Last checked and updated on 05 Jun 2026
What is new in Elasticsearch 9.0
Elasticsearch 9.0 marks the start of the 9.x series with significant improvements in vector search, semantic capabilities, and overall performance. The release brings generally available features for semantic text and binary quantization, new field types for advanced ranking, enhanced ES|QL capabilities, and important security and runtime changes.
It delivers faster and more efficient search, especially for AI-powered and semantic workloads, while laying a strong foundation for future enhancements.
Key Highlights in Elasticsearch 9.0
| Feature | Description | Benefit |
|---|---|---|
| semantic_text Field Type (GA) | Generally available field type for seamless semantic search using built-in or custom inference models. | Simplifies building AI-powered search applications with high-quality semantic relevance. |
| rank_vectors Field Type | New experimental field type supporting late-interaction multi-vector ranking (e.g., ColBERT-style). | Enables more accurate ranking for complex retrieval scenarios. |
| Better Binary Quantization (BBQ) | Improved binary quantization for vectors, significantly faster than previous options and competitive alternatives. | Reduces memory usage and speeds up indexing and search for large vector datasets. |
| ES|QL Enhancements | New commands and functions including LOOKUP JOIN (tech preview), COMPLETION, and better cross-cluster support. | Makes data analysis and enrichment more powerful and flexible directly in Elasticsearch. |
| Entitlements Security Model | Permanent replacement of Java SecurityManager with the new Entitlements system. | Provides stronger and more maintainable security boundaries. |
| Lucene 10 Upgrade | Updated to Lucene 10 with performance gains in vector and lexical search. | Improves overall query speed and indexing efficiency. |
Semantic Search and Vector Improvements
The semantic_text field type is now generally available. It integrates inference models directly into mapping, making semantic search simple to set up and use. Support for bit vectors and sparse vector options has also been added.
The new rank_vectors field type (experimental) enables advanced late-interaction ranking techniques. Binary quantization receives major optimizations, offering better speed and efficiency for dense vector workloads.
Additional vector enhancements include:
- Option to include or exclude vectors from
_sourceretrieval. - Off-heap statistics for dense_vector fields in stats APIs.
- Panama-accelerated operations for scalar quantization and vector computations.
- Improved rescoring and oversampling controls.
These changes make Elasticsearch more capable for modern AI and semantic search use cases while keeping resource usage manageable.
ES|QL Query Language Updates
ES|QL gains several practical features in 9.0:
LOOKUP JOIN(tech preview) for enriching data with lookup indices, including support for multiple fields and expressions.COMPLETIONcommand (tech preview) andmatch_phrasefunction.- Better support for
date_nanos, geohash/geo tile functions, and new aggregation functions likeSTD_DEVandSAMPLE. - Improvements in cross-cluster querying and partial results handling.
These additions allow analysts to perform more sophisticated data processing and joins directly within Elasticsearch queries.
Storage, Mapping, and Performance
Several mapping and storage optimizations improve efficiency:
- Synthetic source improvements with native array offset storage for many field types.
- Synthetic recovery source enabled by default when synthetic source is active.
- Better handling of nested fields in time series indexes.
- Performance gains from Lucene 10, including faster vector indexing and search.
These changes reduce storage overhead and speed up indexing and querying for high-volume workloads.
Security and Runtime Changes
A major security update replaces the Java SecurityManager with the new Entitlements system. This provides a cleaner, more robust security model going forward.
Other security enhancements include:
- Support for SSHA-256 hashing for API keys.
- Additional permissions for built-in roles.
- Improved handling of inference and model-related security checks.
These updates strengthen cluster security without impacting existing workloads significantly.
Other Notable Enhancements
Elasticsearch 9.0 also includes many smaller but valuable improvements:
- New ingest processors and options for better pipeline control.
- Enhanced retrievers including pinned, linear, and RRF improvements.
- Better support for inference services with additional model providers.
- Various bug fixes and stability improvements, especially around system indices and migration.
- Performance tweaks across search, ingest, and snapshot operations.
While 9.0 focuses on maturing existing capabilities, it sets the stage for rapid innovation in the 9.x series.
How These Changes Help in Real-World Use
Teams building semantic search or RAG applications will appreciate the GA semantic_text field and improved vector performance. Analysts benefit from richer ES|QL features for data exploration and joining. Infrastructure operators gain from better storage efficiency and a more modern security model. The faster binary quantization helps reduce costs for large-scale vector deployments.
Most changes are backward compatible, allowing smooth upgrades from 8.x with minimal adjustments.
FAQ
What is the biggest change in Elasticsearch 9.0?
The general availability of the semantic_text field type and major improvements to binary quantization make semantic and vector search faster and easier to use at scale.
Is rank_vectors ready for production use?
The rank_vectors field type is experimental in 9.0 and intended for advanced late-interaction ranking scenarios.
Does Elasticsearch 9.0 remove the Java SecurityManager?
Yes. It is permanently replaced by the new Entitlements system for improved security and maintainability.
Can I use LOOKUP JOIN in ES|QL in production?
LOOKUP JOIN is available as a tech preview in 9.0. Test it thoroughly before using in critical production queries.
Will vector search be faster in 9.0?
Yes. Optimized binary quantization (BBQ) and Lucene 10 upgrades deliver noticeable performance gains, especially for large vector indexes.
Should I upgrade to Elasticsearch 9.0?
If you use semantic search, vectors, or ES|QL heavily, 9.0 brings valuable improvements. Review the full release notes and test in a staging environment, especially regarding the security model change.