Amazon S3 Vectors adds metadata pre-filtering for higher recall on filtered searches

Amazon S3 Vectors now supports metadata pre-filtering, a feature that evaluates metadata filters before executing similarity searches to improve recall on filtered queries. This update allows applications to scope searches by attributes such as tenant, category, status, or time, ensuring that the similarity search covers all relevant vectors matching the filter rather than a subset of the entire index.

How Pre-Filtering Works

The feature introduces a new index mode called ENHANCED. In this mode, S3 Vectors resolves the filter first and then searches only the vectors that match. This contrasts with the existing CLASSIC mode, where vector search and filter evaluation happen in tandem, potentially missing relevant matches in highly selective queries.

• Higher Recall: On highly selective filters, pre-filtering returns up to 5x more matching vectors compared to CLASSIC indexes.
• No Re-ingestion: Existing indexes can be updated to ENHANCED mode without re-ingesting data or changing queries.
• New Operator: A $startsWith operator is added for prefix matching on paths, URLs, and hierarchical keys.

Use Cases and Implementation

Pre-filtering is designed for applications where results must be both relevant and correctly scoped, such as:

• Legal and Professional Services: Scoping document searches to a single client or matter number.
• Financial Services: Filtering analyst notes by issuer, document type, and date.
• Agentic Applications: Restricting agent searches to a specific user’s session or document set.

To enable this feature, users must update their index mode using the UpdateIndexMode API. New indexes created in vector buckets established on or after September 30, 2026, will default to ENHANCED mode. Each vector can carry up to 2 KB of filterable metadata, and a single query supports up to 100 filter constraints.

Availability and Cost

Metadata pre-filtering is available at no additional cost in all commercial AWS Regions and AWS China Regions where Amazon S3 Vectors is offered. Users continue to pay standard S3 Vectors pricing for storage, PUT requests, and queries.