Vector databases - verified
We read the docs, pricing pages, READMEs and license files of 24 vector databases on October 18, 2026: 11 managed services, 5 cloud provider services, 5 open-source engines you run yourself and 3 Postgres extension entries. Every value carries a verbatim quote and its source. Prices stay in each product's own unit, grouped by unit and never converted or ranked across units.
Choosing one for RAG or an agent? Read the vector database comparison. Still picking the vectors? Compare embedding models and rerankers.
Renames, moves, deprecations and license changes seen on October 18, 2026, each only as far as its quotes go. A research note is the researcher's record of the page, shown with a link, not a quote. The pages give no date for the renames and moves.
Google's Vector Search docs and pricing now live on Gemini Enterprise Agent Platform pages, and the pricing page adds an Agent Retrieval section priced per CU-hour and per read operation. The pages state no date for the change.
“Leveraging the ScaNN algorithm, Vector Search lets you build next-generation search and recommendation systems as well as generative AI applications.”
“Storage-Optimized CU $2.30 / 1 hour”
Research note: Vertex AI documentation and pricing URLs now redirect to "Gemini Enterprise Agent Platform" pages (under docs.cloud.google.com/gemini-enterprise-agent-platform). Source page
MongoDB's docs call the feature MongoDB Vector Search, and the old atlas-vector-search overview URL redirects to the new docs path. No date is stated.
“You can also use MongoDB Vector Search with self-managed or local Atlas deployments that you create with the Atlas CLI.”
Research note: The docs now call the feature "MongoDB Vector Search"; the old atlas-vector-search overview URL redirects to mongodb.com/docs/vector-search/. Source page
The Azure pricing page presents the service as Foundry IQ (Azure AI Search). The vector search docs we read still call it Azure AI Search.
“Foundry IQ (Azure AI Search)”
“Azure AI Search defines hybrid search as the execution of vector search and keyword search in the same request.”
Cloudflare's limits page keeps a separate section for V1 indexes, headed as deprecated, where the maximum is 200,000 vectors per index. The page states no date for the deprecation.
“Limits for V1 indexes (deprecated)”
“Maximum vectors per index · 200,000”
Research note: V1 indexes are deprecated and had lower limits (200,000 vectors per index). Source page
The repository LICENSE file puts most source code under BSD-3-Clause and the "wl" directory under a separate enterprise license only.
“Source code in this repository is variously licensed under the BSD-3-Clause License, or the Weaviate License.”
“Within the "wl" directory and all its sub-directories, source code is Copyright Weaviate BV 2026-present, all rights reserved, and is only available under a separate enterprise license.”
AWS now distinguishes NextGen and Classic vector search collections. The pricing page bills NextGen collections for compute and storage separately; the per-OCU rate itself appears only in the pricing calculator.
“With Amazon OpenSearch Serverless NextGen collections, you only pay for the resources consumed by the workload. OpenSearch Serverless charges for compute and storage separately.”
“With Amazon OpenSearch Serverless Classic collections, you only pay for the resources consumed by the workload.”
Research note: AWS now distinguishes NextGen and Classic vector search collections; NextGen collections have a 10-second read-after-write refresh interval per the docs. Source page
S3 Vectors has two index modes, ENHANCED and CLASSIC, and AWS recommends ENHANCED for almost all workloads. Vector buckets created on or after September 30, 2026 default to ENHANCED and cannot use CLASSIC.
“The modes are ENHANCED and CLASSIC. We recommend ENHANCED for almost all workloads, because it gives you higher recall on filtered queries and additional filter operators.”
“A vector bucket created on or after September 30, 2026 has a default index mode of ENHANCED, and indexes in it cannot use CLASSIC.”
Research note: Index modes: vector buckets created on or after September 30, 2026 default to ENHANCED and cannot use CLASSIC, per the vector index docs. Source page
Each card shows the first price in the product's own unit, then max dimensions, hybrid search, built-in embedding and reranking, BYOC or self-hosting and license. Every value links the page it came from.
Vector databases run for you by the company that builds them, serverless or on provisioned capacity. Several also publish an open-source engine, listed separately below.
managed serverless
How it charges
Free - plan price (Starter)
“Starter For trying out and for small applications. Start for Free Free”
managed provisioned - status not stated
How it charges
$0 - plan price per month (Free)
“Free Forever$0/mo No credit card required”
managed provisioned - status not stated
How it charges
Free - plan price (Free Tier)
“Free Tier Free forever For testing, and prototypes”
managed serverless - status not stated
How it charges
$0 - plan price (Free)
“Free Starting point for learning and personal projects. $0”
managed serverless - status not stated
How it charges
$0 - plan price per month plus usage (Starter)
“Starter $0/month + usage Get up and running quickly.”
managed serverless - status not stated
How it charges
$16 - minimum usage per month (launch plan)
“Sign up $16/month minimum usage scale”
managed provisioned - status not stated
How it charges
No price listed.No public price list: lancedb.com/pricing resolved to the LanceDB contact page on 2026-10-18.
managed provisioned - status not stated
How it charges
$0 - per hour (Atlas Free cluster)Cluster price
“Free $0 /hour Free forever For learning and exploring MongoDB in a cloud environment. STORAGE 512 MB”
managed serverless - status not stated
How it charges
As low as $0.14 - per VCU-hour ingest (Serverless)
“Ingest*Per VCU per hour As low as $0.14”
managed serverless - status not stated
How it charges
$0 - plan price (Free)
“Free $0 - Perfect for prototypes and hobby projects. Daily Query/Update Limit 10K”
managed provisioned - status not stated
How it charges
$0.05 - per vCPU-hour (Startup plan; memory $0.005/GB-hour, disk $0.0002/GB-hour, GPU memory $0.03/GB-hour)
“Unit prices - vCPU$ 0.05 / hour - Memory GB$ 0.005 / hour - Disk GB$ 0.0002 / hour - GPU Memory GB$ 0.03 / hour”
Vector search inside a cloud platform's own services: AWS, Azure, Google Cloud and Cloudflare. Billed on that platform's account.
cloud-native service - status not stated
How it charges
Not statedThe pricing page describes OCU-hour billing ("One OCU comprises approximately 6 GB of RAM and corresponding vCPU"), but the per-OCU rates appear only in its pricing calculator, not in the page text.
Every value, quote and source for Amazon OpenSearch Serverless
cloud-native service - status not stated
How it charges
$2.50 - per 1M queries (US East example)AWS US East worked example
“Queries are charged based on a per-query request fee ($2.50/MM queries)”
managed provisioned - status not stated
How it charges
$0 / $73.73 / $245.28 / $981.12 / $1,962.24 / $2,802.47 / $5,604.21 - per search unit per month (Free, Basic, S1, S2, S3, L1, L2; as displayed with the page default region and currency)Page default region, not stated
“Price per SU (Search unit) $0/month $73.73/month $245.28/month $981.12/month $1,962.24/month $2,802.47/month $5,604.21/month”
managed provisioned - status not stated
How it charges
$3.00 - per GiB processed for index build and update
“Vector Search charges $3.00 per gibibyte (GiB) of data processed in all regions.”
managed serverless - status not stated
How it charges
$0.01 - per 1M queried vector dimensions (Workers Paid, after 50M included)
“First 50 million queried vector dimensions / month included + $0.01 per million”
Engines you install and operate on your own infrastructure under the license shown. No price applies to the software itself.
open source self-host - status not stated
How it charges
No price listed.Open source; no price. Managed Milvus is offered as Zilliz Cloud (separate entry).
open source self-host - status not stated
How it charges
No price listed.Open source; no price.
open source self-host - status not stated
How it charges
No price listed.Open source; no price.
open source self-host - status not stated
How it charges
No price listed.Open source; no price.
open source self-host - status not stated
How it charges
No price listed.This entry covers the open-source engine. Redis Cloud plans are not vector-specific and are outside this entry, so no price is listed.
pgvector adds vector columns, distance operators and ANN indexes to Postgres. The hosted entries bill it as part of their Postgres plans.
database extension - status not stated
How it charges
No price listed.Open source; no price. Hosting cost depends on the Postgres provider.
database extension - status not stated
How it charges
No price listed.Billed as part of Supabase Postgres plans; no vector-specific price on the pages checked.
database extension - status not stated
How it charges
No price listed.Billed as part of Neon Postgres plans; no vector-specific price on the page checked.
Every price row, in each product's own unit, one block per unit. Rows are sorted by amount only inside a block; nothing is converted, and a price in one block is never compared with another. Plan labels in brackets come from the pricing table's column headings, which the quoted cell text does not always repeat.
| Product | Price | Unit, as stated | Quote and source | Where it comes from |
|---|---|---|---|---|
| Per GB-month stored | ||||
| Elasticsearch | As low as $0.047 | per GB-month retained (Serverless) | “Storage & RetentionPer GB retained per month As low as $0.047” | - |
| Amazon S3 Vectors | $0.06 | per GB-month logical storage (US East example) | “Total monthly storage cost = 59 GB * $0.06/GB per month = $3.54” |
|
| Pinecone | $0.33 | per GB-month storage (Standard and Enterprise) | “Storage Up to 2 GB Up to 10 GB Unlimited $0.33/GB/mo Unlimited $0.33/GB/mo” | - |
| Per GiB-month stored | ||||
| Chroma Cloud | $0.33 | per GiB-month stored | “Storageper GiB per month $0.33” | - |
| Per 1M read or write units | ||||
| Pinecone | $4-$4.50 | per 1M write units (Standard; varies by cloud and region) | “Unlimited $4-$4.50 per million (varies by cloud and region)” | - |
| Pinecone | $6-$6.75 | per 1M write units (Enterprise; varies by cloud and region) | “Unlimited $6-$6.75 per million (varies by cloud and region)” | - |
| Pinecone | $16-$18 | per 1M read units (Standard; varies by cloud and region) | “Unlimited $16-$18 per million (varies by cloud and region)” | - |
| Pinecone | $24-$27 | per 1M read units (Enterprise; varies by cloud and region) | “Unlimited $24-$27 per million (varies by cloud and region)” | - |
| Per 1M queries | ||||
| Amazon S3 Vectors | $2.50 | per 1M queries (US East example) | “Queries are charged based on a per-query request fee ($2.50/MM queries)” |
|
| Per 100K requests | ||||
| Upstash Vector | $0.4 | per 100K requests (Pay as You Go; query and upsert priced the same) | “Pay as You Go $0.4 per 100K requests” | - |
| Per 100,000 read operations | ||||
| Google Vector Search | $0.06 | per 100,000 read operations (Agent Retrieval payload database) | “Read Operations $0.06 / 100,000 count” |
|
| Per search unit per month | ||||
| Azure AI Search | $0 / $73.73 / $245.28 / $981.12 / $1,962.24 / $2,802.47 / $5,604.21 | per search unit per month (Free, Basic, S1, S2, S3, L1, L2; as displayed with the page default region and currency) | “Price per SU (Search unit) $0/month $73.73/month $245.28/month $981.12/month $1,962.24/month $2,802.47/month $5,604.21/month” |
|
| Per CU-hour | ||||
| Google Vector Search | $0.065 | per CU-hour (Agent Retrieval index serving, Performance-Optimized) | “Performance-Optimized CU $0.065 / 1 hour” |
|
| Google Vector Search | $2.30 | per CU-hour (Agent Retrieval index serving, Storage-Optimized) | “Storage-Optimized CU $2.30 / 1 hour” |
|
| Per vCPU-hour | ||||
| Vespa Cloud | $0.05 | per vCPU-hour (Startup plan; memory $0.005/GB-hour, disk $0.0002/GB-hour, GPU memory $0.03/GB-hour) | “Unit prices - vCPU$ 0.05 / hour - Memory GB$ 0.005 / hour - Disk GB$ 0.0002 / hour - GPU Memory GB$ 0.03 / hour” | - |
| Vespa Cloud | $0.1 | per vCPU-hour (Basic, initial) | “Initial unit prices - vCPU$ 0.1 / hour - Memory GB$ 0.01 / hour - Disk GB$ 0.0004 / hour - GPU Memory GB$ 0.07 / hour” | - |
| Vespa Cloud | $0.145 | per vCPU-hour (Commercial, initial) | “Initial unit prices - vCPU$ 0.145 / hour - Memory GB$ 0.0145 / hour - Disk GB$ 0.0005 / hour - GPU Memory GB$ 0.1 / hour” | - |
| Vespa Cloud | $0.18 | per vCPU-hour (Enterprise, initial) | “Initial unit prices - vCPU$ 0.18 / hour - Memory GB$ 0.018 / hour - Disk GB$ 0.0007 / hour - GPU Memory GB$ 0.125 / hour” | - |
| Per VCU-hour | ||||
| Elasticsearch | As low as $0.07 | per VCU-hour machine learning (Serverless) | “Machine LearningPer VCU per hour As low as $0.07” | - |
| Elasticsearch | As low as $0.09 | per VCU-hour search (Serverless) | “Search*Per VCU per hour As low as $0.09” | - |
| Elasticsearch | As low as $0.14 | per VCU-hour ingest (Serverless) | “Ingest*Per VCU per hour As low as $0.14” | - |
| Per cluster-hour (whole cluster, not vector-specific) | ||||
| MongoDB Vector Search | $0 | per hour (Atlas Free cluster) | “Free $0 /hour Free forever For learning and exploring MongoDB in a cloud environment. STORAGE 512 MB” |
|
| MongoDB Vector Search | $0.011 | per hour, up to $30/month (Atlas Flex cluster) | “Flex $0.011 /hour Up to $30/month” |
|
| MongoDB Vector Search | $0.08 | per hour, starts at $56.94/month (Atlas Dedicated cluster) | “Dedicated $0.08 /hour Starts at $56.94/month” |
|
| Per 1M vectors per month | ||||
| Zilliz Cloud | From $5 | per million vectors per month (Dedicated Tiered-storage) | “Search Latency 100-1000ms From $5 per million vectors / month” | - |
| Zilliz Cloud | From $16 | per million vectors per month (Dedicated Capacity-optimized) | “Search Latency 50-100 ms From $16 per million vectors / month” | - |
| Zilliz Cloud | From $63 | per million vectors per month (Dedicated Performance-optimized) | “Search Latency 10 ms From $63 per million vectors / month” | - |
| Per 1M queried vector dimensions | ||||
| Cloudflare Vectorize | $0.01 | per 1M queried vector dimensions (Workers Paid, after 50M included) | “First 50 million queried vector dimensions / month included + $0.01 per million” | - |
| Per 100M stored vector dimensions | ||||
| Cloudflare Vectorize | $0.05 | per 100M stored vector dimensions (Workers Paid, after 10M included) | “First 10 million stored vector dimensions + $0.05 per 100 million” | - |
| Per 1M vector dimensions (period not stated) | ||||
| Weaviate Cloud | from $0.00465 | per 1M vector dimensions (Flex; Premium shared from $0.003875, Premium dedicated from $0.002718) | “Vector dimensions Free from $0.00465 / 1M from $0.003875 / 1M from $0.002718 / 1M” |
|
| Plan price or monthly minimum | ||||
| Chroma Cloud | $0 | plan price per month plus usage (Starter) | “Starter $0/month + usage Get up and running quickly.” | - |
| Pinecone | Free | plan price (Starter) | “Starter For trying out and for small applications. Start for Free Free” | - |
| Qdrant Cloud | Free | plan price (Free Tier) | “Free Tier Free forever For testing, and prototypes” | - |
| Upstash Vector | $0 | plan price (Free) | “Free $0 - Perfect for prototypes and hobby projects. Daily Query/Update Limit 10K” | - |
| Weaviate Cloud | $0 | plan price per month (Free) | “Free Forever$0/mo No credit card required” | - |
| Zilliz Cloud | $0 | plan price (Free) | “Free Starting point for learning and personal projects. $0” | - |
| Zilliz Cloud | From $0 | per month (Standard, Serverless) | “From $0 /month (Serverless)” | - |
| turbopuffer | $16 | minimum usage per month (launch plan) | “Sign up $16/month minimum usage scale” | - |
| Pinecone | $20 | plan price per month, flat (Builder) | “Builder NEW For solo developers and small teams. Get Started $20/month flat” | - |
| Weaviate Cloud | $45 | plan minimum per month (Flex, starts at) | “Flex Starts at$45/mo Pay-as-you-go” | - |
| Pinecone | $50 | minimum usage per month (Standard) | “You'll be charged a minimum of $50/month. Once your usage exceeds this amount, you'll pay as you go.” | - |
| Upstash Vector | $60 | plan price per month (Fixed) | “Fixed $60/ month - For consistent loads with predictable costs.” | - |
| Zilliz Cloud | From $197 | per month (Enterprise, Dedicated) | “From $197 /month (Dedicated)” | - |
| Chroma Cloud | $250 | plan price per month plus usage (Team) | “Team $250/month + usage Scale your production use cases. $100 credits then usage-based pricing.” | - |
| turbopuffer | $256 | minimum usage per month (scale plan) | “Sign up $256/month minimum usage enterprise” | - |
| Weaviate Cloud | from $400 | plan minimum per month (Premium) | “The minimum covers the baseline cluster cost and includes charges for Vector Dimensions and Storage: $45/month on Flex and from $400/month on Premium.” | - |
| Pinecone | $500 | minimum usage per month (Enterprise) | “You'll be charged a minimum of $500/month. Once your usage exceeds this amount, you'll pay as you go.” | - |
| turbopuffer | $4,096 | per month, with a 35% usage premium (enterprise plan) | “Contact us =$4,096/month 35% usage premium” | - |
| Vespa Cloud | $20,000 | minimum monthly spend (Enterprise) | “Minimum monthly spend of $20,000” | - |
| Qdrant Cloud | Minimum spend required | Premium Tier | “Premium Tier Minimum spend required” | - |
| Other units, one row eachnot sorted: no two rows share a unit | ||||
| Weaviate Cloud | from $0.12 | per GiB storage (Flex; Premium shared from $0.10, Premium dedicated from $0.33) | “Storage Free from $0.12 / GiB from $0.10 / GiB from $0.33 / GiB” | - |
| Weaviate Cloud | from $0.03036 | per GiB backup (Flex; Premium shared from $0.021, Premium dedicated from $0.014) | “Backup Free from $0.03036/ GiB from $0.021 / GiB from $0.014 / GiB” | - |
| Chroma Cloud | $2.50 | per GiB written | “Writeper GiB written $2.50” | - |
| Chroma Cloud | $0.0075 | per TiB queried | “Queryper TiB queried $0.0075” | - |
| Chroma Cloud | $0.09 | per GiB returned | “Networkper GiB returned $0.09” | - |
| Upstash Vector | $0.25 | per GB storage (monthly, daily-average basis) | “The storage cost is charged at a rate of $0.25 per GB total storage.” | - |
| Upstash Vector | $0.03 | per GB bandwidth beyond 200 GB/month (Pay as You Go) | “Beyond that, we charge $0.03 for each additional GB data transfer.” | - |
| Amazon S3 Vectors | $0.004 / $0.002 | per TB processed by query, first 100K vectors / 100K to 10M vectors (US East example) | “a data processed charge ($0.004/TB for the first 100K vectors, $0.002/TB for 100K” |
|
| Amazon S3 Vectors | $0.0004 | per TB processed by query, 10M+ vectors (US East example) | “$0.0004/TB for 10M+ vectors” |
|
| Amazon S3 Vectors | $0.01 | per GB returned (US East example) | “and a data returned charge ($0.01/GB)” |
|
| Amazon S3 Vectors | $0.20 | per GB uploaded (PUT, US East example) | “Total monthly PUT cost = 1 full upload/6 months * 59 GB total storage * $0.20/GB = $1.97” |
|
| Google Vector Search | $3.00 | per GiB processed for index build and update | “Vector Search charges $3.00 per gibibyte (GiB) of data processed in all regions.” | - |
| Google Vector Search | $0.45 | per GiB ingested (streaming updates) | “Vector Search charges $0.45/GiB ingested for Streaming Update inserts.” | - |
| Google Vector Search | Charged per node hour by machine type and region | per node-hour (index serving) | “The price corresponds to the machine type, by region, and is charged per node hour.” | - |
These rates are not in the pricing page text, so they read Not stated. The billing model is quoted where the page states one.
| Product | Rate | Why | Billing model, quoted |
|---|---|---|---|
| Qdrant Cloud | Not statedStandard Tier unit rates | Standard Tier unit prices are only available through the Qdrant pricing calculator, not as text on the pricing page. | Usage-based “Standard Tier Usage-based pricing For production workloads and scaling applications” |
| Zilliz Cloud | Not statedServerless read and write vCU rates | Serverless read/write vCU unit rates are not in the pricing page text; the page gives them only through its pricing calculator. | See the plan minimums above. |
| turbopuffer | Not statedPer-GB storage, write and query rates | Per-GB storage, write and query rates appear only in the pricing page's cost calculator, not in the page text, so no per-unit rate is listed here. | See the plan minimums above. |
| Amazon OpenSearch Serverless | Not statedPer-OCU compute rates | The pricing page describes OCU-hour billing ("One OCU comprises approximately 6 GB of RAM and corresponding vCPU"), but the per-OCU rates appear only in its pricing calculator, not in the page text. | Compute (OCUs) and storage billed separately “With Amazon OpenSearch Serverless NextGen collections, you only pay for the resources consumed by the workload. OpenSearch Serverless charges for compute and storage separately.” |
Open-source software, Postgres extensions billed inside a Postgres plan, and a contact-only product.
No public price list: lancedb.com/pricing resolved to the LanceDB contact page on 2026-10-18.
Open source; no price. Managed Milvus is offered as Zilliz Cloud (separate entry).
Open source; no price.
Open source; no price.
Open source; no price.
This entry covers the open-source engine. Redis Cloud plans are not vector-specific and are outside this entry, so no price is listed.
Open source; no price. Hosting cost depends on the Postgres provider.
Billed as part of Supabase Postgres plans; no vector-specific price on the pages checked.
Billed as part of Neon Postgres plans; no vector-specific price on the page checked.
Every product against every field, each stated cell linking its source. A cell shows the vendor's words, shortened; hover for the full value, or open the product page for the quote. Not stated is not the same as no. This is not a ranking.
Vector size, distance metrics, index algorithms, keyword or sparse search, filters and tenant isolation.
Embedding and reranking inside the database, where it can run, its license and official SDKs.
16 figures as the vendors state them. They are claims, not measurements, and are not compared with each other.
| Product | Claim | Quote and source |
|---|---|---|
| Pinecone | 99.95% uptime SLA on Enterprise vendor-statedinferredRead as the Enterprise plan: the quoted list opens with Everything in Standard, the pricing page's lead-in for the tier above Standard. | “Everything in Standard 99.95% Uptime SLA” |
| Weaviate Cloud | Availability SLA: best effort (Free), 99.5% (Flex), 99.9% (Premium shared), 99.95% (Premium dedicated) vendor-statedinferredPer-plan SLAs (Free, Flex, Premium shared, Premium dedicated) mapped by the pricing table's column order; the quoted row does not name the plans. | “Availability Best effort 99.5% 99.9% 99.95%” |
| Qdrant Cloud | 99.5% uptime SLA on Standard Tier vendor-statedinferredThe quote does not name the tier; Standard comes from the pricing card it sits in. | “99.5% Uptime SLA” |
| Qdrant Cloud | 99.9% uptime SLA on Premium Tier vendor-statedinferredThe quote does not name the tier; Premium comes from the pricing card it sits in. | “99.9% Uptime SLA” |
| Zilliz Cloud | Performance-optimized CU: about 2 million vectors, 500-1500 QPS, 10 ms search latency (768-dim vectors) vendor-statedinferredThe quoted row does not name the CU type; Performance-optimized and the 768-dimension basis come from the surrounding pricing table and its footnote. | “Per CU Capacity 2 million vectors Search QPS 500-1500 Search Latency 10 ms” |
| Zilliz Cloud | Capacity figures are based on 768-dimensional vectors vendor-stated | “This data is based on evaluations of 768-dimensional vectors.” |
| Zilliz Cloud | 99.95% uptime SLA on Enterprise vendor-stated | “Enterprise-grade reliability and controls. Best for: production applications. From $197 /month (Dedicated) Start 30-day Free Trial - 99.95% uptime SLA” |
| turbopuffer | Vector search recall@10 of 90-100% (limits page) vendor-stated | “Vector search recall@10 90-100%” |
| turbopuffer | 5k+ queries/s per namespace vendor-stated | “Max queries (per namespace) 5k+ queries/s” |
| turbopuffer | 99.95% uptime SLA on enterprise vendor-statedinferredThe quote does not name the plan; enterprise comes from the pricing table column it sits in. | “Uptime SLA99.95%” |
| Elasticsearch | BBQ reduces memory footprint by 96% (32x), at a cost of accuracy vendor-stated | “This reduces the memory footprint by 96% (or 32x) at a larger cost of accuracy.” |
| Upstash Vector | Enterprise plan scales over a billion vectors vendor-statedinferredThe quote does not name the plan; Enterprise comes from the pricing card it sits in. | “Scales over a billion of vectors” |
| Amazon OpenSearch Serverless | NextGen collections create all indexes with 32x compression vendor-statedinferredThe quote itself does not mention NextGen collections. | “All indexes are created with advanced 32x compression technique by default.” |
| Amazon S3 Vectors | Subsecond latency for infrequent queries, as low as 100 ms for frequent queries vendor-stated | “delivers subsecond latency for infrequent queries and as low as 100 milliseconds for more frequent queries” |
| Amazon S3 Vectors | Up to 90% lower costs for uploading, storing and querying vectors vendor-stated | “Amazon S3 Vectors provides cost-effective, elastic, and durable vector storage at up to 90% lower costs for uploading, storing, and querying vectors.” |
| Google Vector Search | A minimal setup under $100 per month can handle high throughput for moderate-sized use cases vendor-stated | “Even a minimal setup (under $100 per month) can accommodate high throughput for moderate-sized use cases.” |
Values that are our reading of a quote rather than the vendor's own words. Each carries an inferred tag wherever it appears.
| Product | Field | Value shown | Our reading |
|---|---|---|---|
| Pinecone | Free tier | Starter plan includes up to 1M read units per month (with storage up to 2 GB and up to 2M write units per month) | The Starter figures are the first column of the pricing table, mapped by the pricing table's column order; the quoted row does not name the plans. |
| Pinecone | BYOC and self-host | BYOC listed on the Enterprise plan | Read as the Enterprise plan: the quoted list opens with Everything in Standard, the pricing page's lead-in for the tier above Standard. |
| Pinecone | Vendor-stated claim | 99.95% uptime SLA on Enterprise | Read as the Enterprise plan: the quoted list opens with Everything in Standard, the pricing page's lead-in for the tier above Standard. |
| Weaviate Cloud | Limits | Free: 100,000 objects and 1 collection; paid plans: unlimited objects, 1000 collections | Free versus paid-plan figures mapped by the pricing table's column order; the quoted row does not name the plans. |
| Weaviate Cloud | Regions and clouds | Free: AWS; Flex and Premium shared: GCP, AWS; Premium dedicated: GCP, AWS, Azure | Per-plan clouds (Free, Flex and Premium shared, Premium dedicated) mapped by the pricing table's column order; the quoted row does not name the plans. |
| Weaviate Cloud | Vendor-stated claim | Availability SLA: best effort (Free), 99.5% (Flex), 99.9% (Premium shared), 99.95% (Premium dedicated) | Per-plan SLAs (Free, Flex, Premium shared, Premium dedicated) mapped by the pricing table's column order; the quoted row does not name the plans. |
| Qdrant Cloud | Vendor-stated claim | 99.5% uptime SLA on Standard Tier | The quote does not name the tier; Standard comes from the pricing card it sits in. |
| Qdrant Cloud | Vendor-stated claim | 99.9% uptime SLA on Premium Tier | The quote does not name the tier; Premium comes from the pricing card it sits in. |
| Zilliz Cloud | Vendor-stated claim | Performance-optimized CU: about 2 million vectors, 500-1500 QPS, 10 ms search latency (768-dim vectors) | The quoted row does not name the CU type; Performance-optimized and the 768-dimension basis come from the surrounding pricing table and its footnote. |
| turbopuffer | Vendor-stated claim | 99.95% uptime SLA on enterprise | The quote does not name the plan; enterprise comes from the pricing table column it sits in. |
| Upstash Vector | Max dimensions | 1,536 (Free), 3,072 (Pay as you go, Fixed), 5,000 (Pro) | Per-plan figures (Free, Pay as you go, Fixed, Pro) mapped by the pricing table's column order; the quoted row does not name the plans. |
| Upstash Vector | Multi-tenancy | Namespaces: 100 (Free), 10K (Pay as you go, Fixed), unlimited (Pro) | Per-plan namespace counts (Free, Pay as you go, Fixed, Pro) mapped by the pricing table's column order; the quoted row does not name the plans. |
| Upstash Vector | Limits | Max vectors x dimensions: 200M (Free), 2B (Pay as you go, Fixed), 100B (Pro) | Per-plan vector-dimension limits (Free, Pay as you go, Fixed, Pro) mapped by the pricing table's column order; the quoted row does not name the plans. |
| Upstash Vector | Regions and clouds | AWS N. Virginia, AWS Ireland, GCP Iowa (varies by plan) | The quote lists three regions; that they vary by plan comes from the pricing table's columns, not the quoted text. |
| Upstash Vector | Vendor-stated claim | Enterprise plan scales over a billion vectors | The quote does not name the plan; Enterprise comes from the pricing card it sits in. |
| Amazon OpenSearch Serverless | Vendor-stated claim | NextGen collections create all indexes with 32x compression | The quote itself does not mention NextGen collections. |
Products and plans considered and left out, with the reason recorded at the time.
Couchbase (vector search)
Optional in scope and not researched in this pass.
Redis Cloud pricing
Redis Cloud plans are not vector-specific and were not checked; the Redis entry covers the open-source engine only.
Supabase Vector Buckets
Not checked; the Supabase entry covers pgvector only.
Our longer reads on vector databases, embeddings and RAG.
Related data
Method
Read from each vendor's docs, pricing pages, READMEs and license files on , with every quote string-matched against the page text that day. Prices and limits change; check the linked source before you commit.
In the middle. An embedding model turns documents and queries into vectors, the vector database stores those vectors and returns the nearest ones for each query, often alongside keyword or sparse search, and a reranker re-scores the top candidates before they reach the model. 12 entries here state built-in embedding and 4 state built-in reranking, so some products cover more than one stage.
Stated on the pages we read: Pinecone, Weaviate Cloud, Qdrant Cloud, turbopuffer, LanceDB Enterprise, MongoDB Vector Search, Elasticsearch, Upstash Vector, Vespa Cloud, Azure AI Search, Milvus and Chroma. How much each does varies, from hosted embedding models to functions that call an external provider; the product pages quote each one.
Stated on the pages we read: Pinecone, Weaviate Cloud, Qdrant Cloud, Zilliz Cloud, Chroma Cloud, turbopuffer, LanceDB Enterprise, MongoDB Vector Search, Elasticsearch, Upstash Vector, Vespa Cloud, Amazon OpenSearch Serverless, Azure AI Search, Google Vector Search, Milvus, Chroma, LanceDB OSS and pgvector. For the rest the pages we checked did not describe it, which is not evidence that it is missing.
It depends on the unit each product bills in: storage per GB-month or GiB-month, read and write units, queries, requests, search units, compute hours, vectors or vector dimensions, or a plan price. Those units do not convert into each other, so this directory groups prices by unit and sorts only inside a unit. Qdrant Cloud, Zilliz Cloud, turbopuffer and Amazon OpenSearch Serverless show some rates only in a pricing calculator, so those read Not stated here.
The open-source engines and the Postgres extension: Milvus (Apache-2.0), Chroma (Apache-2.0), LanceDB OSS (Apache-2.0), OpenSearch (Apache-2.0), Redis (Redis 8+: RSALv2, SSPLv1 or AGPLv3 (your choice); 7.2 and earlier BSD-3) and pgvector (PostgreSQL License). Several managed services also state BYOC or self-managed options; the matrix links each one.
The README says: "Vectors can have up to 16,000 dimensions." That is the stored vector type. For indexed columns it lists: "- vector - up to 2,000 dimensions - halfvec - up to 4,000 dimensions - bit - up to 64,000 dimensions"
Vertex AI Vector Search now sits under Gemini Enterprise Agent Platform. MongoDB Atlas Vector Search is now called MongoDB Vector Search. Azure AI Search is shown as Foundry IQ (Azure AI Search). Cloudflare Vectorize V1 indexes are deprecated. Weaviate's license now splits BSD-3-Clause and an enterprise-licensed directory. Amazon OpenSearch Serverless splits vector collections into NextGen and Classic. S3 Vectors: ENHANCED is the default index mode for buckets created on or after September 30, 2026. Each change is quoted in the What changed section with its source.
No. Every value is a first-party statement copied from the vendor's own pages, with the quote and the link. Performance, scale and SLA figures are labelled vendor-stated; nothing here was measured, and prices in different units are never converted or compared.
Not stated means the pages we checked did not say, so the cell shows the reason rather than a guess; it is not evidence that the feature is missing. Inferred marks our reading of a quote, such as a plan name taken from a pricing table's column order when the quoted row does not name it.
Last verified from each vendor's own docs, pricing pages, READMEs and license files; every quote was string-matched against the page that day. This is a dated snapshot, not a benchmark: prices are shown in each product's own unit and change without notice, so check the linked source before you commit.
Snapshot of October 18, 2026: 24 products, 360 quotes, 66 price rows, 13 with an open-source license, 3 considered and not included.
Published from the verified research file of 2026-10-18 with no change to any value or quote. Before publishing we removed every reference to the researcher's local copies of the source pages (the saved-file path recorded for each source and the folder named in each quoteCheck method), and rewrote the notes that described the reading process so they describe the source instead: the turbopuffer, Amazon OpenSearch Serverless (note and the unit line of its one price row) and Zilliz Cloud notes on calculator-only rates, the Milvus reranking, LanceDB embedding and three Azure AI Search null notes, the MongoDB tabbed-page note, the three Redis notes on what the entry covers, and the Couchbase skip reason. Supabase's two null fields moved from a separate nullNotes map into the field shape every other entry uses, with the same notes.
Newsletter
One email per week. New model releases, retrieval tooling and pricing changes and what they mean for what you build.
The step before the database: dimensions, max input and price per 1M tokens.
The step after it: re-score the candidates a vector search returns.
LangGraph, CrewAI, OpenAI Agents SDK and more, for wiring retrieval into an agent.
Our guide to choosing a vector database for RAG and agents.