![[Update] Amazon S3 Vectors reduces query processing costs for large-scale vector indexes by up to 80%](https://images.ctfassets.net/ct0aopd36mqt/wp-thumbnail-066beb776f0c57ce64255fadcc072f60/82b2f6687ab5774cd73f9176dcac7855/amazon-s3?w=3840&fm=webp)
[Update] Amazon S3 Vectors reduces query processing costs for large-scale vector indexes by up to 80%
This page has been translated by machine translation. View original
This is Ishikawa from the Cloud Business Division. For Amazon S3 Vectors, query processing charges (data processed charges) for indexes with more than 10 million vectors have been reduced by up to 80%.
For those running large-scale AI, RAG, and semantic search workloads, this is a welcome update that can further reduce operating costs. No application-side changes are required, and it is applied automatically.
Update Details
This update is a reduction in the "data processed" charges for S3 Vectors queries.
The main changes are as follows:
- Query processing charges (data processed) for large indexes with more than 10 million vectors reduced by up to 80%
- No application-side changes required; the new pricing is applied automatically
- While costs decrease for large indexes, distributing vectors across multiple indexes continues to be recommended for improved query performance
Supported Regions
The pricing change is effective from the announcement date (June 16, 2026) in all AWS Regions where S3 Vectors is available. S3 Vectors is available in Asia Pacific (Tokyo), Asia Pacific (Osaka), and other regions.
Impact on Pricing
Since query processing charges are reduced, the cost of similarity searches targeting large indexes in particular will decrease. Be sure to check the "S3 pricing page" for specific unit prices.
How S3 Vectors Query Pricing Works
To understand the significance of this update, let's first organize the structure of S3 Vectors query pricing. The charge per query is composed of the following three main elements:
The key point is "data processed (data processing charge)". The data processed for S3 Vectors is calculated as average vector size (vector data + key + filterable metadata) × number of vectors in the index. In other words, the more vectors stored in an index, the more data is processed per query, resulting in a higher charge.
This data processed uses a tiered pricing structure (volume discount) based on vector count. With this update, the unit price for the largest tier—"over 10 million vectors"—has been reduced, delivering up to an 80% reduction. Workloads that consolidate large numbers of vectors into a single index will benefit the most.
Reference: Current Pricing
For reference, the S3 Vectors pricing in the Asia Pacific (Tokyo) region (ap-northeast-1) as of June 2026 is as follows. Since pricing is subject to change, always check the latest values on the "S3 pricing page" (select "Asia Pacific (Tokyo)" for the region under the Vectors tab).
| Item | Unit Price (Reference: Tokyo Region / ap-northeast-1) |
|---|---|
| Vector storage | $0.066 / GB per month |
| Upload (PUT) | $0.219 / GB (minimum 128KB charged per PUT) |
| Requests (GET / LIST, etc., excluding PUT and queries) | $0.06 / 1,000 requests |
| Query (request charge) | $0.0027 / 1,000 queries (≈ $2.70 / 1 million queries) |
| Query (data processed: first 100,000 vectors) | $0.0044 / TB |
| Query (data processed: 100,000 to 10 million vectors) | $0.0022 / TB |
| Query (data processed: over 10 million vectors) | $0.00044 / TB |
| Returned data | $0.01 / GB (first 512KB per query is free; each result is charged a minimum of 256 bytes) |
Why Distribution Is Still Recommended Even for Large Indexes
You might think, "If pricing has gone down, why not put everything into one huge index?" However, AWS continues to recommend distributing vectors across multiple indexes.
This is easier to understand when you consider the data processed calculation formula. The more vectors consolidated into a single index, the more data is processed per query (= average vector size × number of vectors), which affects query latency. By splitting indexes by access pattern, tenant, or use case, you can narrow down the scan target for each query and maintain high query performance.
Note that this pricing reduction is meant to "reduce the cost burden of large indexes" and does not mean "you no longer need to distribute." We recommend continuing to consider index design from both cost optimization and performance optimization perspectives separately.
Usage Notes
- This reduction applies only to the data processed charge for queries. Storage charges, upload (PUT) charges, and request charges are not included.
- The maximum reduction benefit applies to large indexes exceeding 10 million vectors. The effect is limited or does not apply to smaller scales.
- Unit prices and eligible regions are subject to change. Always check the latest official page when estimating costs.
Closing
Amazon S3 Vectors query processing charges (data processed) have been reduced by up to 80% for large indexes exceeding 10 million vectors. No application changes are required, and this is automatically applied in all regions where S3 Vectors is available.
For workloads handling large volumes of vectors—such as RAG, semantic search, and AI agent memory—there is potential to further reduce operating costs. Those already using S3 Vectors may want to review their current cost structure, while those considering adoption can incorporate this pricing change into their cost estimates.
