Storing 10 million 1536-dimension vectors and serving 1 million queries a month costs about $63 on turbopuffer, $79 on Weaviate Flex, $509 on Chroma Cloud and $1,166 on Pinecone Standard as of October 30, 2026 (our arithmetic on list rates). The spread comes from how each vendor meters reads: Pinecone and Chroma charge by the size of the namespace each query touches, Weaviate bills stored dimensions and not queries. Splitting data into small namespaces drops Pinecone to its monthly minimum.
Storage, query and minimum-spend rates from eight managed vector databases, priced on one RAG workload at three sizes, plus what self-hosting Qdrant, Milvus or pgvector saves.
Some links on this page are affiliate links: if you sign up through one, Toolradar may earn a commission. Commissions never change our ranking or what we write. How we make money
A vector database bill is mostly decided by one design choice: does the vendor charge you per query, and if so, does the price grow with the size of the index the query searches? Pinecone, Chroma and turbopuffer all meter reads, but in different ways, so the same 10-million-vector index can cost under $100 or over $1,000 a month depending on where it lives and how you partition it.
Toolradar data: our vector database category lists 21 tools, and 81% of them offer a free or freemium plan, which here usually means a capped starter cluster or an open-source engine you host yourself. The vector database statistics page covers the wider category.
Methodology: we read the published rates of eight managed services and three open-source engines on each vendor's own pricing page and documentation as of October 30, 2026, in US dollars, and took the lowest listed rate where a range varies by cloud or region. Every monthly total below is our arithmetic on one workload: 1536-dimension float32 embeddings with 1 KB of metadata each (7.14 KB per record, the size Pinecone's own cost docs give for that shape), one namespace or collection, and 1 million queries a month. Totals exclude tax, support plans, embeddings and data returned to the client unless stated. Ranking is by cost fit for that workload; no vendor paid for a position.
Top Picks
Selected by pricing the same workload (1536-dimension embeddings with 1 KB of metadata, 1 million queries a month) on each managed vector database's published rates, then ranking by what that bill looks like at 1M, 10M and 100M vectors. How we rate
Serverless vector and full-text search engine billed on the logical GB you store, write and query.
Starting price
From $16 /mo minimum (Launch)
Best for
Large or multi-tenant indexes where the lowest metered bill matters more than a free tier.
Open-source vector database with a managed cloud billed on stored vector dimensions, not on queries.
Starting price
Flex from $45/mo
4.4 (47 reviews)
Best for
Teams with high query volume on one large index who want a bill that does not move with traffic.
Fully managed serverless vector database billed on storage plus read and write units.
Starting price
Builder $20/mo flat; Standard $50/mo min
4.2 (68 reviews)
Best for
Multi-tenant SaaS apps that can give each customer a small namespace.
| Tool | Starting price | Rating | Best for |
|---|---|---|---|
| turbopuffer Visit turbopuffer website | From $16/mo minimum (Launch) | n/a | Best for: Large or multi-tenant indexes where the lowest metered bill matters more than a free tier. |
| Weaviate Cloud Visit Weaviate Cloud website | Flex from $45/mo | 4.4 47 reviews | Best for: Teams with high query volume on one large index who want a bill that does not move with traffic. |
| Pinecone Visit Pinecone website | Builder $20/mo flat; Standard $50/mo min | 4.2 68 reviews | Best for: Multi-tenant SaaS apps that can give each customer a small namespace. |
| Zilliz Cloud Visit Zilliz Cloud website | Free 5 GB; Dedicated from $197/mo | 4.7 53 reviews | Best for: Teams scaling past 100 million vectors that want a tiered-storage option. |
| Qdrant Cloud Visit Qdrant Cloud website | Free 1 GB RAM node; usage billed hourly | 4.5 12 reviews | Best for: Engineers who want dedicated resources and the option to self-host the same engine later. |
| Chroma Cloud Visit Chroma Cloud website | Starter free + usage; Team $250/mo | 4.2 from 6 reviews | Best for: Prototypes and apps with many small collections, such as per-user memory for agents. |
| MongoDB Atlas Vector Search Visit MongoDB Atlas Vector Search website | Flex $8 to $30/mo; M10 from $0.08/hr | 4.5 901 reviews | Best for: Teams on MongoDB that want vectors next to their documents without a second database. |
| Upstash Vector Visit Upstash Vector website | $0.40 per 100K requests | n/a | Best for: Low-traffic or bursty apps under about 1.3 million 1536-dim vectors. |
| pgvector Visit pgvector website | Free extension; host from $25/mo | n/a | Best for: Teams already on Postgres with under a few million vectors. |
| Milvus Visit Milvus website | Free, open source | 4.8 310 reviews | Best for: Platform teams that can run Kubernetes and want the engine behind Zilliz without the managed fee. |
turbopuffer
Serverless vector and full-text search engine billed on the logical GB you store, write and query.
- Starting price
- From $16/mo minimum (Launch)
- Free plan
- No
Best for: Large or multi-tenant indexes where the lowest metered bill matters more than a free tier.
- Pro: Queried data above 32 GB gets an 80% marginal discount, and 96% above 128 GB.
- Pro: Batch writes of about 3.2 MB or more get 50% off.
- Pro: Attributes marked non-filterable are billed at half rate.
- Con: No free tier: every plan has a monthly minimum, and unused minimums do not roll over.
- Con: Enterprise carries a 35% usage premium on top of a $4,096 monthly minimum.
The 'Enterprise' tier at an estimated $4096/month is also very high.
Weaviate Cloud
Open-source vector database with a managed cloud billed on stored vector dimensions, not on queries.
- 4.4 on G2 (47 reviews)
- Starting price
- Flex from $45/mo
- Free plan
- Yes
- Rating
- 4.4 (47 reviews)
Best for: Teams with high query volume on one large index who want a bill that does not move with traffic.
- Pro: No per-query fee: dimensions, storage and backups make up the bill.
- Pro: Built-in embeddings from $0.025 per 1M tokens.
- Pro: Free cluster with 100,000 objects and 10 GB of disk.
- Con: The dimension rate varies by index type, compression and region; only the console shows yours.
- Con: Data transfer is free only for a promotional period.
Weaviate's pricing is fair, offering an always-free plan and a flexible entry point with the Flex plan from $45/month.
Watch out
Flex is usage-based: cost grows with vector dimensions and storage above the $45/month minimum, and backups are charged on top
Pinecone
Fully managed serverless vector database billed on storage plus read and write units.
- 4.5 on G2 (66 reviews)
- 2.9 on Trustpilot (2 reviews)
- Starting price
- Builder $20/mo flat; Standard $50/mo min
- Free plan
- No
- Rating
- 4.2 (68 reviews)
Best for: Multi-tenant SaaS apps that can give each customer a small namespace.
- Pro: Standard allows 100,000 namespaces per index, so per-tenant partitioning is built in.
- Pro: 3-week Standard trial with $300 in credits.
- Pro: Integrated embedding and reranking models on the same bill.
- Con: Each query costs 1 read unit per GB searched, so one large namespace gets expensive fast.
- Con: HIPAA on Standard is a $190 monthly add-on with a 6-month minimum.
Pinecone's pricing model is fair and offers good flexibility.
Watch out
Usage-based fees for storage, reads, writes on Standard
Zilliz Cloud
Managed Milvus from its main contributor, sold as serverless or as dedicated compute units.
- 4.7 on G2 (53 reviews)
- Starting price
- Free 5 GB; Dedicated from $197/mo
- Free plan
- Yes
- Rating
- 4.7 (53 reviews)
Best for: Teams scaling past 100 million vectors that want a tiered-storage option.
- Pro: Capacity-optimized clusters from $16 and tiered-storage from $5 per million vectors a month.
- Pro: Free tier with 5 GB storage and 2.5M vCUs a month.
- Pro: 30-day free trial on Standard and Enterprise.
- Con: Per-CU capacity figures assume 768-dim vectors, so 1536-dim workloads need your own sizing.
- Con: Serverless read and write rates are not shown as a simple table on the main pricing page.
Zilliz's pricing is fair, offering a generous free tier for initial exploration.
Watch out
Overage fees for exceeding storage limits
Qdrant Cloud
Open-source vector search engine with a managed cloud billed hourly on the resources your cluster uses.
- 4.5 on G2 (12 reviews)
- Starting price
- Free 1 GB RAM node; usage billed hourly
- Free plan
- Yes
- Rating
- 4.5 (12 reviews)
Best for: Engineers who want dedicated resources and the option to self-host the same engine later.
- Pro: Free node with 0.5 vCPU, 1 GB RAM and 4 GB disk.
- Pro: Same Apache 2.0 engine runs on your own servers.
- Pro: Hybrid Cloud runs managed clusters on your own infrastructure.
- Con: Rates appear only through a calculator, not as a list on the pricing page.
- Con: Premium needs a minimum spend for SSO and private VPC links.
Qdrant occupies the sweet spot between developer-friendly open source and production-grade managed cloud.
Watch out
Opaque per-unit pricing: Qdrant Cloud does not publish per-vCPU, per-GB-RAM, or per-GB-storage rates on its pricing page. You must use the pricing calculator at cloud.qdrant.io/calculator or contact sales to estimate costs. This makes budget planning harder compared to competitors with transparent rate cards
Chroma Cloud
Serverless hosting for the open-source Chroma database, billed per GiB written, stored and queried.
- 4.2 on G2 (6 reviews)
- Starting price
- Starter free + usage; Team $250/mo
- Free plan
- Yes
- Rating
- 4.2 from 6 reviews
Best for: Prototypes and apps with many small collections, such as per-user memory for agents.
- Pro: Starter plan has no monthly fee and includes $5 in credits.
- Pro: Query rate of $0.0075 per TiB queried is cheap for small collections.
- Pro: Collection forks cost $0.03 each and share unchanged data.
- Con: One 66.5 GiB collection costs about $487 a month in query fees at 1M queries.
- Con: Each metadata or full-text filter in a query counts as another query.
It's best for individual developers, startups, and small to medium-sized businesses looking for a flexible and scalable vector database solution.
MongoDB Atlas Vector Search
Vector search inside the MongoDB Atlas cluster that already holds your application data.
- 4.5 on G2 (901 reviews)
- Starting price
- Flex $8 to $30/mo; M10 from $0.08/hr
- Free plan
- Yes
- Rating
- 4.5 (901 reviews)
Best for: Teams on MongoDB that want vectors next to their documents without a second database.
- Pro: Free M0 cluster with 512 MB for testing.
- Pro: Dedicated M30 with 8 GB RAM at $0.54 an hour.
- Pro: Separate search nodes let vector queries scale apart from the database.
- Con: You pay for RAM sized to the index around the clock, whatever the traffic.
- Con: Holding 61 GB of raw 1536-dim vectors in memory points to the 64 GB M60 tier at $3.95 an hour.
MongoDB Atlas offers a fair and flexible pricing model.
Watch out
Dedicated tier storage and I/O are additional.
Upstash Vector
Serverless vector index billed per request, with scale to zero and a REST API.
- 4.5 on G2 (2 reviews)
- Starting price
- $0.40 per 100K requests
- Free plan
- Yes
Best for: Low-traffic or bursty apps under about 1.3 million 1536-dim vectors.
- Pro: Queries and upserts cost the same $0.40 per 100K requests.
- Pro: Storage at $0.25 per GB, bandwidth free up to 200 GB a month.
- Pro: Fixed plan at $60 a month covers up to 1M queries or updates a day.
- Con: Pay as you go and Fixed cap out at 2 billion vector dimensions and 50 GB of data.
- Con: Larger indexes need the Pro plan, priced by quote.
Upstash offers a very generous free tier, making it highly accessible for developers.
Watch out
Redis storage beyond 100GB on Pay as You Go
- Starting price
- Free extension; host from $25/mo
- Free plan
- Free to use
Best for: Teams already on Postgres with under a few million vectors.
- Pro: No licence fee and no second database to secure or back up.
- Pro: Supports half-precision and binary vectors to cut storage.
- Pro: Runs on most managed Postgres hosts, including Supabase.
- Con: HNSW indexes on the vector type stop at 2,000 dimensions; larger embeddings need halfvec.
- Con: Supabase's benchmark needs 2XL compute ($410 a month) for 1 million 1536-dim vectors.
pgvector's pricing is exceptionally generous, being entirely free and open-source.
- Starting price
- Free, open source
- Free plan
- Yes
- Rating
- 4.8 (310 reviews)
Best for: Platform teams that can run Kubernetes and want the engine behind Zilliz without the managed fee.
- Pro: Standalone mode runs on a single machine; Milvus Lite installs with pip.
- Pro: Compute and storage scale separately in the distributed version.
- Pro: Same engine as Zilliz Cloud, so moving to managed later is direct.
- Con: The distributed version is a set of microservices on Kubernetes to operate yourself.
- Con: No licence cost, but the hardware, backups and on-call time are all yours.
Milvus offers a very generous pricing model, especially with its robust open-source option and the Zilliz Cloud Free tier providing 5GB storage and 2.5M vCUs/month at no cost.
Watch out
Overage fees for vCUs on Serverless tier
Other Vector Databases tools worth considering
More published tools from our Vector Databases category, ordered by our category ranking. They are not part of the editorial picks above.
How vector database pricing works
A vector database stores embeddings (lists of numbers that represent text, images or code) and returns the nearest matches to a query vector, which is the retrieval step behind RAG chatbots, semantic search and recommendations.
Managed services price that work in four different ways, and the model matters more than the headline rate:
| Pricing model | What you pay for | Vendors |
|---|---|---|
| Usage units | Storage per GB plus read and write units per request | Pinecone, Chroma Cloud, turbopuffer |
| Stored dimensions | Total vector dimensions held, plus storage and backups; queries not metered | Weaviate Cloud |
| Provisioned compute | Cluster size per hour (CPU, RAM, disk) whatever the traffic | Zilliz Dedicated, Qdrant Cloud, MongoDB Atlas, Supabase |
| Per request | A flat fee per query or upsert, plus storage | Upstash Vector |
Why the model changes the bill. Under usage units, a Pinecone query costs 1 read unit for every 1 GB of namespace it searches, so query cost rises with the index. Under stored dimensions, Weaviate charges the same whether you run 1,000 or 10 million queries. Under provisioned compute, you pay for RAM sized to hold the index even at 3 a.m. when nobody searches.
Minimums set the floor. Pinecone Standard, Weaviate Flex and turbopuffer's Launch plan each bill a monthly minimum, while Chroma's Starter plan charges usage only. A small app that would cost $8 in raw usage on a plan with a minimum still pays the minimum.
How much do 10 million vectors cost per month?
One workload, priced on every service: 10 million 1536-dimension vectors with 1 KB of metadata (71.4 GB, or 66.5 GiB), kept in a single namespace or collection, serving 1 million queries a month. Rates as of October 30, 2026; totals are our arithmetic.
| Service and plan | Arithmetic | Monthly total |
|---|---|---|
| turbopuffer Launch | 71.4 GB at the Launch storage rate + 39.9 GB billable per query x 1M at $1 per PB | About $63.44, plus data returned |
| Weaviate Flex | 15,360M dimensions at $0.00465 per 1M + 66.5 GiB at $0.12 | About $79.40, plus backups |
| Chroma Cloud Starter | 66.5 GiB stored + 1M queries x 0.065 TiB at $0.0075 per TiB | About $509.01, plus data returned |
| Pinecone Standard, one namespace | 71.4 GB x $0.33 + 71.4M read units at the low end of the Standard read rate | $1,165.96 |
| Pinecone Standard, 100 namespaces | Each query searches 0.714 GB, so 714,000 read units | $34.98 in usage, billed at the Standard plan minimum |
| Zilliz Dedicated, capacity-optimized | Capacity-optimized rate per million 768-dim vectors; 1536 dims need about twice the capacity | Roughly $320, our estimate |
| Upstash Vector, pay as you go | 15.36 billion vector dimensions exceeds the 2 billion plan cap | Pro plan, contact sales |
Loading the data is a separate, one-off charge. Writing the 10 million records costs $285.60 on Pinecone (71.4M write units at $4 per million), $166.25 on Chroma ($2.50 per GiB written) and up to $142.80 on turbopuffer, which halves the write rate for batches of about 3.2 MB or more. Weaviate lists no write fee.
What the table hides. Pinecone's and Chroma's totals are driven almost entirely by reads, because each query is billed on the full size of the namespace or collection it searches. A multi-tenant app that gives each customer its own small namespace pays a fraction of that. Weaviate's figure uses the lowest listed dimension rate; the real rate depends on index type, compression and region, which the console shows per cluster.
Key Features to Look For
Read pricing that scales with namespace size (Essential)
On Pinecone, a query costs 1 read unit per GB of namespace (minimum 0.25). On turbopuffer, data queried is the namespace size or 1.28 GB, whichever is larger. Partitioning by tenant is the biggest cost lever on both.
Monthly minimum spend (Essential)
Pinecone Standard charges at least $50 a month and Enterprise $500. turbopuffer's Scale plan starts at $256, and Weaviate Premium from $400. Below those thresholds you pay for capacity you do not use.
Free tier limits (Essential)
Pinecone Starter holds 2 GB, Weaviate's free cluster 100,000 objects, Zilliz Free 5 GB with 2.5M vCUs a month, Qdrant 1 GB RAM and 4 GB disk. A 1M-vector index of 1536-dim embeddings (about 7 GB) outgrows most of them.
Write and import fees (Important)
Initial loads are not free: Pinecone bills 1 write unit per KB written and $0.25 per GB imported from object storage, Chroma $2.50 per GiB written, turbopuffer bills every write as at least 10 KB.
Built-in embedding models (Important)
Generating embeddings inside the database avoids a second vendor. Weaviate Embeddings start at $0.025 per 1M tokens and Pinecone's llama-text-embed-v2 costs $0.16 per 1M tokens on Standard.
Backups and egress (Important)
Pinecone charges $0.10 per GB a month for backups and $0.10 per GB of egress beyond the included 100 GB. Weaviate bills backups from $0.03036 per GiB on Flex, and they are on by default.
What to work out before you pick a vector database
-
Size your records first: a 1536-dim float32 vector is 6,144 bytes before metadata, so 10 million of them is over 61 GB before any index overhead.
-
Count queries per month and the size of the namespace each one searches. On metered services that product, not storage, sets the bill.
-
Decide whether your data splits naturally by customer, workspace or document set. Many small namespaces make Pinecone, Chroma and turbopuffer far cheaper.
-
Check the monthly minimum against your realistic usage for the first six months, not your year-two forecast.
-
Price the initial load and every full re-embedding: switching embedding models means rewriting every vector.
-
If your data already lives in Postgres or MongoDB, price the vector index there before adding a second database.
Evaluation Checklist
Load a 10% sample of your real data and read the storage figure the console reports, then multiply.
Run a day of production-like queries and check the read units or GB queried that the invoice preview shows.
Test recall at your target latency with compression on, since compression changes both quality and price on Weaviate and Zilliz.
Confirm the embedding dimension limit for the index type you plan to use (pgvector's HNSW index on the vector type goes up to 2,000 dimensions).
Set a usage limit or budget alert before you point production traffic at a metered service.
Export a collection and restore it elsewhere to check how hard leaving would be.
Pricing Overview
Free tiers
Prototypes under about 2 to 5 GB: Pinecone Starter, Weaviate's free cluster, Zilliz Free, Qdrant's 1 GB RAM node, MongoDB M0, Upstash Free at 10K requests a day.
$0
Low-minimum usage plans
Small production apps paying only for what they store and query: Chroma Starter, turbopuffer Launch, Upstash pay as you go, Pinecone Builder at a flat rate.
Up to $20 a month floor
Production pay as you go
Teams that need every region, RBAC, backups and SSO options: Pinecone Standard and Weaviate Flex, both billing usage above the minimum.
$45 to $50 a month minimum
Dedicated clusters
Steady high-QPS workloads on reserved compute: Zilliz Dedicated, Qdrant Cloud Standard, MongoDB Atlas dedicated tiers.
From $197 a month
Enterprise
SLAs, private networking, customer-managed keys and HIPAA: Pinecone Enterprise, turbopuffer Scale and Enterprise, Weaviate Premium, Qdrant Premium.
$256 to $4,096+ a month minimum, or contact sales
Pricing Comparison
| Service | Free tier | Paid entry | Storage | Reads (queries) | 10M vectors, 1M queries a month |
|---|---|---|---|---|---|
Pinecone | Starter: 2 GB, 1M read units a month | Builder $20/month flat; Standard $50/month minimum | $0.33 per GB a month | $16 to $18 per 1M read units; 1 RU per GB searched | $1,165.96 in one namespace |
100,000 objects, 1 GB memory, 10 GB disk | Flex from $45/month | From $0.12 per GiB | Not metered; from $0.00465 per 1M stored dimensions | About $79.40 | |
None | Launch $16/month minimum; Scale $256/month | Up to $0.33 per GB | Up to $1 per PB queried, 1.28 GB minimum per query | About $63.44 plus data returned | |
Starter: no fee, $5 in credits | Team $250/month with $100 credits | $0.33 per GiB a month | $0.0075 per TiB queried; $0.09 per GiB returned | About $509.01 | |
5 GB, 2.5M vCUs a month | Serverless usage-based; Dedicated Enterprise from $197/month | Included in dedicated CU capacity | Serverless vCU rates vary by region; dedicated not metered | Roughly $320 on capacity-optimized, our estimate | |
0.5 vCPU, 1 GB RAM, 4 GB disk | Standard, billed hourly by vCPU, RAM and disk | Billed by GB of disk | Not metered | Check the current price in Qdrant's calculator | |
10K queries or updates a day | $0.40 per 100K requests; Fixed $60/month | $0.25 per GB | $0.40 per 100K requests | Over the 2 billion dimension cap; Pro plan by quote | |
MongoDB Atlas | M0, 512 MB | Flex $8 to $30/month; M10 from $0.08/hour | Included in cluster tier | Not metered; sized by cluster RAM | Depends on the tier that holds the index in RAM |
Supabase (pgvector) | Free plan, 500 MB database | Pro $25/month plus compute | 8 GB disk included, then $0.125 per GB | Not metered | $425 for 1M vectors on 2XL compute |
Prices as of October 30, 2026 from Pinecone pricing and Pinecone cost docs, Weaviate pricing, turbopuffer pricing, Chroma pricing and Chroma Cloud pricing docs, Zilliz Cloud pricing, Qdrant pricing, Upstash Vector pricing, MongoDB Atlas pricing, Supabase pricing and Supabase's compute guide for vector workloads. The last column is our arithmetic on 10 million 1536-dim vectors with 1 KB of metadata (71.4 GB) and 1 million queries a month, at the lowest listed rate. Pinecone's figure uses the low end of its regional range. The Zilliz figure is an estimate: its per-CU capacity is quoted for 768-dim vectors. The Supabase figure is for 1 million vectors, the size its own benchmark fits on 2XL compute.
Mistakes to Avoid
- ×
Putting every document in one big Pinecone namespace. With all 10M records together, each query costs 71.4 read units; split into 100 tenant namespaces, the same query costs under 1.
- ×
Comparing Weaviate's per-dimension rate with Pinecone's per-GB storage rate. Weaviate's dimension charge replaces query fees; Pinecone's storage rate is the smallest part of its bill.
- ×
Forgetting that switching embedding models means re-writing every vector: a full reload of 10M records costs $285.60 in Pinecone write units alone.
- ×
Choosing Upstash Vector for a 10M-vector index: pay-as-you-go caps at 2 billion vector dimensions, about 1.3 million 1536-dim vectors (our arithmetic).
- ×
Adding metadata filters on Chroma without pricing them: each metadata or full-text predicate counts as an extra query.
Expert Tips
- →
Use 768 or 1024-dimension embeddings when recall allows. Dimensions drive storage on every service and the dimension charge on Weaviate, so halving them roughly halves those lines.
- →
On Pinecone Builder, the $20 flat plan includes 2M read units a month. With a 7 GB namespace that covers about 280,000 queries (our arithmetic), so it suits internal tools, not public search.
- →
If you already run Postgres, price pgvector first: Supabase's own benchmark fits 1 million 1536-dim vectors on its 2XL compute, $410 a month on top of the $25 Pro plan. See our serverless Postgres pricing guide for other hosts.
- →
Compare the leaders head to head: Pinecone vs Weaviate, Pinecone vs Qdrant and Pinecone alternatives. For the ranked list by features rather than cost, see the best vector databases guide.
- →
Running your own embedding or reranking model too? The GPU cloud pricing guide prices the hardware side of the same stack.
Red Flags to Watch For
-
A pricing page that shows only a calculator: you cannot compare rates without entering a workload, and the estimate is not a quote.
-
"Free for a promotional period" on data transfer, as Weaviate states for now, with charges possible later after notice.
-
Per-query pricing with no published rule for how the size of the searched data affects the charge.
-
Unused monthly minimums that do not roll over, as turbopuffer states for its commitments.
The Bottom Line
For a single 10-million-vector index with steady traffic, turbopuffer and Weaviate produce the smallest bills at list price, under $80 a month each in our arithmetic. Weaviate is the easier one to forecast because queries are not metered; turbopuffer gets cheaper per GB as namespaces grow past 32 GB.
Pinecone is the market default and costs the most for one large namespace, but its read pricing rewards multi-tenant designs: split by customer and the same workload sits inside the Standard minimum. Chroma Cloud suits small or heavily partitioned collections. Teams with steady high query volume should price dedicated capacity on Zilliz or Qdrant, and teams already on Postgres should start with pgvector on Supabase or their current host. Browse every option in the vector database category.
Frequently Asked Questions
How much does a vector database cost per month in 2026?
For 10 million 1536-dim vectors and 1 million queries a month, list rates as of October 30, 2026 give about $63 on turbopuffer, $79 on Weaviate Flex, $509 on Chroma Cloud and $1,166 on Pinecone Standard with one namespace (our arithmetic). Small apps pay the plan minimum instead, which on these four ranges from nothing on Chroma's Starter plan to Pinecone Standard's minimum.
What does 1 million vectors cost?
With 1536-dim vectors, 1 KB of metadata and 1 million queries a month: about $5.79 on Upstash pay as you go, $9.50 of usage on turbopuffer (billed at the Launch minimum), $7.94 on Weaviate (billed at the Flex minimum), $50.90 on Chroma Cloud and $116.60 on Pinecone Standard. All figures are our arithmetic on October 2026 list rates.
Is there a free vector database?
Why is Pinecone so expensive for large indexes?
Because a Pinecone query costs 1 read unit for every GB of the namespace it searches. A 71.4 GB namespace makes each query cost 71.4 read units, which at the Standard read-unit rate comes to over $1,100 a month for 1 million queries. Splitting the data into many small namespaces cuts the per-query cost to a fraction of a unit.
What does 100 million vectors cost?
At 714 GB of 1536-dim vectors and 1 million queries a month, our arithmetic on list rates gives up to $310.26 on turbopuffer, which discounts queried data above 32 GB and 128 GB, about $794 on Weaviate Flex, and about $11,660 on Pinecone Standard in one namespace. Zilliz's tiered-storage clusters start at $5 per million 768-dim vectors a month.
Is self-hosting a vector database cheaper?
On licences, yes: Qdrant and Milvus are Apache 2.0 and pgvector uses the PostgreSQL license. You still pay for servers with enough RAM for the index (raw 1536-dim vectors alone take 6,144 bytes each) plus the time to run upgrades, backups and monitoring, which managed services include.
Pinecone or Weaviate: which is cheaper?
For one large index with heavy traffic, Weaviate Flex, because it does not meter queries: about $79 a month against $1,166 on Pinecone for our 10M-vector workload. For many small tenant namespaces, Pinecone can fall to its Standard minimum, a few dollars above Weaviate's Flex floor.
Cite this page: Toolradar, "Vector Database Pricing Compared (2026)", updated November 2026, https://toolradar.com/guides/vector-database-pricing-compared
Spotted an outdated price or a wrong fact? Suggest a correction: we check it against the vendor's own page. Work for one of these vendors? Claim your listing to update it yourself.
Related Guides
Ready to Choose?
Compare features, read reviews, and find the right tool.
Some offers on this page may be paid placements or contain affiliate links.