[Step Edge Released: On-Device Agent Tool Latency as Low as 0.1 Seconds] According to monitoring by Beating, StepStar has launched the on-device model family Step Edge, which includes foundational models, Audio, GUI, and Gen, targeting devices such as smartphones and automobiles. It can process text, images, and voice locally, enabling tasks like screen understanding, speech recognition, interface operations, and image generation. The on-device tool invocation latency is as low as 0.1 seconds. Simple, high-frequency, or weak-network tasks can be completed locally, while complex inference is delegated to the cloud. Sensitive data does not need to be uploaded, reducing reliance on network and cloud computing power. StepStar has also introduced its self-developed Step Inference NPU inference engine, designed to reduce latency for terminal chips. According to the official statement, the Step Edge series achieved first place in 29 core evaluation metrics. [Original Link]