Elasticsearch 9.0 Released: Vector Quantization GA and 2x Faster Aggregations

Elastic has released Elasticsearch 9.0, its first major version in three years, focused on vector search maturity and analytical performance.

Vector Quantization GA

Int8 and int4 scalar quantization is now GA for dense vector fields, reducing index size by 4x and 8x respectively while preserving 95%+ recall on typical benchmarks. Binary quantization (1-bit) is available as preview for extreme compression.

2x Faster Numeric Aggregations

A rewritten doc-values accessor using SIMD for numeric range aggregations doubles throughput on date histograms, range buckets, and percentile calculations. Dashboards on large time-series datasets feel noticeably snappier.

Breaking Changes

• Transport protocol v9, older clients incompatible
• Removed types from mapping (finally)
• Java 21+ required
• _doc replaced with _id in all APIs