Home / Glossary / Reranking
What is Reranking?
Reranking is a second-pass step in RAG that re-orders the top-k retrieved chunks by relevance to the query, using a smaller dedicated model. osFoundry’s RAG pipelines include reranking via Voyage rerankers by default.
Detail
Embedding-based retrieval is fast but coarse — it returns chunks that are semantically near the query, but the top-3 aren’t always the most relevant. A reranker scores each (query, chunk) pair more precisely (often using a cross-encoder), then re-orders.
Reranking adds 50-200 ms latency but dramatically improves top-3 precision, especially on technical content with overlapping vocabulary. Common cost: a few cents per million tokens reranked.
How osFoundry approaches Reranking
osFoundry’s default pipeline reranks with Voyage. Configurable threshold and top-k. BYOK to Cohere or any rerank API.
Related terms
Related features