Huawei Flagship Phones Reportedly Support Offline Local AI Models Up to 15GB
According to a tech leak, Huawei's upcoming flagship smartphones like the Pura X View and Mate XT 2 Ultimate Design will allow users to voluntarily download and deploy local AI models. Users can choose between a ~6GB multimodal enhanced model for media editing and a ~15GB multimodal Mixture of Experts (MoE) model for complex offline task management. Running large multimodal and MoE models locally on smartphones marks a major shift toward privacy-focused, offline-first mobile AI features. It highlights how mobile chipsets and operating systems like HarmonyOS are evolving to support intensive edge computing without relying on cloud servers. The 6GB multimodal enhanced model handles tasks like AI super-resolution camera enhancement, gallery photo editing, and offline voice synthesis, while the 15GB MoE model manages complex application control and offline photo organization. Recent updates to HarmonyOS 7 allow compatible devices to view installed local model versions, parameter sizes, and service scenarios directly.
## BACKGROUND
On-device AI involves deploying machine learning models directly on hardware like smartphones rather than routing queries through cloud data centers, offering faster response times and improved data privacy. A Mixture of Experts (MoE) architecture is an AI model design that selectively activates only specific sub-networks ("experts") for a given task, enabling large-scale capabilities while maintaining computational efficiency on edge devices.