Benchmarking Overton Pluralism in LLMs

TL;DR

This paper introduces OVERTONBENCH, a framework for measuring the diversity of viewpoints in LLM outputs, including a formal metric, large-scale human study, and an automated benchmark that correlates well with human judgments.

Contribution

The paper formalizes Overton pluralism as a set coverage metric, conducts a large-scale human evaluation, and develops an automated benchmark that closely matches human assessments.

Findings

Models achieve OVERTONSCOREs of 0.35--0.41

DeepSeek V3 performs best among tested models

Automated benchmark correlates highly with human judgments (ρ = 0.88)

Abstract

We introduce OVERTONBENCH, a novel framework for measuring Overton pluralism in LLMs--the extent to which diverse viewpoints are represented in model outputs. We (i) formalize Overton pluralism as a set coverage metric (OVERTONSCORE), (ii) conduct a large-scale U.S.-representative human study (N = 1208; 60 questions; 8 LLMs), and (iii) develop an automated benchmark that closely reproduces human judgments. On average, models achieve OVERTONSCOREs of 0.35--0.41, with DeepSeek V3 performing best; yet all models remain far below the theoretical maximum of 1.0, revealing substantial headroom for improvement. Because repeated large-scale human studies are costly and slow, scalable evaluation tools are essential for model development. Hence, we propose an automated benchmark that achieves high rank correlation with human judgments ($\rho = 0.88$), providing a practical proxy without replacing human assessment. By turning pluralistic alignment from a normative aim into a measurable benchmark, our work establishes a foundation for systematic progress toward more pluralistic LLMs.