Alibaba has unveiled Qwen Intelligence, a full-stack agentic AI solution designed to help smartphone manufacturers bring agentic AI capabilities to their devices. HONOR is the first one to do so, with HONOR Magic9 Series and HONOR Robot Phone among the first devices to incorporate the feature.
Alibaba Qwen Intelligence
Qwen Intelligence is structured across three layers: at the foundation, a Qwen-based model is optimized for smartphones, with stronger on-device capabilities at lower cost compared to porting a general-purpose model. The middle platform layer is modular and customizable, supporting “independent model deployment, harness customization, unified tool governance, and automated evaluation,” with “device-cloud coordination” balancing performance, latency, and cost. At the top, premium agents address vertical domain needs by packaging capabilities into solutions for real-world scenarios.
The first phase of Qwen Intelligence features three agents. The Mobile Planner Agent acts as the phone’s brain, handling planning and orchestration tasks – for a request such as arranging a business trip, it breaks the task down into booking flights, reserving a hotel, setting calendar events, and planning the itinerary. When disruptions occur, such as a flight cancellation, memory and proactive services are designed to step in with rebooking and alternatives.
The Mobile-Use Agent carries out actions across apps through a hybrid API-first, GUI-fallback model, enforcing strict security boundaries and privacy protection. The Mobile Creative Agent is a lightweight model optimised for content creation on smartphones, handling image generation and editing. Additional services in the pipeline include personalized memory with proactive assistance, multimodal interaction, and vertical agents from ecosystem partners.
Alibaba and HONOR have co-developed domain-specific models and solutions built on Qwen Intelligence and HONOR’s MagicOS operating system. Alibaba claims HONOR smartphones achieve a task accuracy rate of up to 91.8% and can orchestrate complex flows exceeding 100 steps.
To address the lack of standardised evaluation methods for agentic phones, the Qwen Intelligence team has built an open evaluation system comprising four benchmarks. MobilePA-Bench evaluates how well agents plan and carry out complex tasks, covering more than 1,000 real-scenario tasks and more than 200 commonly used mobile tools. Other benchmarks test long-horizon, cross-app tasks in cloud-phone environments and real-device tasks that reflect everyday usage patterns.
Pokdepinion: Your phone booking flights and hotels automatically sounds great, until it picks the middle seat.
