On September 30, 2026, Amazon Web Services (AWS) announced metadata pre-filtering for Amazon S3 Vectors. In ENHANCED mode, the service applies a metadata filter before similarity search; CLASSIC evaluates the search and filter in tandem.
How ENHANCED differs from CLASSIC
A metadata filter narrows the vectors a query considers—for example, to a particular tenant or category. With ENHANCED, S3 Vectors resolves that filter first, then searches for similar vectors within the matching set. With CLASSIC, filtering and similarity search proceed together.
This order matters for selective filters, which leave a smaller set of eligible vectors. AWS says recall—the share of relevant items a search retrieves—can improve with the filter-first approach. The two modes handle the search differently:
| Dimension | CLASSIC | ENHANCED |
| Filter evaluation | Evaluates filtering and similarity search in tandem. | Resolves the metadata filter before similarity search. |
| Existing indexes | Remains the mode until the index is updated. | Can be enabled in place without re-ingestion or query changes, according to AWS. |
The prefix operator and filter limits
AWS added $startsWith, a prefix-matching operator for paths, URLs, and hierarchical keys. S3 Vectors also supports equality, numeric ranges, set membership, existence checks, and boolean combinations using $and and $or. Metadata fields are filterable by default.
The limits measure two different things: each vector can carry up to 2 KB of filterable metadata, while a query can use up to 100 filter constraints. The first is a per-vector metadata capacity; the second is a per-query limit.
Switching an existing index
AWS says an existing index can switch to ENHANCED in place through UpdateIndexMode. The change does not require vector re-ingestion or query changes, though the operation requires appropriate IAM permissions.
The default depends on when the vector bucket was created. Indexes in buckets created on or after September 30, 2026, use ENHANCED by default. Buckets created earlier retain CLASSIC as their default until that bucket default is changed, including for indexes created in those buckets later.
AWS’s recall claim and feature charges
AWS says highly selective filters can return up to five times as many matching vectors as the same query on a CLASSIC index. That is AWS’s performance claim for this comparison; the announcement does not establish it as a result for every workload.
AWS says metadata pre-filtering carries no additional feature charge in commercial Regions where S3 Vectors is available. Standard charges for storage, PUT requests, and queries still apply.
Availability in US Regions
AWS lists Amazon S3 Vectors in four commercial US Regions: us-east-1, us-east-2, us-west-1, and us-west-2. AWS says pre-filtering is available in commercial Regions where S3 Vectors is available.
