新闻
OSWorld-Pro: Process-based Evaluation for Computer Use Agents
arXiv cs.AI · 发布于 · 阅读约3分钟
30秒读懂
- 发生了什么
- OSWorld-Pro introduces over 300 tasks with 2800 subgoals and 67000 human annotations to evaluate computer-use agents step-by-step rather than only final outcomes.
- 为何重要
- Engineers building or testing AI agents that interact with graphical interfaces need granular failure analysis beyond pass-fail metrics to guide improvements.
- 注意
- Top models like Claude Opus 5 score 75.7% on OSWorld-Pro versus 83.4% on OSWorld, suggesting the benchmark is substantially harder and may not reflect real-world difficulty.
- agent
- eval
- phi
这条新闻背后的模式
每个模式都讲清楚技术如何运作、何时值得投入，以及在哪里会失效。
The Agent Architect
每周一个模式、一个权衡、一个生产事故案例。为构建智能体系统的人准备的每周简报。
每周一封邮件,一键退订。您的地址仅用于发送简报。