HuggingPapers· @HuggingPapers · X·· 7 小时前AI 评分
TraceTrial-Bench 评测智能体与框架可靠性
任务成功并不意味着框架运行正常 TraceTrial-Bench 分别评估智能体表现和框架可靠性，涵盖 95 项任务、7 个维度、4 个框架 × 9 个模型和 3,420 次运行。31.1% 的运行在任务成功的同时，未通过框架标准。
Task success doesn't mean the harness worked
TraceTrial-Bench scores agent performance and harness reliability separately, across 95 tasks, 7 dimensions, 4 harnesses × 9 models and 3,420 runs. 31.1% of runs succeed while failing a harness criterion.
来源：HuggingPapers · x.com