WebUniverse：为智能体网页开发扩展可验证训练数据
原标题：WebUniverse: Scaling Verified Training Data for Agentic Web Development
AI 摘要
该论文提出 WebUniverse，一个从开源代码仓库和真实网页构建可验证前端开发任务与开发轨迹的流水线，覆盖文本需求、迭代修改和多视口视觉复现等场景。作者发布了包含 1.5 万个可执行任务环境的 WebUniverse-Env 和对应 1.5 万条带验证记录的开发轨迹 WebUniverse-Mix。在 WebUniverse-Mix 上微调 Qwen3.5-35B-A3B 后，Vision2Web 平均分从 51.7 提升至 68.3，WebGen-Bench 准确率从 34.5% 升至 63.9%，Web-Bench pass@2 从 12.2% 升至 23.1%，且优于同配置同样本量的公开数据基线。
正文节选
WebUniverse: Scaling Verified Training Data for Agentic Web Development Abstract Training agents for front-end web development requires demonstrations of code editing, execution, browser inspection and revision, yet public trajectories and reusable task environments remain limited. We introduce WebUniverse, a pipeline that constructs verifiable tasks with verified development trajectories from open-source repositories and real web pages. The tasks cover text-grounded application development in s