CASE EVIDENCE / A RECORD
Princeton NLP / SWE-agent (Karol Rajewski et al.) 公开的代码代理与软件工程案例，来源为 公开代码库，复核于 2026-06-26T13:55:38Z。
原始记录：SWE-agent 1.0 achieves state-of-the-art on SWE-bench using Claude 3.7 Sonnet as backbone model
Chinese Brief
Princeton NLP / SWE-agent (Karol Rajewski et al.) 公开的代码代理与软件工程案例，来源为 公开代码库，复核于 2026-06-26T13:55:38Z。 Model Atlas 将它标记为 A 类证据，因为它同时具备具体使用者、具体任务、公开原始证据和可访问产物。 Model Atlas 不把 benchmark、教程、发布说明或集合页包装成真实案例。
任务
这是一个围绕软件工程任务执行的真实任务，公开材料可以回溯到具体使用者和具体产物。 原始资料写作：The SWE-agent team at Princeton NLP used Claude 3.7 Sonnet as the backbone LLM for their open-source autonomous software engineering agent. SWE-agent 1.0 + Claude 3.7 achieved state-of-the-art results on both SWE-bench …
公开材料提供公开代码、README 或项目配置，可用于核验任务结果、项目形态和模型绑定关系。 原始资料写作：SWE-agent 1.0 paired with Claude 3.7 Sonnet achieved the highest resolve rates on SWE-bench Verified and Full leaderboards at the time of release, surpassing all prior agent systems. The source code explicitly handles C…
Claude 3.7 Sonnet 在该案例中承担软件工程任务执行相关的生成、分析、编排或实现角色。 原始资料写作：Claude 3.7 Sonnet served as the reasoning backbone for SWE-agent's autonomous software engineering pipeline. Its code understanding, multi-step reasoning, and tool-use capabilities enabled the agent to autonomously navi…
当前判断基于公开材料；若产物下线、仓库变更或模型参与比例仅来自作者自述，需要在引用前重新复核。 原始资料写作：SWE-bench results are benchmark evaluations, but SWE-agent is a real open-source tool used for autonomous software engineering on real codebases. The model's contribution is demonstrated through a concrete, reproducible…