Benchmark^2: Systematic Evaluation of LLM Benchmarks

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Qian, Qi, Huang, Chengsong, Xu, Jingwen, Lv, Changze, Wu, Muling, Liu, Wenhao, Wang, Xiaohua, Wang, Zhenghua, Huang, Zisu, Tian, Muzhao, Xu, Jianhan, Hu, Kun, Wang, He-Da, Hu, Yao, Huang, Xuanjing, Zheng, Xiaoqing
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909983704088576
author Qian, Qi
Huang, Chengsong
Xu, Jingwen
Lv, Changze
Wu, Muling
Liu, Wenhao
Wang, Xiaohua
Wang, Zhenghua
Huang, Zisu
Tian, Muzhao
Xu, Jianhan
Hu, Kun
Wang, He-Da
Hu, Yao
Huang, Xuanjing
Zheng, Xiaoqing
author_facet Qian, Qi
Huang, Chengsong
Xu, Jingwen
Lv, Changze
Wu, Muling
Liu, Wenhao
Wang, Xiaohua
Wang, Zhenghua
Huang, Zisu
Tian, Muzhao
Xu, Jianhan
Hu, Kun
Wang, He-Da
Hu, Yao
Huang, Xuanjing
Zheng, Xiaoqing
contents The rapid proliferation of benchmarks for evaluating large language models (LLMs) has created an urgent need for systematic methods to assess benchmark quality itself. We propose Benchmark^2, a comprehensive framework comprising three complementary metrics: (1) Cross-Benchmark Ranking Consistency, measuring whether a benchmark produces model rankings aligned with peer benchmarks; (2) Discriminability Score, quantifying a benchmark's ability to differentiate between models; and (3) Capability Alignment Deviation, identifying problematic instances where stronger models fail but weaker models succeed within the same model family. We conduct extensive experiments across 15 benchmarks spanning mathematics, reasoning, and knowledge domains, evaluating 11 LLMs across four model families. Our analysis reveals significant quality variations among existing benchmarks and demonstrates that selective benchmark construction based on our metrics can achieve comparable evaluation performance with substantially reduced test sets.
format Preprint
id arxiv_https___arxiv_org_abs_2601_03986
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Benchmark^2: Systematic Evaluation of LLM Benchmarks
Qian, Qi
Huang, Chengsong
Xu, Jingwen
Lv, Changze
Wu, Muling
Liu, Wenhao
Wang, Xiaohua
Wang, Zhenghua
Huang, Zisu
Tian, Muzhao
Xu, Jianhan
Hu, Kun
Wang, He-Da
Hu, Yao
Huang, Xuanjing
Zheng, Xiaoqing
Computation and Language
The rapid proliferation of benchmarks for evaluating large language models (LLMs) has created an urgent need for systematic methods to assess benchmark quality itself. We propose Benchmark^2, a comprehensive framework comprising three complementary metrics: (1) Cross-Benchmark Ranking Consistency, measuring whether a benchmark produces model rankings aligned with peer benchmarks; (2) Discriminability Score, quantifying a benchmark's ability to differentiate between models; and (3) Capability Alignment Deviation, identifying problematic instances where stronger models fail but weaker models succeed within the same model family. We conduct extensive experiments across 15 benchmarks spanning mathematics, reasoning, and knowledge domains, evaluating 11 LLMs across four model families. Our analysis reveals significant quality variations among existing benchmarks and demonstrates that selective benchmark construction based on our metrics can achieve comparable evaluation performance with substantially reduced test sets.
title Benchmark^2: Systematic Evaluation of LLM Benchmarks
topic Computation and Language
url https://arxiv.org/abs/2601.03986