arXiv:2509.23045v3 Announce Type: replace-cross 
Abstract: 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.

تم تقديم Kimi-Dev كنموذج لغة مفتوح المصدر (LLM) مصمم لهندسة البرمجيات (SWE)، حيث حقق نسبة 60.4% ملحوظة في معيار SWE-bench Verified. يستخدم هذا النموذج تدريبًا بدون وكيل لتطوير أولويات المهارات التي تعزز أداء SWE-Agents، مما يدل على تقدم كبير في دمج أساليب التدريب المنظمة في تطوير الذكاء الاصطناعي.

Kimi-Dev se ha presentado como un modelo de lenguaje de código abierto (LLM) diseñado para la ingeniería de software (SWE), logrando un notable 60.4% en el benchmark SWE-bench Verified. Este modelo utiliza entrenamiento sin agente para desarrollar prioridades de habilidades que mejoran el rendimiento de los SWE-Agents, demostrando un avance significativo en la integración de métodos de entrenamiento estructurados en el desarrollo de IA.

Kimi-Dev a été introduit comme un modèle de langage open-source (LLM) conçu pour l'ingénierie logicielle (SWE), atteignant un score notable de 60,4 % sur le benchmark SWE-bench Verified. Ce modèle utilise une formation sans agent pour développer des priorités de compétences qui améliorent les performances des SWE-Agents, démontrant une avancée significative dans l'intégration de méthodes de formation structurées dans le développement de l'IA.

Kimi-Dev has been introduced as an open-source large language model (LLM) designed for software engineering (SWE), achieving a notable 60.4% on the SWE-bench Verified benchmark. This model utilizes agentless training to develop skill priors that enhance the performance of SWE-Agents, demonstrating a significant advancement in the integration of structured training methods in AI development.

Kimi-Dev: Agentless Training as Skill Prior for SWE-Agents

Was this article worth reading? Share it

Chattermate

LucidQuery AI

Teammately