← All models · MS MARCO reranker
use case · basics
The core move: give the cross-encoder a query and a set of candidate passages, and it scores each one by how well it answers the query — then you sort by that score. Unlike embeddings, it reads the query and passage together, so it judges genuine relevance, not just surface similarity.
Notice how "The cheetah can reach speeds of up to 120 km/h … the fastest land animal" wins over the peregrine-falcon line — even though the falcon passage contains the word "fastest" too. The cross-encoder keys on land animal, not just word overlap.