StepFun Showcases Terminal Agent Route at WAIC

APPSO · wechat · 2026-07-18

StepFun Showcases "Model + Terminal" Route at WAIC

StepFun brought phones, cars, and robots to its WAIC booth. The main focus wasn't a single model capability, but the integrated implementation of "Large Model + StepAOS + Terminal": deploying a unified agent logic across phones, PCs, car infotainment systems, and robots.

Key Highlights

Model Matrix: The Step series is divided into Pro, Flash, and Edge tiers, covering reasoning, Agent workflows, on-device perception, and instant response. It also spans text, multimodal, voice, image, video, and music capabilities.
On-Device Capabilities: The StepEdge on-device model ranked first in its class across 29 benchmarks, emphasizing low latency and privacy for local processing, while offloading complex tasks to the cloud.
Agent Phone STEPXNeo: Features "result-based interaction," where users simply state a goal, and the Amoo agent handles cross-app planning, scheduling, and execution at the system level, secured by four safety gates: "trusted, visible, controllable, and reversible."
Cross-Device Collaboration: Tasks initiated on a phone can be seamlessly handed off to a PC, demonstrating a "task portability" capability that goes beyond standard multi-device syncing.
In-Car Agent: Partnering with Geely and Qianli Tech, they demonstrated linking the cabin, assisted driving, and lifestyle services into a task chain on the Zeekr 8X.
Embodied AI: Six robots collaborated to assemble a 3-meter Lego Great Wall, highlighting multi-robot division of labor, sub-millimeter precision operations, and an architecture where the "brain handles planning, and the cerebellum handles motion."

Conclusion

StepFun aims to prove that a single "thinking brain" can be installed into different physical forms—phones, cars, robots—continuously completing tasks across various terminals.

Original post →

StepFun Showcases Terminal Agent Route at WAIC

More from Companies & People