Amazon S3 Vectors introduces metadata pre-filtering for up to 5x higher recall on filtered search
Amazon S3 Vectors now supports metadata pre-filtering, boosting recall up to 5x for selective filters, and adds a prefix match operator ($startsWith) for filtering values like paths and URLs.
Amazon S3 Vectors supports metadata pre-filtering, evaluating metadata filters before similarity searches, which can increase matching vectors by up to five times for selective filters. This improves filtering completeness for RAG, agent-based, and semantic search applications. S3 Vectors natively stores and queries vectors on Amazon S3, offering purpose-built, cost-optimized vector storage and queries at petabyte scale. New vector bucket indexes use metadata pre-filtering by default, with no changes to writing vectors using PutVectors or running filtered queries using QueryVectors. To enable pre-filtering on existing indexes, update the index using the UpdateIndexMode API. QueryVectors provides per-query parameters to compare current and pre-filtering on the same index. Metadata pre-filtering is available at no extra cost in all commercial AWS regions and AWS China regions. The feature is currently being rolled out and will complete within days. You can get started using the AWS CLI, AWS SDK, or Amazon S3 console. For more information, see the Amazon S3 Vectors documentation and the AWS News Blog.
Why it matters
Amazon S3 Vectors provides native support for storing and querying vectors in Amazon S3, useful for RAG, agentic, and semantic-search applications. The addition of metadata pre-filtering improves the precision and efficiency of filtering operations.