mteb-useb-mostly-reranking

IN premise — summaries/2026/08/24/muennighoff-2022-mteb-s2-related-work.md

Created 2026-08-25T02:58:18+00:00

USEB (Wang et al., 2021) consists mostly of reranking tasks and does not cover retrieval or classification, limiting its scope compared to MTEB.

Summary

USEB is essentially a test of one narrow skill, reordering a small set of candidates, rather than a broad check of how well embeddings perform across different types of tasks. This means it can't tell you whether a model is actually good at retrieval or classification, so it gives a much thinner picture of general embedding quality than MTEB does.