Computer-Using Agent (CUA)

由大型多模态模型驱动的 AI 智能体，能像真人一样直接操作图形界面（点击按钮、打开窗口、执行界面级动作）来完成任务。

English Definition

"An AI agent, typically powered by a Large Multimodal Model, that can directly manipulate graphical user interfaces (e.g., clicking buttons, opening windows) to complete tasks on a computer, functioning like a human operator."

用法说明 · 针对中文母语者

CUA 是 2024–2025 年随着 LMM 兴起才出现的新词，常与 Operator（OpenAI）、Claude Computer Use 等产品名同框出现；不要简单等同于 chatbot。

常见错误

·误写为 Computer Use Agent 或 Computer-Use Agent；首字母大写后缩写 CUA 应保持全大写。

真实用例 · 3 条

"The recent advent of Computer-Using Agents (CUA) based on Large Multimodal Models, which can directly manipulate graphical user interfaces, offers new opportunities for such accessibility."

论文Pipeline extracted — review needed·摘要里点出 CUA 是为视障用户带来无障碍机会的新技术

"Recent advances in Large Multimodal Model (LMM) have led to the emergence of Computer-Using Agent (CUA), which can operate computers like human users (e.g., opening windows, clicking buttons, and executing interface-level actions)"

论文Pipeline extracted — review needed·引言介绍 CUA 的技术背景与能力

"We employed Operator, a CUA model released by OpenAI [75] in January 2025"

论文Pipeline extracted — review needed·实验部分说明使用的具体 CUA 模型

由 pipeline 自动采集，待人工 review