Retrieval model · Established · Advanced
Reranker models
Cross-encoder models that read a query and a candidate passage together to score relevance precisely.
What Reranker models is
Rerankers are slower than embedding search but far more accurate, so they operate on a shortlist rather than the whole corpus.
How it works
Retrieve a few dozen candidates by vector or keyword search, score each with the reranker, and pass the top few to the generator.
Why it matters
Adding a reranker is often the single largest quality improvement available to an existing RAG pipeline.
Common uses
- →RAG precision
- →Enterprise search
- →Question answering
Strengths
- ✓Large relevance gains
- ✓Model-agnostic addition
Watch for
- ✓Latency and cost per query
- ✓Scales poorly to large candidate sets
Continue exploring
More in this collection
Browse all AI Models