Kimi-Dev: Agentless Training as Skill Prior for SWE-Agents
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , , , , , , , , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866912753811193856 |
|---|---|
| author | Yang, Zonghan Wang, Shengjie Fu, Kelin He, Wenyang Xiong, Weimin Liu, Yibo Miao, Yibo Gao, Bofei Wang, Yejie Ma, Yingwei Li, Yanhao Liu, Yue Hu, Zhenxing Zhang, Kaitai Wang, Shuyi Chen, Huarong Sung, Flood Liu, Yang Gao, Yang Yang, Zhilin Liu, Tianyu |
| author_facet | Yang, Zonghan Wang, Shengjie Fu, Kelin He, Wenyang Xiong, Weimin Liu, Yibo Miao, Yibo Gao, Bofei Wang, Yejie Ma, Yingwei Li, Yanhao Liu, Yue Hu, Zhenxing Zhang, Kaitai Wang, Shuyi Chen, Huarong Sung, Flood Liu, Yang Gao, Yang Yang, Zhilin Liu, Tianyu |
| contents | Large Language Models (LLMs) are increasingly applied to software engineering (SWE), with SWE-bench as a key benchmark. Solutions are split into SWE-Agent frameworks with multi-turn interactions and workflow-based Agentless methods with single-turn verifiable steps. We argue these paradigms are not mutually exclusive: reasoning-intensive Agentless training induces skill priors, including localization, code edit, and self-reflection that enable efficient and effective SWE-Agent adaptation. In this work, we first curate the Agentless training recipe and present Kimi-Dev, an open-source SWE LLM achieving 60.4\% on SWE-bench Verified, the best among workflow approaches. With additional SFT adaptation on 5k publicly-available trajectories, Kimi-Dev powers SWE-Agents to 48.6\% pass@1, on par with that of Claude 3.5 Sonnet (241022 version). These results show that structured skill priors from Agentless training can bridge workflow and agentic frameworks for transferable coding agents. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2509_23045 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Kimi-Dev: Agentless Training as Skill Prior for SWE-Agents Yang, Zonghan Wang, Shengjie Fu, Kelin He, Wenyang Xiong, Weimin Liu, Yibo Miao, Yibo Gao, Bofei Wang, Yejie Ma, Yingwei Li, Yanhao Liu, Yue Hu, Zhenxing Zhang, Kaitai Wang, Shuyi Chen, Huarong Sung, Flood Liu, Yang Gao, Yang Yang, Zhilin Liu, Tianyu Artificial Intelligence Computation and Language Software Engineering Large Language Models (LLMs) are increasingly applied to software engineering (SWE), with SWE-bench as a key benchmark. Solutions are split into SWE-Agent frameworks with multi-turn interactions and workflow-based Agentless methods with single-turn verifiable steps. We argue these paradigms are not mutually exclusive: reasoning-intensive Agentless training induces skill priors, including localization, code edit, and self-reflection that enable efficient and effective SWE-Agent adaptation. In this work, we first curate the Agentless training recipe and present Kimi-Dev, an open-source SWE LLM achieving 60.4\% on SWE-bench Verified, the best among workflow approaches. With additional SFT adaptation on 5k publicly-available trajectories, Kimi-Dev powers SWE-Agents to 48.6\% pass@1, on par with that of Claude 3.5 Sonnet (241022 version). These results show that structured skill priors from Agentless training can bridge workflow and agentic frameworks for transferable coding agents. |
| title | Kimi-Dev: Agentless Training as Skill Prior for SWE-Agents |
| topic | Artificial Intelligence Computation and Language Software Engineering |
| url | https://arxiv.org/abs/2509.23045 |