Woofun AI reports that LeapStar has unveiled Step Edge, an on-device model family encompassing Audio, GUI, and Gen base models designed for mobile phones, automobiles, and other terminals. The system enables local processing of text, images, and speech to facilitate screen understanding, speech recognition, interface operation, and image generation, with tool invocation latency reduced to a minimum of 0.1 seconds.

Simple or low-network tasks are executed locally, while complex inference is offloaded to the cloud, ensuring sensitive data remains off-network and reducing dependency on cloud computing resources. LeapStar also introduced its self-developed Step Inference NPU Engine to lower terminal chip latency, claiming the Step Edge series leads in 29 core evaluation metrics.