Qwen Intelligence claims 91.8% phone-task accuracy

Alibaba launched Qwen Intelligence on September 25, and its HONOR deployment reports 91.8% task accuracy while orchestrating flows longer than 100 steps. Alibaba’s announcement says the first release includes three ready-to-use agents and four benchmarks spanning more than 1,000 real-world tasks and 200 mobile tools. Builders evaluating mobile agents should demand an app-by-app task suite and permission audit before treating a cross-app accuracy claim as production evidence; Alibaba’s earlier Qwen3.8-Max preview analysis shows why model availability and deployment evidence matter.