WebArena

What it tests: A self-hosted replica of real websites (shopping, forums, maps, CMS). Agents must complete multi-step tasks by actually clicking and typing in a browser.

考什么:自托管的仿真网站环境(购物/论坛/地图/CMS),agent 要在浏览器里真点真输入,完成多步任务。

Why it counts: The standard for 'can this model operate a computer' — note: like all agent benchmarks, verify the setup; some reported scores use heavy scaffolding.

为什么算数:'模型会不会操作电脑'的标准考场。注意:Agent 类成绩普遍受脚手架影响大,我们只收录标注了测试方式的成绩。