English

LoCoBench-Agent: An Interactive Benchmark for LLM Agents in Long-Context Software Engineering

Software Engineering 2025-11-19 v1 Artificial Intelligence

Abstract

As large language models (LLMs) evolve into sophisticated autonomous agents capable of complex software development tasks, evaluating their real-world capabilities becomes critical. While existing benchmarks like LoCoBench~\cite{qiu2025locobench} assess long-context code understanding, they focus on single-turn evaluation and cannot capture the multi-turn interactive nature, tool usage patterns, and adaptive reasoning required by real-world coding agents. We introduce \textbf{LoCoBench-Agent}, a comprehensive evaluation framework specifically designed to assess LLM agents in realistic, long-context software engineering workflows. Our framework extends LoCoBench's 8,000 scenarios into interactive agent environments, enabling systematic evaluation of multi-turn conversations, tool usage efficiency, error recovery, and architectural consistency across extended development sessions. We also introduce an evaluation methodology with 9 metrics across comprehension and efficiency dimensions. Our framework provides agents with 8 specialized tools (file operations, search, code analysis) and evaluates them across context lengths ranging from 10K to 1M tokens, enabling precise assessment of long-context performance. Through systematic evaluation of state-of-the-art models, we reveal several key findings: (1) agents exhibit remarkable long-context robustness; (2) comprehension-efficiency trade-off exists with negative correlation, where thorough exploration increases comprehension but reduces efficiency; and (3) conversation efficiency varies dramatically across models, with strategic tool usage patterns differentiating high-performing agents. As the first long-context LLM agent benchmark for software engineering, LoCoBench-Agent establishes a rigorous foundation for measuring agent capabilities, identifying performance gaps, and advancing autonomous software development at scale.

Keywords

Cite

@article{arxiv.2511.13998,
  title  = {LoCoBench-Agent: An Interactive Benchmark for LLM Agents in Long-Context Software Engineering},
  author = {Jielin Qiu and Zuxin Liu and Zhiwei Liu and Rithesh Murthy and Jianguo Zhang and Haolin Chen and Shiyu Wang and Ming Zhu and Liangwei Yang and Juntao Tan and Roshan Ram and Akshara Prabhakar and Tulika Awalgaonkar and Zixiang Chen and Zhepeng Cen and Cheng Qian and Shelby Heinecke and Weiran Yao and Silvio Savarese and Caiming Xiong and Huan Wang},
  journal= {arXiv preprint arXiv:2511.13998},
  year   = {2025}
}

Comments

54-pages

R2 v1 2026-07-01T07:42:24.757Z