AWS has added metadata pre-filtering to Amazon S3 Vectors, so searches can narrow to matching records before comparing their similarity. AWS says highly selective filters can return up to five times more matching vectors than the previous approach on indexes using CLASSIC mode, a practical change for scoped RAG and agent searches.
AWS AI Watch analysis
What happened
The new ENHANCED index mode applies metadata filters before the similarity search. Filters can target fields such as tenant, category or time, and the update adds $startsWith for matching paths, URLs and hierarchical keys. Each vector can carry up to 2 KB of filterable metadata, and a query supports up to 100 filter constraints.
AWS says existing indexes remain in CLASSIC mode until operators switch them using UpdateIndexMode. New indexes in vector buckets created from 30 September 2026 use ENHANCED mode by default; older buckets retain their existing default unless changed. The feature is available at no additional cost in commercial AWS Regions and AWS China Regions where S3 Vectors is available. Read AWS’s announcement.
Why it matters
A vector search that first finds the right tenant or document set can draw results from across that narrower pool, rather than filtering a smaller set of candidates pulled from the whole index. That matters when an agent needs to search one customer’s records or a RAG application must stay within a particular collection: relevance is useful, but relevance from the wrong account is not.
The upgrade is in place, with no re-ingestion or query changes required, according to AWS. Existing indexes do need an explicit mode update, so the improvement is not simply switched on everywhere by magic. A rare victory for checking the settings.
Our read
This is a meaningful infrastructure improvement for teams building filtered semantic search, especially where filters are highly selective. Operators should check which mode their indexes use and test the change against their own queries; AWS’s reported improvement is a product claim, not a guarantee that every workload will see the same gain.
What to watch
- How much recall improves on real workloads with selective filters.
- Whether teams update existing indexes or continue using CLASSIC mode.
- How applications use $startsWith to scope searches across hierarchical data.
Discussion spark: For a multi-tenant AI search system, would you switch existing indexes to pre-filtering promptly, or wait for workload-specific testing before changing the search path?
Sources and evidence
- Amazon S3 Vectors now supports metadata pre-filtering for higher recall on filtered searches (30 September 2026, 20:03 UTC)
not affiliated with or endorsed by Amazon Web Services (AWS)