Nubia NaviX Ultra Teases Core AI Features Ahead of September 16 Launch

On September 14, Nubia released a video demonstrating the core AI features of its upcoming NaviX Ultra, a device the report describes as the world's first AI agent smartphone. The official launch and first sale are set for September 16, just two days later, which makes the clip a key warm-up ahead of the formal debut.

The report says Nubia defines four capability dimensions for the phone: understanding what users say, getting work done, remembering context, and staying secure. Those labels map to voice understanding, task execution, memory and safety, and the video leads with the understanding dimension.

For voice interaction, the NaviX Ultra carries what the report calls the industry's only fingerprint AI button, which wakes the assistant with a single press and is presented as both fast and secure. In real-time call mode the phone connects to Seed's newest full-duplex speech model, letting users interrupt and hold continuous conversations rather than taking rigid turns.

Recognition performance is treated as a headline strength. According to the coverage, the Doubao phone assistant can follow instructions in noisy streets, accented speech and multi-person conversations, and the device already supports more than ten Chinese dialects including Hokkien, Hakka, Shanghainese and Cantonese, alongside multilingual interaction in English and Japanese.

Beyond voice, the video shows multimodal behaviour: screen-based Q&A built on a multimodal model, and natural interaction driven by live visual content during video calls, so the phone can both hear the user and see the surrounding situation.

The more interesting part is the packaging. Doubao's phone assistant appears here as a consumer edition embedded in Nubia hardware, a sign that large-model assistants are moving from standalone apps toward deeply device-bound products. That shifts competition from raw model benchmarks toward whether an agent can actually finish a task for its owner.

What to watch next is the September 16 launch and opening sale. How the remaining three dimensions, namely getting work done, remembering and security, are demonstrated on real hardware, and how the work is split between on-device and cloud processing, will decide whether the AI agent phone positioning holds up.