---
title: "What is hybrid search? BM25 plus vector search"
description: "Hybrid search combines keyword search such as BM25 with vector search and merges the results. Learn how it works and how it compares to pure vector search."
type: "glossary-term"
canonical_url: "https://theaiengineering.com/glossary/hybrid-search"
markdown_url: "https://theaiengineering.com/glossary/hybrid-search.md"
topic: "RAG and vector search"
author: "The AI Engineering Editors"
publisher: "The AI Engineering"
updated: "2026-11-06"
keywords: ["Hybrid search","hybrid retrieval","BM25 plus vector search","keyword and semantic search"]
citation: "The AI Engineering Editors. \"What is hybrid search? BM25 plus vector search.\" The AI Engineering, updated Nov 06, 2026. https://theaiengineering.com/glossary/hybrid-search"
---

# What is hybrid search? BM25 plus vector search

Also called hybrid retrieval, BM25 plus vector search, keyword and semantic search.

> **Definition of Hybrid search:** Hybrid search is a retrieval method that runs keyword search, typically BM25, and vector similarity search on the same query and merges the two ranked lists into one result set.

Hybrid search is a retrieval method that combines lexical (keyword) search with semantic (vector) search for the same query. The keyword side, usually [BM25](https://theaiengineering.com/glossary/bm25), matches exact terms. The vector side compares [embeddings](https://theaiengineering.com/glossary/embeddings) to match meaning. A fusion step merges both ranked lists into one.

## How hybrid search works

Hybrid search works by running two retrievers in parallel and fusing their results:

1. **Keyword retrieval**: BM25 scores each document by how often the query terms appear in it, weighted by how rare each term is across the corpus and normalized for document length.
2. **Vector retrieval**: the query is embedded and matched against document vectors in a [vector database](https://theaiengineering.com/glossary/vector-database).
3. **Fusion**: the two lists are merged into one ranking.

Reciprocal rank fusion (RRF) is a common fusion method because it uses only rank positions, so the incompatible score scales of BM25 and [cosine similarity](https://theaiengineering.com/glossary/cosine-similarity) do not matter. Each document receives 1 / (k + rank) from every list it appears in, and the sums are sorted. [Elasticsearch's RRF implementation](https://www.elastic.co/docs/reference/elasticsearch/rest-apis/reciprocal-rank-fusion) sets k to 60 by default. The alternative is weighted score fusion, which normalizes both scores and blends them with a tunable weight. The fused list often goes to a [reranking](https://theaiengineering.com/glossary/reranking) model before it reaches the LLM.

## Hybrid search vs vector search

Hybrid search tends to beat pure vector search when queries contain exact identifiers, such as error codes, product SKUs, function names, or legal citations, which embeddings blur together with similar-looking strings. Pure vector search handles paraphrases and synonyms well but can miss a document that shares the one rare term the user typed. Anthropic's [contextual retrieval](https://www.anthropic.com/news/contextual-retrieval) write-up makes the same point: embeddings can miss exact matches such as unique identifiers, and combining BM25 with embeddings beat embeddings alone in Anthropic's tests.

| | Vector search | Hybrid search |
|---|---|---|
| Matches | Meaning | Meaning and exact terms |
| Handles synonyms | Yes | Yes |
| Handles rare identifiers | Often poorly | Yes |
| Moving parts | One index | Two indexes plus fusion |

The cost of hybrid search is a second index to build and keep in sync, plus a fusion step to tune. Many vector databases and search engines now support both index types in one system, which reduces that overhead; the [vector databases directory](https://theaiengineering.com/platforms/vector-databases) lists options. For the full pipeline, see [What is RAG](https://theaiengineering.com/articles/what-is-rag) and the [retrieval-augmented generation](https://theaiengineering.com/glossary/retrieval-augmented-generation) entry.

## Related terms

- [Embeddings](https://theaiengineering.com/glossary/embeddings): Embeddings are fixed-length vectors of numbers that represent data such as text or images so that items with similar meaning sit close together, which lets software compare meaning with arithmetic.
- [Vector database](https://theaiengineering.com/glossary/vector-database): A vector database is a database built to store embedding vectors and quickly return the vectors most similar to a query vector, usually with approximate nearest neighbor indexes such as HNSW.
- [Reranking](https://theaiengineering.com/glossary/reranking): Reranking is a second retrieval stage that rescores an initial list of candidate documents with a more accurate but slower model, usually a cross-encoder, and keeps only the most relevant ones.
- [Retrieval-augmented generation (RAG)](https://theaiengineering.com/glossary/retrieval-augmented-generation): Retrieval-augmented generation (RAG) is a technique that retrieves relevant passages from an external knowledge source at query time and adds them to a language model's prompt so the answer is grounded in that data.
- [Chunking](https://theaiengineering.com/glossary/chunking): Chunking is the step in a retrieval pipeline that splits documents into smaller passages, called chunks, so each one can be embedded, indexed, and retrieved on its own.
- [BM25](https://theaiengineering.com/glossary/bm25): BM25 is a keyword ranking function that scores a document against a query using how often each query term appears, how rare the term is across the corpus, and how long the document is.

## Guides that cover hybrid search

- [What is retrieval-augmented generation (RAG)? How RAG works](https://theaiengineering.com/articles/what-is-rag): Retrieval-augmented generation (RAG) grounds LLM answers in your documents. Learn the RAG pipeline, hybrid search, agentic RAG, evaluation, and failure modes.
- [Chunking strategies for RAG: how to choose, tune, and evaluate them](https://theaiengineering.com/articles/chunking-strategies-for-rag): Chunking strategies for RAG compared: fixed-size, recursive, document-aware, semantic, parent-child, and late chunking, with tested Python and an eval method.
- [AI engineer roadmap: skills, projects, and resources by stage](https://theaiengineering.com/articles/ai-engineer-roadmap): A practical AI engineer roadmap in seven stages, each with the skills to learn, a project to ship, and resources worth the time, from Python to production.

Part of the [RAG and vector search](https://theaiengineering.com/topics/rag) topic hub.

## Cite this definition

The AI Engineering Editors. "What is hybrid search? BM25 plus vector search." The AI Engineering, updated Nov 06, 2026. https://theaiengineering.com/glossary/hybrid-search

- Title: What is hybrid search? BM25 plus vector search
- Author: The AI Engineering Editors
- Publisher: The AI Engineering
- Updated: 2026-11-06
- URL: https://theaiengineering.com/glossary/hybrid-search
- Markdown: https://theaiengineering.com/glossary/hybrid-search.md
