← All models · BGE-reranker-v2-m3
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. Because it reads the query and passage together on the M3/large backbone, it judges genuine relevance, not surface similarity.
Notice how the cheetah line wins over the peregrine-falcon line — even though the falcon passage also contains the word "fastest." The cross-encoder keys on land animal, not just word overlap.