Amazon S3 Vectors introduces metadata pre-filtering for up to 5x higher recall on filtered search
Amazon S3 Vectors now supports metadata pre-filtering to improve filtered search recall by up to five times, alongside a new prefix match operator for path and URL filtering.
Amazon S3 Vectors has introduced metadata pre-filtering, which evaluates metadata filters before executing similarity searches. This change enables applications to return up to five times more matching vectors when filters are selective, improving the completeness of results for retrieval-augmented generation, agentic, and semantic-search use cases. The feature is designed to enhance the relevance of answers by ensuring more vectors are considered during filtered queries.
A new prefix match operator, $startsWith, has been added to support filtering on values such as paths and URLs. This operator complements metadata pre-filtering by providing more granular control over how filters are applied during queries. Together, these updates aim to streamline the process of retrieving relevant vectors while maintaining the cost-optimized and scalable nature of Amazon S3 Vectors.
Indexes in new vector buckets now use metadata pre-filtering by default, with no changes required to existing workflows for writing vectors with PutVectors or running filtered queries with QueryVectors. For existing indexes, users can enable pre-filtering in place using the UpdateIndexMode API, allowing for a seamless transition without data migration or downtime.
Metadata pre-filtering is available at no additional cost in all commercial AWS Regions where Amazon S3 Vectors is supported, as well as in the AWS China Regions. The feature is being rolled out gradually, with deployment expected to complete in the coming days. Users can access it via the AWS CLI, AWS SDKs, or the Amazon S3 console, with documentation and additional details available on the Amazon S3 Vectors documentation and AWS News blog.