HOME·SCIENCE·HYBRID VS VECTOR VS SEMANTIC SEARCH
SCIENCE · RETRIEVAL

Hybrid vs vector vs semantic search

AUTHOR · XERJ documentation team · REVIEWED BY · XERJ engineering team · PUBLISHED · 2026-11-13 · UPDATED · 2026-11-14
SHORT ANSWER · UPDATED 2026-11-14

What is the difference between hybrid, vector, and semantic search? Vector search ranks passages by closeness in a vector space. Semantic search is the usual name for ranking by meaning, and it almost always uses vectors. Hybrid search also runs a full-text arm, such as BM25, and fuses the two lists.

TL;DR Vector search ranks by closeness in a vector space. Semantic search is the usual name for ranking by meaning, and it almost always uses vectors. Hybrid search also runs a full-text arm, such as BM25, and fuses the two lists.

Try it: three ways to rank

Hybrid, vector, and semantic search

Two lists, then a merge

A hybrid query runs a full-text arm, such as BM25, and a vector arm on the same question, then fuses the ranks. You keep exact names and close meanings in one list.

This comparison is a teaching account, not a new benchmark. Measured hybrid scores stay on the hybrid-search how-to and the published quality answers.
PASTE THIS TO YOUR AI AGENT — IT DOES THE REST Act as a coding agent. Read https://xerj.org/llms.txt, install XERJ, start a local node, index one folder I point you to, and run one search so I can see a cited passage.

How the three differ

A full-text ranker such as BM25 scores a document by the query terms it contains, how rare those terms are, and document length. It is strong on exact tokens: names, error strings, identifiers.

Vector search turns the query and each passage into vectors and ranks by closeness. It can match a paraphrase. It is weaker where two similar identifiers sit close together.

Semantic search is the usual name for ranking by meaning. In practice that almost always means a vector query. The names are close, and product pages sometimes treat them as the same thing. This page keeps both phrases because people search for both.

Hybrid search runs both a full-text arm and a vector arm on the same question, then merges the ranks. Reciprocal rank fusion is a common way to merge them without mixing raw scores. The formula is on the reciprocal rank fusion page.

How XERJ uses it

XERJ answers a full-text query with BM25 by default. Neural or hybrid mode adds a vector arm. The default embedder is lexical and runs offline. The how-to, including the request shape, is how to run hybrid search in XERJ. A side-by-side of the two arms is BM25 vs vector search.

This page does not add a new quality number.

History and research

Dense vector retrieval is described by Karpukhin and colleagues in Dense Passage Retrieval for Open-Domain Question Answering (EMNLP 2020). BM25 is surveyed by Robertson and Zaragoza in The Probabilistic Relevance Framework: BM25 and Beyond (2009). Fusion is described by Cormack, Clarke and Buettcher in Reciprocal Rank Fusion outperforms Condorcet and individual Rank Learning Methods (SIGIR 2009).

Those papers are the research sources. They are not XERJ measurements.

Guides

For a hands-on view, read vector database vs full-text search, BM25 vs vector search and the vectors reference.

Related terms

Hybrid search, BM25, semantic search, and reciprocal rank fusion.

FAQ

What is vector search?

It ranks passages by closeness in a vector space. A paraphrase can match even when it shares few words with the query.

Is semantic search the same as vector search?

In practice, almost always. Semantic search means ranking by meaning, and the usual method is a vector query. Some writers use the names as if they were identical.

What does hybrid search add?

It also runs a full-text arm, such as BM25, and fuses the two ranked lists. Exact names and close meanings can both appear in the results.

Where should I start in XERJ?

The concept pages are hybrid search, BM25, and semantic search. The how-to is the hybrid search answer. This page is only the comparison.

Related