_version_ 1866912809272475648
author Ding, Jingzhe
Long, Shengda
Pu, Changxin
Zhou, Huan
Gao, Hongwan
Gao, Xiang
He, Chao
Hou, Yue
Hu, Fei
Li, Zhaojian
Shi, Weiran
Wang, Zaiyuan
Zan, Daoguang
Zhang, Chenchen
Zhang, Xiaoxu
Chen, Qizhi
Cheng, Xianfu
Deng, Bo
Gu, Qingshui
Hua, Kai
Lin, Juntao
Liu, Pai
Li, Mingchen
Pan, Xuanguang
Peng, Zifan
Qin, Yujia
Shan, Yong
Tan, Zhewen
Xie, Weihao
Wang, Zihan
Yuan, Yishuo
Zhang, Jiayu
Zhao, Enduo
Zhao, Yunfei
Zhu, He
Zhu, Liya
Zou, Chenyang
Ding, Ming
Jiao, Jianpeng
Liu, Jiaheng
Liu, Minghao
Liu, Qian
Tao, Chongyang
Yang, Jian
Yang, Tong
Zhang, Zhaoxiang
Chen, Xinjie
Huang, Wenhao
Zhang, Ge
author_facet Ding, Jingzhe
Long, Shengda
Pu, Changxin
Zhou, Huan
Gao, Hongwan
Gao, Xiang
He, Chao
Hou, Yue
Hu, Fei
Li, Zhaojian
Shi, Weiran
Wang, Zaiyuan
Zan, Daoguang
Zhang, Chenchen
Zhang, Xiaoxu
Chen, Qizhi
Cheng, Xianfu
Deng, Bo
Gu, Qingshui
Hua, Kai
Lin, Juntao
Liu, Pai
Li, Mingchen
Pan, Xuanguang
Peng, Zifan
Qin, Yujia
Shan, Yong
Tan, Zhewen
Xie, Weihao
Wang, Zihan
Yuan, Yishuo
Zhang, Jiayu
Zhao, Enduo
Zhao, Yunfei
Zhu, He
Zhu, Liya
Zou, Chenyang
Ding, Ming
Jiao, Jianpeng
Liu, Jiaheng
Liu, Minghao
Liu, Qian
Tao, Chongyang
Yang, Jian
Yang, Tong
Zhang, Zhaoxiang
Chen, Xinjie
Huang, Wenhao
Zhang, Ge
contents Recent advances in coding agents suggest rapid progress toward autonomous software development, yet existing benchmarks fail to rigorously evaluate the long-horizon capabilities required to build complete software systems. Most prior evaluations focus on localized code generation, scaffolded completion, or short-term repair tasks, leaving open the question of whether agents can sustain coherent reasoning, planning, and execution over the extended horizons demanded by real-world repository construction. To address this gap, we present NL2Repo Bench, a benchmark explicitly designed to evaluate the long-horizon repository generation ability of coding agents. Given only a single natural-language requirements document and an empty workspace, agents must autonomously design the architecture, manage dependencies, implement multi-module logic, and produce a fully installable Python library. Our experiments across state-of-the-art open- and closed-source models reveal that long-horizon repository generation remains largely unsolved: even the strongest agents achieve below 40% average test pass rates and rarely complete an entire repository correctly. Detailed analysis uncovers fundamental long-horizon failure modes, including premature termination, loss of global coherence, fragile cross-file dependencies, and inadequate planning over hundreds of interaction steps. NL2Repo Bench establishes a rigorous, verifiable testbed for measuring sustained agentic competence and highlights long-horizon reasoning as a central bottleneck for the next generation of autonomous coding agents.
format Preprint
id arxiv_https___arxiv_org_abs_2512_12730
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle NL2Repo-Bench: Towards Long-Horizon Repository Generation Evaluation of Coding Agents
Ding, Jingzhe
Long, Shengda
Pu, Changxin
Zhou, Huan
Gao, Hongwan
Gao, Xiang
He, Chao
Hou, Yue
Hu, Fei
Li, Zhaojian
Shi, Weiran
Wang, Zaiyuan
Zan, Daoguang
Zhang, Chenchen
Zhang, Xiaoxu
Chen, Qizhi
Cheng, Xianfu
Deng, Bo
Gu, Qingshui
Hua, Kai
Lin, Juntao
Liu, Pai
Li, Mingchen
Pan, Xuanguang
Peng, Zifan
Qin, Yujia
Shan, Yong
Tan, Zhewen
Xie, Weihao
Wang, Zihan
Yuan, Yishuo
Zhang, Jiayu
Zhao, Enduo
Zhao, Yunfei
Zhu, He
Zhu, Liya
Zou, Chenyang
Ding, Ming
Jiao, Jianpeng
Liu, Jiaheng
Liu, Minghao
Liu, Qian
Tao, Chongyang
Yang, Jian
Yang, Tong
Zhang, Zhaoxiang
Chen, Xinjie
Huang, Wenhao
Zhang, Ge
Computation and Language
Recent advances in coding agents suggest rapid progress toward autonomous software development, yet existing benchmarks fail to rigorously evaluate the long-horizon capabilities required to build complete software systems. Most prior evaluations focus on localized code generation, scaffolded completion, or short-term repair tasks, leaving open the question of whether agents can sustain coherent reasoning, planning, and execution over the extended horizons demanded by real-world repository construction. To address this gap, we present NL2Repo Bench, a benchmark explicitly designed to evaluate the long-horizon repository generation ability of coding agents. Given only a single natural-language requirements document and an empty workspace, agents must autonomously design the architecture, manage dependencies, implement multi-module logic, and produce a fully installable Python library. Our experiments across state-of-the-art open- and closed-source models reveal that long-horizon repository generation remains largely unsolved: even the strongest agents achieve below 40% average test pass rates and rarely complete an entire repository correctly. Detailed analysis uncovers fundamental long-horizon failure modes, including premature termination, loss of global coherence, fragile cross-file dependencies, and inadequate planning over hundreds of interaction steps. NL2Repo Bench establishes a rigorous, verifiable testbed for measuring sustained agentic competence and highlights long-horizon reasoning as a central bottleneck for the next generation of autonomous coding agents.
title NL2Repo-Bench: Towards Long-Horizon Repository Generation Evaluation of Coding Agents
topic Computation and Language
url https://arxiv.org/abs/2512.12730