OpenRouter documents reranking boosts RAG precision with voyageai/rerank-3-lite

Reranking in RAG applications fixes vector search retrieval errors by passing top candidate chunks through a dedicated cross-encoder model to re-score them against the user's question. Testing on a controlled dataset showed that adding a reranker raised precision@1 from 81.0% to 97.1% while allowing pipelines to send fewer chunks to the LLM, reducing overall answering costs. Open article →