Coding-agent benchmarks often describe a task in natural language but grade it with tests taken from the upstream project. Passing those tests can require in…
机构:阿里
来源:arXiv 2610.30635 | AI4Papers 论文推荐平台