AI News Feed
Market watch
Products & Applications

Nubia NaviX Ultra Ships With Doubao AI Agent, Claims 100 Million Yuan in First Second

Nubia's NaviX Ultra, the first retail flagship carrying the consumer version of Doubao's phone assistant, went on sale Sept. 16 at prices starting at 5,999 yuan. Nubia said first-second sales topped 100 million yuan, as phone makers and model companies bet that AI agents, not foldable screens, become the next battleground.

The 12GB+512GB version is priced at 5,999 yuan, or 5,499 yuan with the national subsidy applied; the 16GB+512GB version costs 6,499 yuan and the 16GB+1TB version 7,499 yuan. Buyers who order and pay within seven days of the first sale can get up to 24 interest-free installments.

Doubao supplies the model understanding and agent functions, while nubia builds the device and hardware. The company has framed the shift as one from a voice assistant that acts like a remote control — "open the camera" triggers a single action — to an agent that handles a chain of them, such as tidying up a meeting recording, generating a document, adding a review session to the calendar and sending the file to attendees. That requires the system to decide what to do first, which tools to call, and how to react when a page changes, a login prompt appears or a step fails.

The hardware is organized around keeping the agent working rather than around a single benchmark. The NaviX Ultra carries a 7,100mAh battery, communication enhancements intended to keep agent interactions stable, a 200-megapixel sensor that can also serve as a visual input, and a dedicated AI button with fingerprint recognition that combines wake-up and identity verification. The phone supports voice, close-range speech and AI earbuds, with continuous conversation, natural interruption and dialect recognition aimed at letting users speak normally rather than learn a command syntax. Memory is handled as a separate layer: the assistant can retain information the user points at on screen or states directly, so it does not ask the same question twice.

Permissions are the harder constraint. Under nubia's design, high-risk actions — payments, deleting files, sending messages, reading information — require user confirmation or outright takeover, drawing a line between tasks such as compiling meeting minutes, which can run automatically, and distributing them, which may not. Evaluation of such products therefore depends less on how many tasks an agent can attempt than on task success rates, the number of manual takeovers and how long users must watch. Leiphone argued that an agent taking 90 seconds while the user does something else can still beat a 60-second manual process, while an agent that stalls every few steps turns the user into a supervisor rather than saving attention.

Nubia is not alone in pursuing that ground. OpenAI paid about $6.5 billion for io, the hardware team co-founded by Jony Ive, to build its own AI device; Google has pushed agent capabilities inside Android and, through AppFunctions, lets agents call specific in-app features. In China, Doubao has chosen to work with phone makers while StepFun has built its own AI terminal system. On the display side, Huawei continues to develop tri-fold phones and Apple has released its own foldable, but Leiphone's reading is that foldables change the shape of the phone while agents change how people operate it.

The interaction model itself is shifting. Rather than opening an app, finding a page and filling in fields, a user can state an outcome — for example, buying a ticket to Shanghai tomorrow afternoon — and let the agent locate the relevant service, complete the search and form entry, and hand back the parts that need a decision or confirmation. Tickets still come from airlines and travel platforms, food delivery still comes from restaurants and their apps, and content still runs on the applications that carry it. What changes is who performs part of the searching, comparing, filling and page operation.

That raises a new competitive question, according to Leiphone: when a user has something to get done, which assistant gets the task first, since whoever understands the request first is better placed to organize the steps that follow. The outlet added that agents are expected to reduce operations rather than make decisions, with final choices left to the user. For nubia, the test is whether users grow accustomed to handing a share of their phone tasks to Doubao's assistant, across network fluctuations, app redesigns and varied usage scenarios that an engineering prototype never had to face.