---
format: "aidr-story-markdown/v1"
id: "d1a13212a136ab34306f9862c95f2eb507167955594eccd632d9791faaaa99eb"
canonical_url: "https://aidr.today/d1a13212?lang=en"
title: "Random Is Almost Enough? Stable and Low-Cost Benchmark Compression via BestRand"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-11-06T00:00:00.000Z"
category: "Research"
topics: ["llm"]
source_urls: ["https://huggingface.co/papers/2611.00883"]
summary: "Benchmark compression evaluates models on a small benchmark subset to reduce time and cost of model evaluation. BestRand is introduced as a stable and low-cost benchmark compression method that reduces the high-error tail of random sampling. BestRand was validated on 70+ text, multimodal, and agent benchmarks and extended to VLA evaluation."
---

# Random Is Almost Enough? Stable and Low\-Cost Benchmark Compression via BestRand

> [Open the canonical story](<https://aidr.today/d1a13212?lang=en>)

**Published:** 2026-11-06T00:00:00.000Z
**Category:** Research
**Topics:** llm

## Summary

Benchmark compression evaluates models on a small benchmark subset to reduce time and cost of model evaluation\. BestRand is introduced as a stable and low\-cost benchmark compression method that reduces the high\-error tail of random sampling\. BestRand was validated on 70\+ text, multimodal, and agent benchmarks and extended to VLA evaluation\.

## Sources

- [Story source](<https://huggingface.co/papers/2611.00883>)