mteb-clustering-reranking-strongest-correlation

IN premise — summaries/2026/08/24/muennighoff-2022-mteb-s4-additional-modalities-text-embeddings-are-chunk-1.md

Created 2026-08-25T02:58:20+00:00

Among MTEB tasks, clustering and reranking show the strongest inter-task performance correlation, while summarization is an outlier with weaker correlation to other tasks.

Summary

If a model performs well on clustering, it will almost certainly do well on reranking too, so the two tasks largely confirm each other. Summarization, however, breaks that pattern and should be evaluated on its own rather than inferred from other task scores, which means benchmark comparisons and model selection need to treat it as a separate axis of capability.