Back to the catalogue
RerankingRetrievalPopular
Cohere Rerank 3
Cross-encoder reranker that reorders retrieved chunks by true relevance before they reach the model's context.
Vector search is fast and approximate; a reranker is slow and exact. Ragenta runs them in order — fetch fifty candidates cheaply, then let the reranker score each one against the actual question and keep the six that earn their place.
On a large knowledge base this is usually the single biggest jump in answer quality, because the model stops being handed passages that merely share vocabulary with the question.