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

This article announces metadata pre-filtering for Amazon S3 Vectors, which improves recall on filtered queries by evaluating metadata filters before similarity search.

• Pre-filtering evaluates metadata filters before similarity search, returning more relevant matches from filtered subsets
• Supports up to 2 KB of filterable metadata per vector and 100 filter constraints per query
• Adds prefix matching with $startsWith operator for paths, URLs, and hierarchical keys
• No additional cost, no re-ingestion required, and existing queries remain unchanged
• Update existing indexes to ENHANCED mode using UpdateIndexMode API
• Improves recall up to 5x on highly selective filters compared to CLASSIC indexes
• Ideal for multi-tenant RAG, legal document search, financial services, and agentic applications

Pre-filtering enables scoped semantic searches for applications requiring both relevance and correct data scoping without sacrificing recall.