About SRGTechnical articleRecruitment Information
About SRGTechnical articleRecruitment Information
HOME/Technical article/Amazon S3 Vectors has been released in preview! I tried out vector search on S3.

Amazon S3 Vectors has been released in preview! I tried out vector search on S3.

This is Yuta Kikai (@fat47) from the Service Reliability Group (SRG) of the Media Management Division.
#SRGThe Service Reliability Group primarily provides comprehensive support for the infrastructure surrounding our media services, focusing on improving existing services, launching new ones, and contributing to open-source software (OSS).
This article is a review of the preview release of the S3 Vector Store feature.
 

What is Amazon S3 Vectors?

This is a preview release of new S3 features announced on July 16, 2025 (Japan time).
 
This feature allows you to use S3 as a vector store, and it's a promising feature for low-cost vector storage.
The current preview release does not support the Japan region; it is only available in select regions such as the US and Sydney regions.
For more details, please refer to the release blog and official documentation mentioned above.

I tried out Amazon S3 Vectors.

Let's try using S3 Vectors right away.
Let's try embedding a Markdown file in the vector store and searching for it.
💡
Currently, this is a preview release. You can create vector buckets and vector indexes from the console, but you cannot delete vector indexes or add/delete vector data.
Operations will need to be performed via the AWS CLI or SDK.
 

Creating a vector bucket

The "Vector Bucket" option has appeared in the S3 menu. Select "Create Vector Bucket" from here.
 
Create a unique vector bucket by specifying its name.
 

Creating a vector index

Once you've opened the vector bucket you created, select "Create Vector Index".
 
We will now enter the necessary information for the vector index.
 
  • Vector index name: A unique index name within the bucket
  • dimension: 1024
    • 1024,512,256
  • Distance Metric
    • You can choose between cosine and Euclidean functions. This time, we'll select cosine.
 
Once you have entered the above information, select "Create Vector Index".
 

Insertion of vector data

This cannot currently be done via the console.
This time, it's AWS officials3vectors-embed-cli Use this.
amazon.titan-embed-text-v2:0
 
Please refer to the above document for detailed instructions on how to implement the system.
Basically, this is all you need to use it.
 
We will PUT the data into the vector bucket we created.
💡
Large Markdown files need to be split into multiple files beforehand.
If you are using amazon.titan-embed-text-v2:0 for your embedded model, the limit is 8192 tokens per file.
File splitting sample
## 見出し
The above PUT command sends all files within the chunk file directory created here.
 

Queries into vector data

Let's try running a query on the registered vector data.
--k
--return-distance
 
Example results
The results were returned in order of closest distance.

This was a feature I was really looking forward to seeing in the GA release!

That completes the vector search!
Regarding pricing, since prices for the Japan region are not yet available, I will show you the S3 Vectors price list for the Northern Virginia region.
 
As shown in this table, charges are based only on data storage size, queries, and requests; unlike other vector stores, there are no charges based on startup time.
This might be the perfect solution for those who want to try out RAG without any hassle.
 
Since it's still a preview release, there are many areas where the console functionality is lacking, but it's a feature I'm really looking forward to seeing in the GA release!
 
If you are interested in SRG, please contact us here.