StepFun Launches STEPX Neo, a Model-Native Agent Phone, at RMB 3,999

Launch highlights: a RMB 3,999 "pioneer edition"

On the evening of October 13, StepFun's device arm launched STEPX Neo, billed as the industry's first "foundation-model-native agent phone", priced at RMB 3,999 with open sales starting October 16. According to ifeng, MyDrivers and Securities Times, it is the first phone in China designed and led by a foundation-model company, with the agent as the core interaction paradigm.

What "model-native" actually means

Unlike mainstream "AI phones" that bolt an assistant onto an existing OS, InfoQ summarized STEPX Neo's approach as "AI first at unlock, no more hunting for apps". Caixin likewise asked what truly separates it from other agent phones. The answer lies in the interaction layer: StepFun's Step-series multimodal models sit in front of the system, so users are greeted by an agent that understands intent and orchestrates tasks across apps, rather than a grid of icons. Wallstreetcn noted that nearly 100 ecosystem partners — including WorkBuddy and Alipay — have built scenarios covering collaboration, payment and daily services.

Why the full-stack pricing matters

Guangzhou Daily reported that StepFun pioneered a "software-hardware-model" full-stack pricing model: model subscription, hardware and services are bundled into one price, with agent capabilities included in the RMB 3,999 tag rather than sold as a monthly add-on. In effect, the MaaS business logic of the LLM industry is compressed into the price tag of a single device.

Notably, third-party tracking in the week before launch showed StepFun's Step 5 Preview entering the global top four by weekly model invocation volume, with China's weekly LLM API traffic up roughly 51% — software momentum fueling a hardware gamble.

The NineZenith view

A model company moving into hardware is essentially pulling agent delivery from cloud APIs to the device edge: owning the entry point closes the loop between task orchestration, multimodal perception and real-time response. NineZenith's practice with the Tianxing agent-orchestration platform points the same way — cross-system task coordination is where agent value is realized. An edge-cloud agent architecture is likely to become the shared standard for both consumer devices and industry deployments.

(Compiled from public reports by ifeng, MyDrivers, InfoQ, Wallstreetcn, Guangzhou Daily, Securities Times, etc.)