WebArena

WebArena: A Realistic Web Environment for Building Autonomous Agents

812 个长程自然语言任务,运行于自托管、功能完整的网站(电商、论坛、GitLab、CMS)以及地图、计算器和 wiki。智能体通过浏览器行动,以对最终站点状态或答案的程序化功能检查评分,而非匹配动作序列。是首个可复现的端到端网页智能体基准,至今仍是文本/DOM 网页智能体的参考。

论文 · 官网 · 榜单 · 数据集 · 代码

74.3% · 人类 78.24%
WebTactix + Deepseek v3.2
官方榜单 2026-02-01

发布: 2023-07
维护者: Carnegie Mellon University (Zhou, Xu et al.)
状态: active
污染风险: high
指标: success rate (percent, ↑)
题量: 812
领域: web tool-use computer-use
人类: 78.24% human annotators (paper, end-to-end task success)

备注. Leaderboard is a maintainer-curated Google Sheet of self-reported results (trajectories required since September 2024). Environments are public Docker images, so task configs are widely available. Top systems now approach the 78% human figure; the maintainers point to WebArena-Infinity and TheAgentCompany as successors, neither of which is tracked here yet.

完整账本

系统	开发者	分数	日期	来源	条件
WebTactix + Deepseek v3.2	WebTactix	74.3%	2026-02-01	官方榜单	scaffold: WebTactix Top of the maintainer sheet as of access date. Month only; 1st used.
ColorBrowserAgent	ColorBrowserAgent	71.2%	2025-12-01	官方榜单	scaffold: ColorBrowserAgent Open-source entry with trajectories. Month only; 1st used.
OpenAI Operator	OpenAI	58.1%	2025-01-01	官方榜单	scaffold: OpenAI CUA Self-reported by OpenAI (system card), listed on the maintainer sheet. Month only; 1st used.
SteP	SteP (leaderboard result source)	33.5%	2024-04-01	官方榜单	scaffold: SteP Sheet notes high-level plans are human-derived. Month only; 1st used.
gpt-4-0613	OpenAI	14.9%	2023-06-01	官方榜单	WebArena team baseline without the 'not achievable' hint (paper reports 14.41% with it). Sheet gives month only; 1st used.