OpenMobile-2 证明开源手机智能体可媲美闭源模型
原标题：OpenMobile-2 proves open-source smartphone agents can rival proprietary models
AI 摘要
南京大学、商汤等机构的研究者发布开源项目 OpenMobile-2，基于 Qwen3.5（90 亿参数）和 Qwen3.6（270 亿参数）训练出两个手机智能体模型，并公开完整训练环境、数据和复现说明。他们重建了淘宝、京东、滴滴、美团、Slack 等 110 多个模拟应用，生成约 1.16 万个示例工作流，由 Gemini 3.1 Pro 作为教师模型产出训练样本。结果显示，在相同数据量下增加应用多样性比单纯增加数据更有效，未见应用成功率从 60.7% 升至 67.5%；大模型在真实 Android 设备的 SPA-Bench 上从 31.9 分升至 59.6 分，为开源模型最高。
正文节选
OpenMobile-2 proves open-source smartphone agents can rival proprietary models Key Points - The open-source OpenMobile-2 project trains AI smartphone agents in an environment with more than 110 simulated apps. - App diversity matters more than data volume. With the same amount of training data, increasing the number of apps pushed the success rate on unseen applications from 60.7 to 67.5 percent. - On real Android devices, the larger model scored 59.6 points on SPA-Bench, the highest result amon