Hybrid search made simple — One API. Exceptional relevance.

Elasticsearch gives you all the tools to add hybrid search through a single API, so you can quickly improve results and tune for exceptional relevance without stitching together multiple systems.

Start free trial

Hybrid search is an information retrieval technique that improves relevance by blending two or more search methods into a single ranked list.

  • Lexical search is based on specific keywords. It’s ideal for specific facts, rare terms, and structured content.

    Try hands-on tutorial

  • Vector search is based on semantic meaning. It’s ideal for unstructured content and meaning-based retrieval.

    Try hands-on tutorial

  • Hybrid search combines multiple retrieval methods, such as lexical, semantic, geo, and more, to achieve superior relevance that’s ideal for real-world use cases.

    Hybrid search examples

Hybrid search with Elasticsearch

Search every data type in one datastore, and power retrieval augmented generation (RAG) and AI agents with results that balance accuracy (with scoring algorithms like BM25F) and semantic understanding.

  • Build hybrid search easily with the speed of a single API by blending lexical and semantic retrieval. Balance exact matches with contextual meaning and achieve high-quality relevance without added complexity.

    Set up hybrid search

  • Customize relevance with full control

    Combine lexical search with production-ready models for semantic search such as ELSER, jina-embeddings-v3, or your own to build hybrid retrieval pipelines. Then, refine relevance by experimenting with advanced techniques such as filters, boosts, ranking, and reranking.

    Learn more

  • Build hybrid search that spans text, image, and geospatial data

    With Elasticsearch, hybrid search adapts to whatever combination you need. Combine keyword, semantic, geospatial, and multimodal approaches and deliver results that are as precise as they are relevant.

    Build hybrid search with multimodal and geo signals

  • Improve AI agent reliability with relevant context

    Use hybrid search to engineer high-quality context across your data, giving AI agents the information they need to reason, plan with, act on, and build more accurate and reliable workflows.

    Get started with Agent Builder

Why developers choose Elasticsearch

Get the best tools for precision, explainability, and control. Lexical search excels at structured queries, rare terms, and out-of-domain data. Semantic search adds fuzziness and recall when exact matches fall short. Control how they work together with tune scoring, filters, and boosts.

For exact, structured, and explainable queries
For flexible, semantic, high-recall search
For production-grade relevance from both worlds
Scoring that makes sense

Use BM25F scoring with full control over field weights and term boosts — no model required.

Retrieve semantically related results via dense_vector or semantic_text fields.

Combine results via reciprocal_rank_fusion or <options> in the rank API.

Full control in your query DSL

Tune relevance using combined_fields, boost, fuzziness, synonyms, and analyzers.

Bring your own embeddings or use built-in inference with ELSER, OpenAI, etc.

Use a single hybrid query with shared filters, weights, and rerank logic.

Filters that just work

Get native support for geo, term, range, and ACL filters — fast and stable at scale.

ACORN-1 enables fast filtered kNN even on large datasets with filter clause support.

The shared filtering layer works across both retrievers — no pipeline stitching required.
Debug and inspect capabilities

Use explain, profile, and the _rank_features field to understand how docs score.

Vector scores are fully exposed — inspect similarity math or weight contributions.
Gain end-to-end debug visibility across both search paths — down to each reranker’s impact.
Good for when ...
You need precision, filtering, and control — for logs, catalog, identifiers, and compliance.
You're handling vague queries, new terms, semantic drift, or unknown phrasing.
You want robust, tunable, explainable results — even when queries get weird.
Scoring that makes sense
Full control in your query DSL
Filters that just work
Debug and inspect capabilities
Good for when ...
For exact, structured, and explainable queries
For flexible, semantic, high-recall search
For production-grade relevance from both worlds

Use BM25F scoring with full control over field weights and term boosts — no model required.

Retrieve semantically related results via dense_vector or semantic_text fields.

Combine results via reciprocal_rank_fusion or <options> in the rank API.

Tune relevance using combined_fields, boost, fuzziness, synonyms, and analyzers.

Bring your own embeddings or use built-in inference with ELSER, OpenAI, etc.

Use a single hybrid query with shared filters, weights, and rerank logic.

Get native support for geo, term, range, and ACL filters — fast and stable at scale.

ACORN-1 enables fast filtered kNN even on large datasets with filter clause support.

The shared filtering layer works across both retrievers — no pipeline stitching required.

Use explain, profile, and the _rank_features field to understand how docs score.

Vector scores are fully exposed — inspect similarity math or weight contributions.
Gain end-to-end debug visibility across both search paths — down to each reranker’s impact.
You need precision, filtering, and control — for logs, catalog, identifiers, and compliance.
You're handling vague queries, new terms, semantic drift, or unknown phrasing.
You want robust, tunable, explainable results — even when queries get weird.

Rightsize your relevance journey

Elasticsearch gives you relevance control at every level — from zero-config to full customization. Explore the full tuning journey on Elasticsearch Labs.

  • Use BM25F: the original no-LLM-needed technology.

  • Use high-performance models available out of the box, such as ELSER and jina-embeddings-v3, with lexical search for better recall on complex queries.

  • Expert mode

    Use rerankers, retrievers, and Better Binary Quantization (BBQ) to ship domain-specific retrieval pipelines.

Best in class? Built right in

Start with Elastic's first-party ELSER and Jina AI models, built into Elasticsearch. Or plug into the models you already use through native integrations across the AI ecosystem.

A four-column ecosystem diagram displaying the logos of leading AI and machine learning partners across Model Providers, Platform Providers, MLOps and orchestration tools, and Open Standard API clients. The visual shows Elastic connecting natively to the full AI stack to enhance search and power intelligent applications.

Frequently asked questions

Hybrid search combines keyword (lexical) precision with vector (semantic) similarity, so users get relevant results even when queries don’t match exact text.