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

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Qiu, Jielin, Liu, Zuxin, Liu, Zhiwei, Murthy, Rithesh, Zhang, Jianguo, Chen, Haolin, Wang, Shiyu, Zhu, Ming, Yang, Liangwei, Tan, Juntao, Ram, Roshan, Prabhakar, Akshara, Awalgaonkar, Tulika, Chen, Zixiang, Cen, Zhepeng, Qian, Cheng, Heinecke, Shelby, Yao, Weiran, Savarese, Silvio, Xiong, Caiming, Wang, Huan
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866909909588639744
author Qiu, Jielin
Liu, Zuxin
Liu, Zhiwei
Murthy, Rithesh
Zhang, Jianguo
Chen, Haolin
Wang, Shiyu
Zhu, Ming
Yang, Liangwei
Tan, Juntao
Ram, Roshan
Prabhakar, Akshara
Awalgaonkar, Tulika
Chen, Zixiang
Cen, Zhepeng
Qian, Cheng
Heinecke, Shelby
Yao, Weiran
Savarese, Silvio
Xiong, Caiming
Wang, Huan
author_facet Qiu, Jielin
Liu, Zuxin
Liu, Zhiwei
Murthy, Rithesh
Zhang, Jianguo
Chen, Haolin
Wang, Shiyu
Zhu, Ming
Yang, Liangwei
Tan, Juntao
Ram, Roshan
Prabhakar, Akshara
Awalgaonkar, Tulika
Chen, Zixiang
Cen, Zhepeng
Qian, Cheng
Heinecke, Shelby
Yao, Weiran
Savarese, Silvio
Xiong, Caiming
Wang, Huan
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2511_13998
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle LoCoBench-Agent: An Interactive Benchmark for LLM Agents in Long-Context Software Engineering
Qiu, Jielin
Liu, Zuxin
Liu, Zhiwei
Murthy, Rithesh
Zhang, Jianguo
Chen, Haolin
Wang, Shiyu
Zhu, Ming
Yang, Liangwei
Tan, Juntao
Ram, Roshan
Prabhakar, Akshara
Awalgaonkar, Tulika
Chen, Zixiang
Cen, Zhepeng
Qian, Cheng
Heinecke, Shelby
Yao, Weiran
Savarese, Silvio
Xiong, Caiming
Wang, Huan
Software Engineering
Artificial Intelligence
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.
title LoCoBench-Agent: An Interactive Benchmark for LLM Agents in Long-Context Software Engineering
topic Software Engineering
Artificial Intelligence
url https://arxiv.org/abs/2511.13998