StepFun on July 13 introduced STEPX Neo, a smartphone built around a native AI agent, as it pushes beyond standalone apps and toward a system-level entry point for large-model services. The company’s pitch centers on a different way to handle cross-app actions: not simulated taps, but a protocol stack made up of the Step Edge on-device model and GUI-MCP.
Why StepFun moved into hardware
At the launch event, chairman Yin Qi said the company had asked many people in the device industry for advice and was repeatedly told not to get into hardware, but decided to build the phone anyway. That remark captured a broader pressure facing model companies: if they remain inside existing mobile operating systems, they still rely on platform owners for permissions and traffic distribution.
According to VCBeat, StepFun raised more than 5 billion yuan in a B+ round in January 2026, the largest single financing in the large-model sector over the previous 12 months. Founded in April 2023 and led by Jiang Daxin, the company has built its base in multimodal models. Even with fresh capital, it moved into a hardware business with a long supply chain, a sign of how hard cloud-side monetization remains and how valuable a device-side gateway has become.
In that context, building a phone is less about adding another app and more about trying to place the agent at the center of the operating system.
What is known about STEPX Neo
STEPX Neo runs Step AOS and includes the Amoo agent. Cailian Press reported that the phone is manufactured by Huaqin Technology, has a secondary interactive display on the back, and will make its first public appearance at the World Artificial Intelligence Conference on July 17. Detailed hardware specifications and pricing have not been disclosed.
Based on what has been announced so far, the secondary screen appears intended to keep agent status visible and offer quick interactions, reinforcing StepFun’s positioning of the device as agent-native hardware rather than a conventional smartphone with AI features layered on top.
The Doubao lesson
The launch also arrives after another route to cross-app control ran into resistance. In December 2025, ByteDance released the Doubao phone assistant. Media reports at the time said it debuted on the ZTE Nubia M153, priced at 3,499 yuan, with cross-app actions handled through system-level simulated clicks.
Users could issue a command and let the assistant open other apps, skip splash ads, and move directly to an information feed or a transaction page. The demos drew attention early on. The limits showed up quickly. In early December 2025, many users reported that Doubao triggered WeChat security controls during operation, causing abnormal app exits or login failures. The capability was later taken offline.
The source article framed that episode as more than a technical glitch. Super apps control traffic distribution and ad exposure, and simulated clicking bypasses those layers. On the technical side, accessibility-based click simulation can also be detected as abnormal behavior, and app operators can block it with interface changes or additional verification steps.
The article also cited legal experts who said forcing cross-app operations through accessibility services can create unfair competition risks, making an adversarial path hard to sustain.
GUI-MCP as a cooperative model
StepFun’s answer is GUI-MCP. Based on the company’s open-source documentation, the protocol uses a layered dual-stack architecture and supports a high-privacy mode in which raw screenshots stay on the device while only semantic summaries are sent to the cloud.
StepFun said its open-source GUI-specific model has 4 billion parameters and supports local recognition and operation across more than 200 apps. In this setup, the on-device model handles screenshot understanding and local execution, while the cloud model takes on higher-level planning and more complex reasoning.
That separates it from the add-on style used in simulated clicking. Instead of taking over a screen by force, the protocol route aims to let app developers expose capabilities through standardized interfaces. If that works, an agent can call core app functions directly rather than searching for buttons visually step by step.
Partners are in, but the larger ecosystem is unresolved
StepFun named Meituan, WPS, CapCut, Trip.com, Amap, Alipay, Baidu, Didi and JD.com as its first ecosystem partners. Those names cover food delivery, office software, travel, maps, payments and e-commerce, giving the company an initial base for multi-step tasks across services.
Still, the source article argues that the real barrier is not the protocol alone but how value is shared. AI agents aim to take users straight to task completion, reducing the role of super apps in traffic routing and display. Without a workable revenue-sharing structure, app operators may still respond with code changes or stronger verification flows.
The article points to another possible route as well: apps can package their own capabilities as Skills or MCP endpoints and open them to agents voluntarily. That lowers the level of confrontation, but only if developers believe the gains from a new entry point outweigh the loss of a closed moat. For platforms with stronger walled-garden characteristics, including Tencent’s ecosystem, there has been no clear official response on whether they would join such a standard.
Tool value will depend on real-world execution
For users, the central question is not raw specs. It is whether STEPX Neo can reliably complete multi-step tasks across apps. The source article used one example: a user could ask the agent to book a flight to Beijing for the next day and arrange a ride to the office, with the system calling Trip.com and Didi to complete both steps.
If that kind of flow works in practice, phone interaction changes in a meaningful way. But the gap between a launch-stage presentation and real-world reliability remains. STEPX Neo has only appeared at the announcement event so far, and there is no hands-on test data yet for complex scenarios involving pop-ups, CAPTCHAs or payment risk controls. The article says the device’s first public showing at WAIC will be an early test of whether GUI-MCP can hold up under those conditions.
Hardware is another open question. The source notes that Huaqin can support mass production, but StepFun does not have the offline channels or after-sales structure that established phone brands already built.
What STEPX Neo clearly shows is that model companies are trying to move from cloud services to end devices and from the app layer to the system layer. Whether StepFun can turn protocol design into real cross-app access will depend on the phone’s performance after its WAIC debut and on whether more major apps decide to connect to the interface standard.