Cross-Encoder
Also called Cross Encoder
A model that reads query and document together to judge relevance — accurate, but too slow to search with.
Think of it like
A judge who reads both essays side by side before scoring, rather than grading each in isolation.
Example
After fast search returns 100 candidates, a cross-encoder reranks them by reading each one next to the query, surfacing the true best few.
How it actually works
A cross-encoder takes the query and a candidate document as a single joint input, letting them attend to each other, and outputs one relevance score. That interaction makes it far more accurate than independent embeddings, but you must run it fresh for every pair — impossible over a whole corpus. So it lives in the reranking stage, scoring a small candidate set from a faster first-pass retriever.
For product teams
The precision boost at the end of the pipeline; used on a shortlist, never the whole database.
For engineers
Joint query-document encoding → single relevance score; O(candidates) cost confines it to reranking.
Related
- Reranking — The engine behind reranking.
- Bi-Encoder — Contrasts with the faster bi-encoder.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome