AWS S3 Vectors alternatives

19 tools we tested head to head against AWS S3 Vectors, ranked — and what each one actually does differently.

last reviewed 23 jul 2026 · from our best 20 vector databases ·list curated by Onur Ozcanx in

first — what you'd be leaving

AWS S3 Vectors ranks #7 of 20 in our vector databases testing. storage at six cents a gigabyte, and it reprices the floor of the market.

80/100

roughly five times cheaper per gigabyte than the premium managed options, and explicitly built for cost rather than latency — choose it deliberately, not by default.

why people look for an alternative

−sub-second latency — wrong for interactive agents
−128KB minimum per PUT punishes unbatched writes ~21×
−index algorithm and internals undocumented

stay with AWS S3 Vectors if $0.06/GB/month — the cheapest credible storage here is the thing you care about most — nothing below beats it on that.

the short version

best alternativepgvectoralmost everyone — start here and migrate on a measured threshold, not a vendor's table92/100

pgvector

#1 in vector databases · the correct default for most teams, and it quietly got good

92/100

verdictfree, permissive, runs on every managed postgres, and the only option here that can update an embedding and its parent row in one transaction.

pgvector vs AWS S3 Vectors

	AWS S3 Vectors	pgvector
price	$0.06 per GB per month storage	free — PostgreSQL License
free tier	no	yes
license	proprietary, managed only	PostgreSQL License — permissive
deployment	aws managed	extension on any postgres
index types	not documented	hnsw, ivfflat, binary quantization
hybrid search	not a strength	via postgres full-text; you write the fusion
pricing model	per gb stored, per million queries	free — you pay for postgres

switch foralmost everyone — start here and migrate on a measured threshold, not a vendor's table

pros

+free and permissively licensed, on every managed postgres
+transactional consistency between embeddings and business rows
+0.8.0 iterative index scans fixed the filtered-recall flaw
+half-precision and binary quantization cut footprint dramatically

cons

−index builds are slow and lock-heavy at large scale
−no native distributed sharding
−you own all the tuning yourself

visit pgvector →full pgvector review →

Qdrant

#2 in vector databases · filtering built into the graph traversal, not bolted around it

90/100

verdictthe best dedicated option: apache-2.0 with no strings, the strongest filtered-search design in the category, and the easiest serious self-host.

Qdrant vs AWS S3 Vectors

	AWS S3 Vectors	Qdrant
price	$0.06 per GB per month storage	free tier; cloud billed hourly on compute and storage
free tier	no	yes
license	proprietary, managed only	Apache-2.0
deployment	aws managed	self-host or managed cloud
index types	not documented	hnsw + scalar/binary/product/turboquant
hybrid search	not a strength	sparse vectors, native
pricing model	per gb stored, per million queries	hourly compute + storage, rates unpublished

switch forteams who genuinely need a dedicated vector database and want to self-host it

pros

+filters applied inside hnsw traversal — best filtered search here
+apache-2.0 with no revenue caps or service restrictions
+four quantization modes including turboquant
+single rust binary — easiest serious self-host

cons

−cloud per-unit rates are not published
−hybrid and private cloud are quote-only
−smaller enterprise support footprint than the incumbents

visit Qdrant →full Qdrant review →

Milvus / Zilliz Cloud

#3 in vector databases · the credible answer at a billion vectors

88/100

verdictthe widest index selection anywhere and genuine billion-scale — and version 2.6 removed the external message queue that made it painful to run.

Milvus / Zilliz Cloud vs AWS S3 Vectors

	AWS S3 Vectors	Milvus / Zilliz Cloud
price	$0.06 per GB per month storage	free self-hosted; zilliz cloud billed per compute unit
free tier	no	yes
license	proprietary, managed only	Apache-2.0
deployment	aws managed	self-host, lite, or zilliz cloud
index types	not documented	hnsw, ivf, diskann, gpu cagra, scann, rabitq
hybrid search	not a strength	full, with sparse indexes
pricing model	per gb stored, per million queries	compute units; rates largely unpublished

switch for100m to 1bn+ vectors, where nothing else on this list is really appropriate

pros

+widest index selection: diskann, gpu cagra, rabitq, ivf family, sparse
+genuinely proven at a billion vectors and beyond
+2.6 removed the kafka/pulsar dependency
+apache-2.0 with over 100k collections for multi-tenancy

cons

−heaviest self-host here despite 2.6's improvements
−over-engineered below roughly 10m vectors
−the vendor also maintains the main public benchmark

visit Milvus / Zilliz Cloud →full Milvus / Zilliz Cloud review →

Elasticsearch

#4 in vector databases · the licence objection to using it for vectors is largely obsolete

85/100

verdictbest-in-class hybrid retrieval with transparent serverless pricing — and its dense and sparse vector code sits in the agpl-licensed tree, not the restricted one.

Elasticsearch vs AWS S3 Vectors

	AWS S3 Vectors	Elasticsearch
price	$0.06 per GB per month storage	serverless from $0.09 per search VCU-hour; storage $0.047/GB/month
free tier	no	yes
license	proprietary, managed only	AGPLv3 / SSPLv1 / Elastic 2.0 triple
deployment	aws managed	self-host, cloud, serverless
index types	not documented	lucene hnsw, int8/int4, bbq
hybrid search	not a strength	best in class — bm25 + rrf
pricing model	per gb stored, per million queries	vcu-hours + storage per gb

switch forteams already running elasticsearch, and anyone who needs real hybrid search

pros

+vector search is available under agplv3, not licence-restricted
+best-in-class hybrid with real bm25 and rank fusion
+transparent serverless pricing with 50gb free on vector profiles
+better binary quantization and mature filtering

cons

−jvm heap tuning and real operational weight
−licence structure is complex even where it's permissive
−the elser sparse model is separately restricted

visit Elasticsearch →full Elasticsearch review →

Weaviate

#5 in vector databases · multi-tenancy and hybrid search, priced per million dimensions

84/100

verdictthe strongest combination of hybrid search and per-tenant isolation here — with a pricing model that punishes large embeddings in a way no competitor's does.

Weaviate vs AWS S3 Vectors

	AWS S3 Vectors	Weaviate
price	$0.06 per GB per month storage	$45 / month minimum; vector dimensions from $0.00465 per million
free tier	no	yes
license	proprietary, managed only	BSD-3-Clause
deployment	aws managed	self-host, cloud, or byoc
index types	not documented	hnsw + product/binary/scalar quantization
hybrid search	not a strength	first-class bm25 + dense fusion
pricing model	per gb stored, per million queries	per million vector dimensions

switch formulti-tenant saas rag where each customer needs isolation

pros

+native multi-tenancy with isolated shards and cold offloading
+first-class hybrid search with configurable fusion
+bsd-3-clause — clean permissive licence
+excellent documentation

cons

−dimension-based pricing doubles when you double embedding size
−$45/month minimum on the entry cloud tier
−historically memory-hungry without quantization

visit Weaviate →full Weaviate review →

Chroma

#6 in vector databases · the best developer experience, and honest per-unit pricing

82/100

verdictpip install and go, with genuinely transparent pricing — the right call up to a few million vectors and the wrong one well before a hundred.

Chroma vs AWS S3 Vectors

	AWS S3 Vectors	Chroma
price	$0.06 per GB per month storage	$0 base with usage-based billing; $0.33 per GiB-month storage
free tier	no	yes
license	proprietary, managed only	Apache-2.0
deployment	aws managed	embedded, self-host, or cloud
index types	not documented	hnsw
hybrid search	not a strength	metadata + full-text; weaker than rivals
pricing model	per gb stored, per million queries	per gib stored, written and queried

switch forprototypes and small-to-mid production where developer speed matters most

pros

+best developer experience in the category
+genuinely transparent per-unit pricing with no base fee
+apache-2.0
+usage-based billing starts at zero

cons

−hnsw only — thinnest tuning surface here
−unproven above roughly 10m vectors
−local mode and cloud are different code paths

visit Chroma →full Chroma review →

turbopuffer

#8 in vector databases · object-storage economics with the best namespace multi-tenancy

78/100

verdictthe architecture the rest of the category converged on, with genuinely good per-tenant isolation — behind unpublished unit rates and no self-host option.

turbopuffer vs AWS S3 Vectors

	AWS S3 Vectors	turbopuffer
price	$0.06 per GB per month storage	$16 / month minimum usage
free tier	no	no
license	proprietary, managed only	proprietary, managed only
deployment	aws managed	managed; byoc at enterprise
index types	not documented	not fully documented
hybrid search	not a strength	supported
pricing model	per gb stored, per million queries	usage, above plan minimums

switch formulti-tenant products with large, mostly-cold corpora

pros

+excellent economics on large, mostly-cold corpora
+namespace-per-tenant isolation is genuinely well designed
+low $16/month entry minimum
+well-regarded engineering team and public writing

cons

−per-unit rates are not published anywhere fetchable
−closed source with no self-host option
−no free tier, and higher tail latency by design

visit turbopuffer →full turbopuffer review →

OpenSearch

#9 in vector databases · apache-2.0 with no licence anxiety, and serverless finally scales to zero

76/100

verdictthe safe institutional choice on aws — genuinely apache-2.0, with faiss and lucene engines, and the always-on cost floor finally removed.

OpenSearch vs AWS S3 Vectors

	AWS S3 Vectors	OpenSearch
price	$0.06 per GB per month storage	free self-hosted; serverless billed per OCU-hour
free tier	no	yes
license	proprietary, managed only	Apache-2.0
deployment	aws managed	self-host or aws managed/serverless
index types	not documented	hnsw, ivf, faiss + lucene, disk modes
hybrid search	not a strength	supported, mature
pricing model	per gb stored, per million queries	ocu-hours; scales to zero on nextgen

switch foraws-committed teams who want vector search without licence questions

pros

+apache-2.0 throughout with no licence ambiguity
+faiss and lucene engines, plus disk-based and quantized modes
+serverless nextgen scales to zero after 10 minutes
+native on aws with full hybrid search

cons

−vector collections can't share compute units with other workloads
−forked-project ergonomics lag elasticsearch
−compute-unit accounting is confusing

visit OpenSearch →full OpenSearch review →

Vespa

#10 in vector databases · the most sophisticated ranking engine here, with the steepest learning curve

75/100

verdicta full search and ranking engine rather than a vector store — if ranking quality is your product, rank it second; if you just need rag, it's overkill.

Vespa vs AWS S3 Vectors

	AWS S3 Vectors	Vespa
price	$0.06 per GB per month storage	$0.05 per vCPU-hour on the startup plan
free tier	no	no
license	proprietary, managed only	Apache-2.0
deployment	aws managed	self-host or vespa cloud
index types	not documented	hnsw + tensor ranking framework
hybrid search	not a strength	best in class, multi-phase ranking
pricing model	per gb stored, per million queries	per vcpu / gb / gpu hour

switch forproducts where retrieval and ranking quality is the differentiator

pros

+unmatched ranking and hybrid sophistication
+most transparent resource pricing in the category
+apache-2.0 and proven at very large scale
+unit prices decline as allocation grows

cons

−steepest learning curve here by a distance
−$20,000/month minimum on the enterprise tier
−overkill for straightforward rag

visit Vespa →full Vespa review →

Pinecone

#11 in vector databases · genuinely zero-ops, with the hardest pricing here to forecast

73/100

verdictthe zero-operations promise is real and worth money to some teams — but you cannot tune it, cannot self-host it, and cannot easily predict the bill.

Pinecone vs AWS S3 Vectors

	AWS S3 Vectors	Pinecone
price	$0.06 per GB per month storage	free tier up to 2GB; standard from $50/month minimum
free tier	no	yes
license	proprietary, managed only	proprietary, managed only
deployment	aws managed	managed serverless only
index types	not documented	proprietary; parameters not exposed
hybrid search	not a strength	native sparse-dense
pricing model	per gb stored, per million queries	read/write units + storage per gb

switch forteams who will pay a premium to never think about infrastructure

pros

+genuinely zero operational burden
+native sparse-dense hybrid and good namespace multi-tenancy
+2gb free tier
+strong enterprise compliance posture

cons

−unit-based pricing is the hardest here to forecast
−hnsw parameters not exposed — no tuning, no recall/latency trade
−no self-host option at all

visit Pinecone →full Pinecone review →

+ 9 more tested, not detailed here

we ranked 20 vector databases in total. the 9 that didn't make this page are written up in the full ranking →

how these were compared

every tool on this page went through the same test as AWS S3 Vectors — same tasks, same order, scored the same way. the comparison tables are the figures from that testing, not vendor spec sheets.

the vector databases test in full →

was this useful?