Code to Think, Think to Code: A Survey on Code-Enhanced Reasoning and Reasoning-Driven Code Intelligence in LLMs

TL;DR

This survey explores how integrating code and reasoning in large language models enhances their ability to perform complex software tasks, emphasizing the mutual reinforcement between structured code and logical reasoning.

Contribution

It provides a comprehensive overview of recent advances in code-enhanced reasoning and reasoning-driven code intelligence in LLMs, highlighting key challenges and future directions.

Findings

Code offers a structured medium for reasoning in LLMs.

Reasoning improves code intelligence from basic completion to complex tasks.

Enhanced reasoning enables better planning and debugging in software engineering.

Abstract

In large language models (LLMs), code and reasoning reinforce each other: code offers an abstract, modular, and logic-driven structure that supports reasoning, while reasoning translates high-level goals into smaller, executable steps that drive more advanced code intelligence. In this study, we examine how code serves as a structured medium for enhancing reasoning: it provides verifiable execution paths, enforces logical decomposition, and enables runtime validation. We also explore how improvements in reasoning have transformed code intelligence from basic completion to advanced capabilities, enabling models to address complex software engineering tasks through planning and debugging. Finally, we identify key challenges and propose future research directions to strengthen this synergy, ultimately improving LLM's performance in both areas.