Thakur, Reimers et al. present BEIR
Presented BEIR — a zero-shot heterogeneous benchmark of 18 datasets showing that BM25 is a robust baseline and that dense and sparse models often fall behind out of distribution
Why it matters
A decisive rebuttal of the narrative that dense retrieval had beaten lexical retrieval. The practical grounds for hybrid search becoming the default
Source — Primary sources first; secondary reports are labelled “Press”
Paperarxiv.orghttps://arxiv.org/abs/2104.08663Added in Week 35, 2026 · See that week