A Fragile Number Sense: Probing the Elemental Limits of Numerical Reasoning in LLMs

2025-09-08

AI Summary

Large language models (LLMs) exhibit fundamental limitations in generative numerical reasoning—excelling at deterministic algorithmic tasks but struggling with complex mathematical problems requiring heuristic search and creative problem-solving. Method: The authors introduce a hierarchical evaluation framework comprising 100 problems across four domains—basic arithmetic, advanced operations, primality testing, and the 24 Game—to systematically assess state-of-the-art LLM-based agents. Contribution/Results: Results show high accuracy (>85%) on structured, deterministic tasks but severe failure on the 24 Game—requiring combinatorial exploration and strategic trial-and-error—with average accuracy below 20%. This study is the first to empirically demonstrate, via controlled, progressively difficult task design, that LLMs’ numerical reasoning relies fundamentally on pattern matching rather than generative, constructive thinking. It establishes a reproducible benchmark and diagnostic methodology to guide future efforts in enhancing LLMs’ mathematical creativity.

Technology Category

• Cognitive Modeling & Cognitive Systems: Computational Creativity
• Knowledge Representation and Reasoning: Computational Complexity of Reasoning
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Large Language Models (LLMs) have demonstrated remarkable emergent capabilities, yet the robustness of their numerical reasoning remains an open question. While standard benchmarks evaluate LLM reasoning on complex problem sets using aggregated metrics, they often obscure foundational weaknesses. In this work, we probe LLM mathematical numeracy by evaluating performance on problems of escalating complexity, from constituent operations to combinatorial puzzles. We test several state-of-the-art LLM-based agents on a 100-problem challenge comprising four categories: (1) basic arithmetic, (2) advanced operations, (3) primality checking, and (4) the Game of 24 number puzzle. Our results show that while the agents achieved high accuracy on the first three categories, which require deterministic algorithmic execution, they consistently failed at the number puzzle, underlining its demand for a heuristic search over a large combinatorial space to be a significant bottleneck. These findings reveal that the agents' proficiency is largely confined to recalling and executing known algorithms, rather than performing generative problem-solving. This suggests their apparent numerical reasoning is more akin to sophisticated pattern-matching than flexible, analytical thought, limiting their potential for tasks that require novel or creative numerical insights.

Problem

• Assessing LLM robustness in numerical reasoning tasks
• Identifying foundational weaknesses beyond standard benchmark metrics
• Evaluating performance on combinatorial puzzles requiring heuristic search

Innovation

• Probing numerical reasoning via escalating complexity
• Testing LLMs on arithmetic, operations, primality, puzzles
• Revealing pattern-matching over generative problem-solving