.7152 GB × $2 = $1.43 (batched, before batch discounts). Sum $11.44, under the minimum$16
Chroma Cloud StarterStorage 6.661 GiB × $0.33 = $2.20. Queried 1M × 0.0065 TiB × $0.0075 = $48.79. Returned 9.31 GiB × $0.09 = $0.84. Writes 0.666 GiB × $2.50 = $1.67≈ $53.50
Weaviate Flex1.536B dimensions × $0.00465/1M = $7.14; storage 6.661 GiB × $0.12 = $0.80; under the minimum≈ $45 plus backups
Zilliz ServerlessReads: Zilliz's docs put 1M reads over 1M 1,536-dim vectors at 25M vCUs = 25M × $4/M = $100. Writes 100k × (6.98 KB × 0.25 + 1) vCU ≈ 274,600 × $4/M = $1.10. Storage per calculator≈ $101.10 plus storage (the docs exclude scalar fields)
Qdrant CloudNo per-unit prices published, and the corpus exceeds the 4 GB free diskUse Qdrant's calculator
pgvectorRuns on your existing Postgres. Size RAM to hold about 6.1 GB of raw vectors plus HNSW graph overhead, or about half that with `halfvec`Your instance cost

In one namespace, the published prices rank turbopuffer ($16), then Weaviate ($45), Chroma ($53.50), Zilliz Serverless (about $101 plus storage) and Pinecone Standard (about $123).

Splitting the namespace changes the ranking. With 100 tenant namespaces of 10,000 records each (about 0.07 GB), a Pinecone query falls to the 0.25 RU minimum, so reads use 250,000 RUs a month ($4 at Standard rates). The whole split workload (7.2 GB, 250k RUs, 1.43M WUs) then fits inside the Builder plan's 10 GB, 2M RU and 5M WU caps for $20 flat. On Standard, the $50 monthly minimum would apply. turbopuffer's 1.28 GB per-query minimum would make its reads $1.28, so it stays at its $16 minimum. Chroma Cloud, with no per-query or monthly minimum, drops to about $5.20 (reads 1M × 0.000065 TiB × $0.0075 = $0.49, plus the same other lines).

pgvector: the honest default

pgvector supports HNSW and IVFFlat indexes. Indexes cover `vector` columns up to 2,000 dimensions and `halfvec` columns up to 4,000. That means 3,072-dimension embeddings need `halfvec` or dimension reduction to be indexed. Its README warns that with approximate indexes, "filtering is applied after the index is scanned." Since 0.8.0 you can turn on iterative index scans to get enough filtered results back. If your app already lives in Postgres (including managed Postgres such as Supabase), pgvector keeps embeddings in the same transactions, backups and row-level security as the rows they describe. Move off it when the vector index no longer fits in memory, or when heavily filtered multi-tenant queries start missing results.

How to choose

  • Under ~5M vectors, Postgres already in the stack: pgvector. If filtered search becomes the bottleneck on self-managed Postgres 17 or 18, look at pg_pinecone. It needs superuser access, so it's not an option on Supabase.
  • Many small tenants, cost-sensitive: turbopuffer or Chroma Cloud. See Pinecone vs turbopuffer and Chroma vs Pinecone.
  • No ops team, predictable managed service: Pinecone. Compare Pinecone vs Weaviate and Pinecone vs Qdrant.
  • Hybrid keyword and vector search with built-in embeddings: Weaviate. See Qdrant vs Weaviate.
  • Hundreds of millions of vectors or more: Milvus or Zilliz. See Milvus vs Pinecone.
  • Enterprise: check where HIPAA sits before pricing. Pinecone lists a $190/month HIPAA add-on, turbopuffer offers a BAA from Scale, Weaviate's HIPAA is on Premium Dedicated, and Zilliz's is on Business Critical.

Frequently asked questions

Is pgvector good enough instead of Pinecone?

For a few million vectors on a Postgres server with enough RAM for the index, usually yes, at no cost beyond the database you already run. Move to a dedicated service when filtered queries miss results or the index outgrows memory.

Is Pinecone cheaper than Weaviate?

It depends on how your data is split. At 1M vectors and 1M queries a month in one namespace, Pinecone Standard works out to about $123, against Weaviate Flex's $45 minimum. Split into 100 small namespaces, the same workload fits Pinecone's $20 flat Builder plan.

Does Qdrant Cloud have a free tier?

Yes. Qdrant's pricing page lists a free-forever single-node cluster with 0.5 vCPU, 1 GB RAM and 4 GB disk, with no high availability.

What is the difference between Milvus and Zilliz Cloud?

Milvus is the Apache-2.0 open-source vector database. Zilliz Cloud is the managed version from Zilliz, with a free tier, Serverless at $4 per million vCUs (about $100 per million reads over 1M 1,536-dim vectors, per Zilliz's docs), and Dedicated clusters from $126 a month.

Does turbopuffer have a free plan?

No. turbopuffer describes itself as commercial-only, and its Launch plan has a $16 monthly minimum that doesn't roll over.

Which vector database is cheapest for a small RAG app?

pgvector, if you already pay for Postgres. Among managed services, turbopuffer's $16 minimum was cheapest for 1M vectors and 1M queries in one namespace, ahead of Weaviate Flex at $45. Split into 100 per-tenant namespaces, Chroma Cloud's usage-only Starter falls to about $5 a month, and Pinecone Builder covers it at $20 flat.

Next step

Put the database behind a working retrieval flow with the RAG support agent workflow, or the enterprise RAG internal knowledge workflow for governed internal search.

Related Workflows

AI-Powered Price Quoting

Generate accurate price quotes in minutes instead of hours by using AI to analyze job details and pull from your pricing database.

View workflow

Shipment Exception & Delay Management

Automatically detect shipment delays and exceptions, notify affected customers, and coordinate carrier resolution — before anyone calls to complain.

View workflow

SOP & Internal Knowledge Assistant

Give your team instant answers from your internal docs — no more "where is that procedure again?"

View workflow

Keep Reading

November 10, 2026

Langfuse vs LangSmith vs Braintrust (and PostHog): LLM Observability Pricing 2026

The six LLM observability tools buyers compare most, priced from each vendor's own pricing page, with a worked monthly bill at 300,000 traces and a five-person team.

March 19, 2026

Vapi vs Bland vs Retell (2026): AI Voice Agent Pricing, Limits and Real Monthly Cost

Vapi, Retell and Bland price AI phone agents in three different ways. We read each vendor's own pricing page and worked out what 2,000 inbound minutes a month actually costs on each.

November 10, 2026

Composio vs Arcade vs MCP in 2026: Agent Tool Integration Pricing, Auth and Security

The managed tool layers for AI agents price the same action four different ways. Here is what Composio, Arcade.dev and Pipedream Connect publish, what running your own MCP servers costs you instead, and a worked monthly bill.

Found this helpful?

Get weekly AI workflow ideas in your inbox.