Alibaba’s newly announced Qwen Intelligence platform is aimed at smartphone manufacturers. Alibaba Cloud says it is designed to let agents plan tasks, act across applications, and handle complex mobile workflows. The offering is a business-facing platform for device makers, not a claim that every smartphone already has the capability.
Alibaba Cloud announced Qwen Intelligence in its September 22 full-stack AI strategy release. The company said the offering gives phone makers a Qwen-powered platform for AI phones, designed to undertake complex tasks across multiple apps. Alibaba’s Alizila publication provides the more detailed product description, while an earlier SCMP report offers independent context for Alibaba’s work with HONOR on agentic devices.
This article is limited to Qwen Intelligence’s mobile-product lane. Alibaba Cloud says the platform gives phone makers access to Qwen-powered agents for complex cross-app tasks.
Qwen Intelligence for Smartphone Manufacturers
Alibaba describes Qwen Intelligence as a full-stack agentic solution optimized for smartphones. Alizila says it stretches from mobile-phone-optimized foundation models through a modular platform layer to ready-to-use agents for particular scenarios. Manufacturers can adopt the complete solution or selected modules, according to the company. Its product description identifies three first-phase agents: Mobile Planner Agent, Mobile-Use Agent, and Mobile Creative Agent. The company describes them as separate components of the platform rather than as evidence that every device already performs each function.
Alizila identifies HONOR as the first partner to adopt Qwen Intelligence and says the two companies will co-develop domain-specific models and solutions. The publication names HONOR’s Magic9 Series and HONOR Robot Phone as the first devices incorporating the platform. It said the Magic9 Series was set to launch on September 28, a future date relative to the September 22 announcement. That is not confirmation that the device had already launched.
Alibaba and HONOR had already been moving toward deeper AI-device collaboration. SCMP reported in July that the companies were preparing to expand their tie-up around an operating system for AI-powered devices and that new agent capabilities were expected at the World Artificial Intelligence Conference. The July report is useful context, but it does not independently verify the detailed September product specifications.
Three Agents Describe the Intended Interaction Model
Alizila describes three ready-to-use agents in Qwen Intelligence’s first phase: Mobile Planner Agent, Mobile-Use Agent, and Mobile Creative Agent. Each points to a different part of the user experience.
The Mobile Planner Agent is described as handling planning, task decomposition, tool orchestration, and dynamic adjustment. Alizila gives a business-trip example: the agent could break a request into flight booking, hotel reservation, calendar events, and itinerary planning, then respond to a canceled flight by offering alternatives. This is a vendor example, not an independently tested account of how reliably the system performs in the real world.
The Mobile-Use Agent is the most consequential piece because it is intended to act across apps. Alizila says it uses an API-first approach with graphical-user-interface fallback, alongside security boundaries and privacy protection. The sources describe the API-first and GUI-fallback approach but do not establish availability across every application or device.
The Mobile Creative Agent is described as a lightweight smartphone-optimized model for image generation and editing. It rounds out the platform by addressing creative tasks rather than navigation or planning, but Alibaba’s announcement does not provide independent quality comparisons or a detailed release timetable for every module.
Benchmarks and Task Claims Need Attribution
Alizila says Qwen Intelligence includes an open evaluation system with four benchmarks. It names MobilePA-Bench, which the company says covers more than 1,000 real-scenario tasks and more than 200 commonly used mobile tools, alongside tests for long-horizon cross-app work in cloud-phone environments and tasks on real devices.
The company also reports a claimed HONOR AI-phone task accuracy of up to 91.8% and says complex flows can exceed 100 steps. Those are Alibaba and Alizila claims, not independently validated rankings.
Alibaba’s announcement sets out the company’s intended mobile-agent functions: planning, cross-app use, and creative work. Alizila says its evaluation system contains four benchmarks and that MobilePA-Bench covers more than 1,000 real-scenario tasks and more than 200 commonly used mobile tools. Those are Alibaba-reported benchmark descriptions. The source material describes the platform, its named agents, its stated evaluation system, and its announced HONOR relationship. It does not establish the availability of every integration or independently validate the performance claims.
The company’s release makes additional platform claims that remain subject to the same attribution. Alizila says the Mobile Planner Agent handles planning, task decomposition, tool orchestration, and dynamic adjustment. It describes a business-trip example involving flight booking, a hotel reservation, calendar events, and itinerary planning. It also says the Mobile-Use Agent can use an API-first approach with a graphical-interface fallback. The source describes the intended functions; it does not prove availability across every app or device.
For historical reader context, EastFrontier has covered Alibaba’s Qwen-UI-Agent screen-control work and HONOR’s MagicOS 11 AI-phone strategy. Those reports do not independently verify the new Qwen Intelligence specifications. The September product materials establish the named platform, planned agent functions, benchmark descriptions, and the reported HONOR collaboration. They do not establish results beyond the company’s stated claims.