Benchmark^2: Systematic Evaluation of LLM Benchmarks

TL;DR

Benchmark^2 introduces a systematic framework with three metrics to evaluate the quality of LLM benchmarks, revealing significant variations and enabling more efficient and reliable model evaluation.

Contribution

It provides a comprehensive, metric-based approach to assess and improve the quality of LLM benchmarks, addressing inconsistencies and inefficiencies in current evaluation practices.

Findings

Significant quality variations among existing benchmarks

Selective benchmark construction can reduce test set size

Metrics correlate with evaluation reliability

Abstract

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.