mteb-text-length-range

IN premise — summaries/2026/08/24/muennighoff-2022-mteb-sR-references.md

Created 2026-08-25T02:58:23+00:00

MTEB dataset average text lengths range from approximately 32 characters (TwentyNewsgroups) to approximately 1981 characters (MedrxivClusteringP2P).

Summary

The MTEB benchmark suite spans roughly a 60x range in input size, from very short text snippets of about 32 characters up to passages of nearly 2,000 characters. This means any model or scoring system evaluated across MTEB must handle inputs that are an order of magnitude different in length, so performance comparisons across tasks are not apples-to-apples in terms of how much context the model is working with.