Reranker
Also: re-ranker, cross-encoder
A model that re-scores an initial set of retrieved passages by how well each actually answers the query, sharpening RAG results before they reach the main model.
Embedding-based retrieval is fast but approximate: it finds passages that are broadly similar to a query, not necessarily the ones that best answer it. A reranker is a second, more careful model that takes the query and each candidate passage together and scores the true relevance, so the best few can be kept.
Adding a reranker after retrieval is one of the highest-value improvements to a RAG pipeline, because it fixes cases where the right passage was retrieved but buried. The cost is a little extra latency per query. See improving RAG retrieval quality.