GeoChallenge: A Multi-Answer Multiple-Choice Benchmark for Geometric Reasoning with Diagrams

TL;DR

GeoChallenge is a large-scale geometry reasoning benchmark with 90K multi-step, diagram-grounded multiple-choice problems designed to evaluate and analyze the reasoning capabilities of large language models.

Contribution

It introduces a novel, extensive dataset with formal annotations and complexity ratings for evaluating LLMs on geometric reasoning tasks involving diagrams.

Findings

GPT-5-nano achieves 75.89% accuracy, below human performance of 94.74%.

Identified common failure patterns in LLM reasoning, including reliance on visual cues and reasoning convergence issues.

Benchmark enables controlled, fine-grained evaluation of geometric reasoning in LLMs.

Abstract

Evaluating the symbolic reasoning of large language models (LLMs) calls for geometry benchmarks that require multi-step proofs grounded in both text and diagrams. However, existing benchmarks are often limited in scale and rarely provide visually grounded multiple-choice questions, limiting reliable evaluation of complex reasoning. We introduce GeoChallenge, a dataset of 90K automatically generated multiple-choice geometry proof problems, each requiring multi-step reasoning over aligned textual descriptions and diagrams. GeoChallenge provides fine-grained complexity ratings and formal language annotations to enable controlled evaluation. Experiments on multiple advanced LLMs show a clear performance gap between models and humans (the best-performing model, GPT-5-nano, achieves 75.89 exact match vs. 94.74 for humans). Further analysis also reveals three common failure patterns of LLMs: (1) exact match failures under the multiple-choice setting; (2) weak visual reliance; and (3) overextended reasoning without convergence.