Brittlebench: Quantifying LLM robustness via prompt sensitivity

TL;DR

This paper introduces Brittlebench, a framework and evaluation pipeline to measure language model sensitivity to prompt variations, revealing significant performance degradation and ranking shifts caused by semantics-preserving perturbations.

Contribution

The work presents a novel theoretical framework and evaluation pipeline for quantifying and analyzing model brittleness due to prompt sensitivity in language models.

Findings

Model performance can degrade by up to 12% due to prompt perturbations.

In 63% of cases, perturbations change the relative ranking of models.

Semantics-preserving input variations can account for up to half of performance variance.

Abstract

Existing evaluation methods largely rely on clean, static benchmarks, which can overestimate true model performance by failing to capture the noise and variability inherent in real-world user inputs. This is especially true for language models, which can face human-generated text queries containing mistakes, typos, or alternative ways of phrasing the same question. In this work, we introduce a theoretical framework for quantifying model sensitivity to prompt variants, or brittleness, that can enable us to disentangle data-induced difficulty from prompt-related variability. Using this framework, we design a novel evaluation pipeline, Brittlebench, to holistically evaluate the sensitivity of frontier models. We apply semantics-preserving perturbations to a suite of popular benchmarks, and observe model performance to degrade as much as 12%. However, these perturbations do not affect all models equally: even a single perturbation alters the relative ranking of models in 63% of cases, impacting conclusions about comparative model performance. Decomposing the total variance of both state-of-the-art open-weight and commercial models, we find that semantics-preserving input perturbations can account for up to half of the performance variance for a given model. Brittlebench highlights the need for more robust evaluations and models, and allows us to systematically understand model brittleness.