Unverified50% confidenceFactExact time
SWE-RPG在SWE-bench基础上引入更细粒度的评分维度，对需求理解、规划质量、代码风格等中间环节进行独立打分
1
Sources
50%
Confidence
Long-term
Relevance
8/24/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedSWE-bench比HumanEval（单函数补全）或MBPP（简单算法题）更贴近生产环境67% similarUnverifiedSWE-bench was launched by a Princeton University research team and is specifically designed to evaluate the ability of large language models to solve real software engineering problems65% similarUnverified业界代码库理解通常采用分层摘要和RAG两类方案63% similarUnverified简单代码补全用轻量模型即可，涉及跨文件重构、架构级推理时才需要旗舰能力63% similarUnverifiedExplicitly constraining output format and style in the system prompt can reduce model style drift, making generated results more consistent with the existing codebase's consistency requirements.63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/794279API
curl https://kongchang.com/api/v1/knowledge/claims/794279MCP
get_claim(id=794279)