VerifyBench: A Systematic Benchmark for Evaluating Reasoning Verifiers Across Domains

TL;DR

This paper introduces VerifyBench, a comprehensive cross-domain benchmark with 4,000 expert-annotated questions to systematically evaluate and compare the performance of various verifiers, revealing fundamental trade-offs and limitations.

Contribution

It presents VerifyBench, the first systematic benchmark for evaluating verifier performance across multiple domains, and provides insights into their strengths, weaknesses, and generalization challenges.

Findings

Specialized verifiers have higher accuracy but lower recall.

General models are more inclusive but less precise.

Verifiers are highly sensitive to input structure and domain shifts.

Abstract

Large language models (LLMs) increasingly rely on reinforcement learning (RL) to enhance their reasoning capabilities through feedback. A critical challenge is verifying the consistency of model-generated responses and reference answers, since these responses are often lengthy, diverse, and nuanced. Rule-based verifiers struggle with complexity, prompting the use of model-based verifiers. However, specialized verifiers lack flexibility, while general LLM judges can be inconsistent. Existing research primarily focuses on building better verifiers, yet a systematic evaluation of different types of verifiers' performance across domains remains lacking, severely constraining the reliable development of Reinforcement Learning with Verifiable Reward (RLVR). To address this, we propose VerifyBench--a cross-domain comprehensive benchmark for systematically evaluating verifiers. We construct 4,000 expert-level questions covering mathematics, physics, chemistry, and biology. Each question is equipped with reference answers and diverse responses. The reliability of the evaluation is ensured through a rigorous annotation process conducted by a multidisciplinary expert team. We design a four-dimensional experimental framework to comprehensively compare the performance boundaries of specialized verifiers and general LLMs under combined conditions of extracted answers vs. complete responses, and short vs. long outputs. Our evaluation uncovers fundamental trade-offs in verifiers: while specialized verifiers achieve leading accuracy, they exhibit deficiencies in recall; general models show stronger inclusivity but unstable precision. More importantly, we discover verifiers' high sensitivity to input structure and inherent limitations in cross-domain generalization, providing critical insights into the bottlenecks of current verifier technology.