SWE-Duel：面向编码智能体的自扩展竞技场
原标题：SWE-Duel: Towards a Self-Scaling Arena for Coding Agents
AI 摘要
论文提出 SWE-Duel，一个自扩展的编码智能体竞技场：两个智能体在镜像的红蓝回合中互相评测，红方在真实仓库中实现新功能并注入隐蔽 bug，蓝方在不知情下审查并修复 PR，任务与判定标准均在评测时生成，避免训练数据泄漏。作者在 5 种语言的 15 个仓库上对 11 个模型-框架组合进行了 1500 回合的锦标赛，GPT-5.5 搭配 Codex CLI 以 98.3% 的缺陷修复率成为最强解题者，Claude Opus 4.8 搭配 mini-swe-agent 以 27.3% 的红方胜率成为最强任务作者。初步证据显示，同一模型家族中最新版本编写的任务比旧版本更能击败固定参考解题者（42.9% vs 4.0%），表明任务难度随任务编写能力扩展。
正文节选
SWE-Duel: Towards A Self-Scaling Arena for Coding Agents Abstract Static coding benchmarks are fixed at release. Over time their tasks leak into training data, and stronger agents eventually saturate them. We argue that a benchmark should instead aim to be self-scaling: its tasks and oracles are created at evaluation time, so no content exists in advance to leak, and its difficulty is set by the participating agents rather than fixed at release. We introduce SWE-Duel, an arena in which two codin