Reranking Models

Reranking models score the relevance between a query and a set of candidate documents, enabling precision refinement in two-stage retrieval pipelines. After an initial recall phase using an embedding model, a reranker re-scores and reorders the candidates using a cross-encoder architecture with cross-attention, delivering more accurate relevance rankings.

• Qwen3 VL Reranker 8B: A multimodal reranking model that scores query-document relevance across text and images, supporting precision refinement in retrieval pipelines across 30+ languages.