Alibaba Qwen launches Qwen Intelligence mobile agent suite with open benchmarks, 90% end-to-end success
At the Apsara Conference on September 22, Alibaba's Qwen team launched Qwen Intelligence, a personal AI suite for phones with three SOTA mobile agents and open benchmarks; Mobile-Use reports 90% end-to-end success.
At the Apsara Conference on September 22, Alibaba's Qwen team officially launched Qwen Intelligence, bringing personal AI to phones with three SOTA mobile agents. Mobile Planner Agent handles complex task planning, decomposition, and orchestration and tops MobilePA-Bench plus its Business and Memory variants. Mobile-Use Agent prefers API calls with GUI fallback, scoring 82.1 on MobileWorld, 92.2 on MobileWorld-Real, and 97.2 on AndroidDaily, with a 90% end-to-end success rate. Mobile Creative Agent produces ready-to-deploy creative assets in about three seconds, roughly 2x faster than peer tools.
Qwen open-sourced its benchmark suite covering MobilePA-Bench, MobileWorld, MobileWorld-Real, and MobileWorld-Safety, spanning planning, cross-app execution, and safety. On the same day, Tongyi-MAI released the Qwen-Planner-Agent technical report (arXiv:2609.29892), introducing CARE (Competence-Aware Reward-and-Advantage Engineering) within a closed-loop AI-for-AI framework spanning AI-for-Data, AI-for-Training, and model-harness co-evolution. Qwen-Planner-Agent posts the best overall score on MobilePA-Bench, beating its base model on tool use, memory, skills, and sub-agent coordination.
The Honor Magic9 series, scheduled for release on September 28, will be the first phones to officially carry the Qwen Intelligence solution, with Honor Robot Phone also supported. The combined solution achieves 91.8% overall task accuracy, 3.6-second GUI operation latency, more than 100 continuous steps for complex long-horizon tasks, and over 90% end-to-end service closure. It pairs tightly with Honor's MagicOS 11, the industry's first commercially deployed system-level Agent Harness for mobile.