Overview
Amazon S3 Vectors adds a new 'vector bucket' type to S3 with dedicated APIs to store embeddings and run similarity queries with metadata filters, without provisioning any cluster. It went GA in December 2025 with up to 2 billion vectors per index and 10,000 indexes per bucket, around 100 ms latency for frequent queries and sub-second for infrequent ones. AWS positions it as the cheap durable tier for RAG and agent memory, integrated with Bedrock Knowledge Bases and able to export into OpenSearch when a workload needs high QPS. It is not a low-latency serving engine and has no keyword search.
Strengths & trade-offs
Strengths
- +Object-storage pricing for billions of vectors with nothing to provision
- +S3 durability, IAM and per-index KMS keys
- +Native backend for Bedrock Knowledge Bases
- +Can tier hot subsets into OpenSearch for low-latency serving
Trade-offs
- −~100 ms warm and up to sub-second cold latency
- −Max 100 results per query and 50 metadata keys per vector
- −No full-text or hybrid keyword search
- −AWS-only, proprietary API
Great for
- +Cheap long-term store for RAG embeddings
- +Agent memory at very large scale
- +Archive tier behind a faster vector engine
- +Bedrock Knowledge Bases backend
Think twice for
- −Low-latency interactive search
- −High-QPS recommendation serving
- −Hybrid keyword plus vector ranking
Capabilities
Native or official first-party support · 6 of 25 tracked
Scale & consistency
- Comfortable data size
- Massive · Multi-PB
- Horizontal scaling
- Native
- Multi-region
- No
- Transactions
- None
- Consistency
- Eventual
- Schema
- Schemaless
- Architecture
- Serverless, Shared storage
- Query languages
- S3 Vectors API (QueryVectors with metadata filters)
Up to 2B vectors per index and 10,000 indexes per vector bucket (20 trillion vectors); available in 14 regions at GA.
Deployment
available on · AWS
Ecosystem
Alternatives to Amazon S3 Vectors
Serverless vector and full-text search on object storage, priced for billions of documents.
amazon-s3-vectors vs turbopuffer →Apache-licensed, foundation-governed fork of Elasticsearch for search, vectors and observability.
amazon-s3-vectors vs opensearch →Fully managed serverless vector database with built-in embedding, reranking and assistants.
amazon-s3-vectors vs pinecone →Embedded, serverless retrieval database on the Lance columnar format for multimodal AI data.
amazon-s3-vectors vs lancedb →Globally distributed serverless vector index for Cloudflare Workers and Workers AI.
amazon-s3-vectors vs cloudflare-vectorize →Google Cloud's managed ANN service built on ScaNN, now part of Gemini Enterprise Agent Platform.
amazon-s3-vectors vs vertex-ai-vector-search →Amazon S3 Vectors FAQ
What is Amazon S3 Vectors?
Amazon S3 Vectors is a vector database by Amazon Web Services, first released in 2025. Amazon S3 Vectors adds a new 'vector bucket' type to S3 with dedicated APIs to store embeddings and run similarity queries with metadata filters, without provisioning any cluster. It went GA in December 2025 with up to 2 billion vectors per index and 10,000 indexes per bucket, around 100 ms latency for frequent queries and sub-second for infrequent ones. AWS positions it as the cheap durable tier for RAG and agent memory, integrated with Bedrock Knowledge Bases and able to export into OpenSearch when a workload needs high QPS. It is not a low-latency serving engine and has no keyword search.
Is Amazon S3 Vectors open source?
No. Amazon S3 Vectors is proprietary, closed-source software.
Does Amazon S3 Vectors support ACID transactions?
No. Amazon S3 Vectors does not provide ACID transactions; its consistency model is eventual.
Does Amazon S3 Vectors scale horizontally?
Amazon S3 Vectors scales horizontally out of the box, with built-in sharding or partitioning. It is comfortable at massive scale (Multi-PB).
How can I run Amazon S3 Vectors?
You can use the vendor's managed cloud and run it serverless. It is available on AWS.
What is Amazon S3 Vectors best for?
Amazon S3 Vectors is a strong fit for cheap long-term store for RAG embeddings, agent memory at very large scale and archive tier behind a faster vector engine. Think twice for low-latency interactive search.
What are the best alternatives to Amazon S3 Vectors?
Common alternatives to Amazon S3 Vectors include turbopuffer, OpenSearch, Pinecone, LanceDB and Cloudflare Vectorize. The right choice depends on your workload, scale and hosting; WhichDB's smart match ranks them against your requirements.
How popular is Amazon S3 Vectors?
Amazon S3 Vectors ranks #491 of 844 databases on the WhichDB Score (2.1/100), and #27 among vector databases. The score combines public adoption signals such as package downloads, GitHub activity, job posts and developer discussion.