RiddleBench: A New Generative Reasoning Benchmark for LLMs

TL;DR

RiddleBench is a new challenging reasoning benchmark with 1,737 puzzles designed to evaluate flexible, multifaceted reasoning abilities in large language models, revealing significant weaknesses in current state-of-the-art models.

Contribution

The paper introduces RiddleBench, a novel benchmark for assessing complex reasoning skills in LLMs, highlighting their limitations and guiding future improvements.

Findings

Top models achieve just over 60% accuracy.

Models exhibit hallucination cascades and poor self-correction.

Performance drops with reordered constraints or irrelevant info.

Abstract

Large Language Models have demonstrated strong performance on many established reasoning benchmarks. However, these benchmarks primarily evaluate structured skills like quantitative problem-solving, leaving a gap in assessing flexible, multifaceted reasoning abilities that are central to human intelligence. These abilities require integrating logical deduction with spatial awareness and constraint satisfaction, which current evaluations do not measure well. To address this, we introduce RiddleBench, a benchmark of 1,737 challenging puzzles in English designed to probe these core reasoning capabilities. Evaluation of state-of-the-art models on RiddleBench shows fundamental weaknesses. Even top proprietary models like Gemini 2.5 Pro, o3, and Claude 4 Sonnet achieve accuracy just above 60% (60.30%, 63.37%, and 63.16%). Analysis further reveals deep failures, including hallucination cascades (accepting flawed reasoning from other models) and poor self-correction due to a strong self-confirmation bias. Their reasoning is also fragile, with performance degrading significantly when constraints are reordered or irrelevant information is introduced. RiddleBench functions as a diagnostic tool for these issues and as a resource for guiding the development of more robust and reliable language models.