StepFun Unveils the StepX Neo, Its First Agentic Smartphone

StepFun unveiled the StepX Neo on Monday in Shanghai, calling it the world’s first mass-market agentic smartphone, one built around an AI agent instead of a chatbot bolted onto Android. The Chinese startup, founded in 2023 by former Microsoft executives, says its Step Amoo assistant can book a flight, translate 32 languages and fill out customs paperwork without the user ever opening a separate app.

It isn’t the first attempt at this. ZTE tried something almost identical with ByteDance seven months earlier, and that agent went just as deep into the operating system, sold out immediately, then triggered a privacy backlash that forced ByteDance to dial back what the assistant could touch. StepX Neo is walking into the same test, with ZTE backing both experiments.

A Phone That Books the Trip Without Being Asked Twice

At the center of the phone is Step Amoo, an assistant StepFun says lives inside the operating system rather than running as a separate download. In its own demo, the company gave the agent a single instruction, plan a trip, and watched it search flights, compare accommodation, complete bookings, organize the details and save the itinerary without the user switching between apps. StepFun says Step Amoo also learns a user’s preferences over time and applies them to future requests.

For travelers, the pitch stretches further. StepFun says the assistant translates live conversations, phone calls, text messages and text captured through the camera viewfinder across 32 languages and regional dialects. One widely shared post on X described the unveiling as marking the beginning of the agentic AI smartphone era, language StepFun’s own event echoed almost exactly.

Beyond translation, StepFun says the assistant is built to:

• Pull up local transit schedules and recommend nearby attractions in an unfamiliar city
• Reach a saved itinerary offline, even without a cell signal
• Flag visa requirements and help complete customs paperwork ahead of a flight
• Send reminders for flight check-in windows based on saved travel details

None of it works without somewhere to plug in, which is why StepFun lined up integrations with Ctrip, Alipay, Didi, Meituan, WPS Office and CapCut spanning travel booking, payments, ride hailing, office work and content editing, months before the phone even has a price tag.

Step AOS Rebuilds the Software Stack Around the Agent

StepFun insists Step AOS, short for Step Agentic-native OS, is not another Android skin with a chatbot stacked on top. The company says it rebuilt significant portions of the stack using pieces of Android, Linux and RTOS, short for real-time operating system, so agents sit at the center of the phone rather than bolted onto it.

The result is what StepFun calls an atomic capability engine. It breaks phone functions into four groups, communication, apps, files and system tools, using the MCP standard, short for Model Context Protocol, so the agent can combine them freely to finish a task on its own. A separate layer called NUI, or Natural User Interface, blends voice and visual input while learning how one specific user behaves over time.

On-device processing runs through Step Edge, StepFun’s own foundation model built to fit inside a phone’s compute budget. The company says it leads comparable edge models across 29 benchmark tests, and that its two-domain memory system, which separates what belongs to the user from what belongs to the agent, retrieves information in as little as 15 milliseconds and tops rankings on LongMemEval. StepFun has not said which 29 benchmarks it used or which competing models it measured against.

ZTE Already Ran This Experiment Once

Seven months before StepX Neo, ZTE partnered with ByteDance on an agentic prototype built around the Doubao assistant. It launched on December 2 as the Nubia M153, sold out immediately and then just as quickly sparked privacy concerns that forced ByteDance to dial back what the assistant was allowed to do.

Yicai Global reported that the Doubao agent on the Nubia M153 used the Android system permission INJECT_EVENTS, typically reserved for system components, letting it simulate taps and read what appeared on screen. Users found that WeChat and some banking apps began throwing up warnings or blocking logins whenever Doubao’s functions were active. ByteDance said the capability was only used with explicit consent and that screen content was never stored in the cloud or used for training.

ZTE is not a bystander this time either. The same company already ships StepFun’s AI features in its own Nubia Z80 Ultra, a separate flagship that has not drawn the same scrutiny, and ZTE has since joined StepFun’s investor roster ahead of a planned stock listing. It is the one supply chain name that watched an agentic phone succeed and stumble in the same year, then backed a second one anyway.

Who Is Actually Bankrolling the Agentic Bet

StepFun did not build any of this on hype alone. The startup closed a Series B+ round worth roughly $718 million in January, backed by Tencent, Alibaba, Qiming Venture Partners, 5Y Capital and several state-owned funds. The same announcement brought in Yin Qi, previously chief executive of the Chinese AI firm Megvii, as chairman, joining founder and chief executive Jiang Daxin, a former Microsoft vice president.

Four months later, StepFun pulled in close to $2.5 billion in a pre-IPO round that drew in the phone industry’s own supply chain, ahead of a planned Hong Kong Stock Exchange listing. The bet fits a broader pattern already reshaping Chinese tech, a build out one estimate puts at $98 billion in AI infrastructure feeding exactly this kind of always-on assistant.

The round valued StepFun at roughly $10 billion, Bloomberg reported, and the company is targeting a Hong Kong listing worth about $500 million by the end of the year.

The Specs and Price StepFun Still Won’t Say

StepFun showed off a dual camera setup and an interactive secondary display on the back of the Neo, but the company has not confirmed RAM, storage, battery capacity or camera resolution. Pricing and a launch date are both still missing.

GSMArena, reviewing the unveiling, called it “mostly buzzwords with little substance” and questioned whether Step AOS is meaningfully different from an Android skin with an AI feature bolted on top.

• What We Know: Step Amoo runs inside Step AOS, translates 32 languages and dialects, and already has integration partners in Ctrip, Alipay, Didi, Meituan, WPS Office and CapCut.
• What We Know: Step Edge claims to lead rivals across 29 benchmark tests and recalls stored memory in as little as 15 milliseconds.

• What’s Unconfirmed: RAM, storage, battery capacity, camera resolution, retail price and a launch date.
• What’s Unconfirmed: Which 29 benchmarks StepFun used, and which competing models it measured Step Edge against.

StepFun has released exactly the numbers that make the Neo look good, and nothing else yet.

The Trust Problem Every Agentic Phone Shares

Agentic phones share one structural issue no benchmark table fixes: an agent needs system-level access to be useful, and system-level access is exactly what makes security researchers uneasy. Chinese and western experts have been sounding the alarm over what that access means for data privacy.

Agentic AI is threatening to break the blood-brain barrier between the application layer and the OS layer.

Meredith Whitaker, president of the Signal messaging app, made that warning at the SXSW conference, part of the same debate over how deeply an agent should be allowed into a phone’s operating system.

Researchers studying agentic systems more broadly warn that an agent’s ability to retain long-term behavioral data raises fresh data minimization and consent questions that existing privacy rules were not built to answer. Gartner separately projects that at least 15% of work decisions will be made autonomously by agentic AI by 2028, up from none in 2024.

The dependency question predates agentic design too. Separate research has already linked compulsive smartphone use to a higher depression risk among older adults, and an assistant built to anticipate a request before it’s made only deepens how much a phone gets trusted to decide.

StepFun still owes the market pricing, hardware specs and a launch date. The trust questions that met ByteDance’s phone in December are still waiting on an answer too.

Frequently Asked Questions

What Makes a Smartphone Agentic Instead of Just Having AI Features?

An agentic smartphone lets an AI agent complete multi-step tasks across apps on its own, rather than answering questions inside a chat window. Apple’s Apple Intelligence, Samsung’s Galaxy AI and Google’s Gemini all run as features layered on top of an existing operating system. StepFun’s pitch is that Step Amoo sits inside Step AOS itself, with system functions broken into pieces the agent can combine without a person tapping through each app one at a time.

How Does Step Amoo Personalize a Task Like Trip Planning?

Step Amoo is designed to remember choices a user made before and apply them automatically on the next request. StepFun says that includes small habits, like automatically prioritizing pet-friendly flights on a future search if that’s what a traveler booked previously, without being asked to repeat the preference each time.

Is StepFun Building the StepX Neo’s Hardware Itself?

Not entirely, based on reporting around the unveiling. GSMArena’s coverage noted StepFun appears to be sourcing the phone from a contract manufacturer rather than producing it fully in-house, which lines up with the investor list: two of the world’s largest phone ODMs, Huaqin Technology and Longcheer Technology, are already among StepFun’s backers.

Will the StepX Neo Launch Outside China?

There is no confirmation yet. Industry watchers expect a China debut sometime next year aimed at the premium segment, and coverage of the unveiling has noted the phone would likely face regulatory hurdles reaching the United States and European Union simply by virtue of being a Chinese-made device. Whether it reaches India or other Asian markets remains unresolved too.

Has StepFun Done Anything to Avoid ByteDance’s Privacy Backlash?

StepFun is pointing to outside validation rather than just its own promises. The company says the Neo has passed the L3 level of China’s national AI terminal intelligence grading standard, currently the highest tier open for testing, and that it co-published an agent security whitepaper with the Shanghai AI Laboratory. StepFun says it is the only phone to hold the L3 certificate so far, though neither claim has been verified outside the company’s own announcement.