Alibaba’s Qwen team introduced Qwen Intelligence on X on September 23, a set of three mobile AI agents built to plan, execute, and create on smartphones. The Mobile Planner Agent breaks down complex tasks and coordinates them. The Mobile-Use Agent completes actions across apps, calling APIs first and falling back to the touchscreen when needed. The Mobile Creative Agent turns a single written prompt into a finished image.
Qwen reported a 90 percent end to end success rate for the Mobile-Use Agent, along with scores of 82.1 on its MobileWorld benchmark, 92.2 on MobileWorld-Real, and 97.2 on AndroidDaily. The Mobile Creative Agent generates an image in three seconds, which Qwen described as roughly twice as fast as competing tools. Every one of these figures comes from Qwen’s own testing; the announcement includes no independent benchmark results.
Qwen also opened its benchmark suite for outside use: MobilePA-Bench, MobileWorld, MobileWorld-Real, and MobileWorld-Safety. Together the four suites score planning skill, cross-app execution, on-device performance, and safety as separate categories. That release invites scrutiny Qwen has not yet faced. Rival labs building mobile agents can now run the same tests and check the numbers directly, instead of taking Qwen’s self-reported results at face value.
Based on Qwen’s official announcement on X, September 23, 2026.
