Rank1

The first test-time-compute ("reasoning") reranker (COLM 2025): generates an explicit reasoning chain before judging document relevance, trained on a released dataset of 600K+ R1 reasoning traces from MS MARCO queries. rank1-7b sits around 75K HF downloads; HLTCOE's Rank-K (arXiv:2505.14432) extends the idea to listwise reranking. JHU (Weller, Yang, Lawrie, Van Durme).