Recently, StepZen officially released the Step Edge Series Terminal Models for mobile phones, in-vehicle devices, and other terminal equipment, including four versions: Basic, Audio, GUI, and Gen. This series of models can process text, images, and audio locally, supporting functions such as screen understanding, speech recognition, interface operations, and image generation.
According to the official introduction, the Step Edge series terminal tools have a call delay as low as 0.1 seconds. In simple, frequent, or weak network environments, tasks can be completed entirely locally; complex reasoning tasks are handled by the cloud. This hybrid architecture significantly reduces the device's reliance on networks and cloud computing, and sensitive data does not need to be uploaded, greatly enhancing privacy and security.
To further optimize terminal performance, LeapStep also released its self-developed Step Inference NPU inference engine, aiming to significantly reduce the inference latency of terminal chips.
The official stated that the Step Edge series models rank first in 29 core evaluation metrics, demonstrating leading multi-modal capabilities on the terminal side.
This move marks a further acceleration of the AI terminalization process, and is expected to bring more real-time and private user experiences to smart devices such as smartphones and cars.
