Ericsson, China Mobile, and Oppo completed China Mobile’s first consumer-oriented user equipment (UE)-level slicing and application-level slicing test on a live 5G standalone (SA) commercial network. According to Ericsson, the trial validated network slicing combined with AI-powered service recognition at the UE level to deliver differentiated service experiences. The field verification, which was carried out in Dezhou, Shandong province, covered uplink livestreaming, short-video download and upload, usage of AI-powered glasses, and mobile gaming. Ericsson said the test achieved fine-grained resource allocation and isolation at both user and application levels, with validated results for typical services in guaranteed throughput and latency optimization on the existing 5G network. The work builds on existing network capabilities by extending UE-level slicing to specific users and terminals, while application-level slicing uses UE route selection policy (URSP) to support an end-to-end process; a complete technical validation chain; and the ability to host applications on slices for enhanced service capabilities in China Mobile’s 5G SA network. Under the hood Within the livestreaming and short-video scenario, the application-level slicing test created a closed-loop process around stutter detection and slice switching. When users experienced buffering while watching short videos or during uplink livestreaming, the Oppo device could inform the user and move traffic to a dedicated slice if the user chose to boost connectivity, while radio resource partitioning prioritized network resources for those applications. In congested network conditions, uplink livestreaming and short-video services reportedly maintained stable uplink and downlink speeds, with performance improving by more than two-times versus standard network performance. For gaming, Ericsson said user-perceived latency fell by at least 30% compared with conventional networks, which it framed as addressing major pain points for video and gaming experiences under heavy network load. The AI-glasses test, meanwhile, focused on device-level slicing for image recognition workflows, where captured images are