Saved in:
Bibliographic Details
Main Authors: Romanou, Angelika, Ibrahim, Mark, Ross, Candace, Shaib, Chantal, Oktar, Kerem, Bell, Samuel J., Ovalle, Anaelia, Dodge, Jesse, Bosselut, Antoine, Sinha, Koustuv, Williams, Adina
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2603.13285
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915917183582208
author Romanou, Angelika
Ibrahim, Mark
Ross, Candace
Shaib, Chantal
Oktar, Kerem
Bell, Samuel J.
Ovalle, Anaelia
Dodge, Jesse
Bosselut, Antoine
Sinha, Koustuv
Williams, Adina
author_facet Romanou, Angelika
Ibrahim, Mark
Ross, Candace
Shaib, Chantal
Oktar, Kerem
Bell, Samuel J.
Ovalle, Anaelia
Dodge, Jesse
Bosselut, Antoine
Sinha, Koustuv
Williams, Adina
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2603_13285
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Brittlebench: Quantifying LLM robustness via prompt sensitivity
Romanou, Angelika
Ibrahim, Mark
Ross, Candace
Shaib, Chantal
Oktar, Kerem
Bell, Samuel J.
Ovalle, Anaelia
Dodge, Jesse
Bosselut, Antoine
Sinha, Koustuv
Williams, Adina
Machine Learning
Artificial Intelligence
I.2
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.
title Brittlebench: Quantifying LLM robustness via prompt sensitivity
topic Machine Learning
Artificial Intelligence
I.2
url https://arxiv.org/abs/2603.13285