LR^2Bench: Evaluating Long-chain Reflective Reasoning Capabilities of Large Language Models via Constraint Satisfaction Problems

TL;DR

LR$^2$Bench is a new benchmark with 850 constraint satisfaction problems designed to evaluate the long-chain reflective reasoning abilities of large language models, revealing current models' limited performance and highlighting the need for further improvements.

Contribution

This paper introduces LR$^2$Bench, the first comprehensive benchmark specifically designed to evaluate the reflective reasoning capabilities of large language models across diverse constraint satisfaction tasks.

Findings

State-of-the-art LRMs achieve only around 20-24% accuracy on LR$^2$Bench.

Reflective reasoning remains a significant challenge for current large language models.

LR$^2$Bench covers diverse constraint types, providing a thorough evaluation framework.

Abstract

Recent progress in Large Reasoning Models (LRMs) has significantly enhanced the reasoning abilities of Large Language Models (LLMs), empowering them to tackle increasingly complex tasks through reflection capabilities, such as making assumptions, backtracking, and self-refinement. However, effectively evaluating such reflection capabilities remains challenging due to the lack of appropriate benchmarks. To bridge this gap, we introduce LR$^2$Bench, a novel benchmark designed to evaluate the Long-chain Reflective Reasoning capabilities of LLMs. LR$^2$Bench comprises 850 samples across six Constraint Satisfaction Problems (CSPs) where reflective reasoning is crucial for deriving solutions that meet all given constraints. Each type of task focuses on distinct constraint patterns, such as knowledge-based, logical, and spatial constraints, providing a comprehensive evaluation of diverse problem-solving scenarios. Our extensive evaluation on both conventional LLMs and LRMs reveals that even the most advanced LRMs, such as DeepSeek-R1 and OpenAI o1-preview, struggle with tasks in LR$^2$Bench, achieving an average Exact Match score of only 20.0% and 23.6%, respectively. These findings underscore the significant room for improvement in the reflective reasoning capabilities of current LLMs.