拍照和翻译救不了AI眼镜？千问2218元起推新硬件，要抢现实世界里..
Can’t photo and translation save AI glasses? Starting at ¥2,218, Qiwen launches new hardware to seize the real world..
Content Summary
Author | Qinghe Editor | April Domestic smart glasses are facing an awkward situation: camera resolution keeps increasing, translation supports dozens of languages, AI functions are getting more packed, but once the novelty wears off, they still can’t escape being left in the corner
Domestic smart glasses are facing an awkward situation: camera resolution keeps increasing, translation languages cover dozens, AI features are getting more and more packed,but once the novelty wears off, they still cannot escape the fate of gathering dust in a corner.
IDC’s latest data already shows some signals: in Q2 2026, global smart‑glass shipments grew 35.3% year over year,while the Chinese market saw its first 8% year‑over‑year decline. Among them,the audio and audio‑recording glasses category fell 35.4%. Functional homogeneity and insufficient product iteration have pushed this category to hit its ceiling.
This does not mean that AI glasses are a false proposition, but it presents a basic fact the industry can no longer ignore:Photo taking and translation are enough to attract users to try them, but they cannot support the essential demand for daily wear, and cannot boost the activity of smart glasses.
On October 13 in Hangzhou, Alibaba’s Qianwen introduced a new round of attempts.
On the same day, Qianwen officially launched the new generation AI glasses N1, N1 Pro, and the first AI ear‑clip headphones co‑developed with Bose. After applying the national subsidy and limited‑time discounts,The N1 starts at ¥2,218, the N1 Pro from ¥2,600, and the headphones are priced as low as ¥1,099.
The moment you start using the N1,the first thing you notice is its long‑awaited lightness. Compared with the previous model that featured dual‑eye displays and a built‑in micro‑optical engine S1, the new N1 series completely removes the display,returning the primary interaction method to voice control.
（Qianwen N1 Pro captures objects in front of it and returns recognition results on the phone, accurately identifying the exact lipstick shade number）
This allows the N1 Pro, even with multiple components such as a 50‑megapixel sensor, eye‑tracking module and vital‑sign sensor, to keep the weight of the device without lenses at roughly 44 g. The most painful aspects of wearing smart glasses in the past—“pressing the nose bridge, leaving red marks, wanting to take them off after just ten minutes”—are greatly alleviated.
Qianwen defines this product roadmap as Personal Agent hardware, with the logic that: today’s large models are becoming increasingly adept at handling complex tasks, but in the real world, they are still an “outsider” who knows nothing about the person in front of them. Therefore, beyond upgrades to imaging and wearing experience, this time Qianwen also tries to give AI a long‑missing capability: understanding the user’s immediate context in the real world.
To this end, Qianwen introduces a continuous visual perception link that consumes only 5 mW, and through eye‑tracking, biometric monitoring, and headset voice interaction, lets the Personal Agent move from a mobile app further into daily life.
In recent years, AI glasses have generally adopted a “click‑to‑wake” interaction model,essentially as a compromise to the physical limits of the battery. If high‑resolution CMOS sensors and multimodal vision models were to run continuously on‑device, the entire pair of glasses would not only drain the battery completely within half an hour,but also become a scorching “hand‑warmer” perched on the bridge of the nose.
The engineering solution offered by Qianwen avoids the hard resistance of large chips and opens acompletely independent low‑power always‑on perception link.
The underlying logic of this system is not to run a complex video model continuously on‑device, but to balance power consumption through extremely stringent compute‑level tiering:
- 5 mW of continuous perception power consumption
: Using the first‑generation Snapdragon AR1, Hengxuan BES2800, and Qijing HX6538 to form a heterogeneous computing platform, where the low‑power perception chip stays on standby, capturing only minimal grayscale frame outlines, the power consumption of the entire continuous pathway is kept at around 5 mW;
- On‑demand tiered wake‑up
: The 5 mW link only monitors characteristic changes in the physical environment. Only when it detects user‑preset or authorized key scenario cues (e.g., entering a meeting room, reaching a station turnstile) does it hierarchically wake the high‑performance chip responsible for communication, audio, and even high‑load image processing.
It is worth emphasizing that low‑power sensing does not equate to “omnipresent spying and 24‑hour recording.”
At the demo booth, staff specifically explained the privacy boundaries of this system: the Always On mode must be manually enabled by the user; when it is continuously on, only ** pixel‑level grayscale frames are captured; the perception stage extracts only the necessary scene features, performs basic judgments on‑device, and does not upload ** by default.
As for the “AI Inspiration Album” that drew attention at the launch event, a daily Vlog feature that can automatically capture highlight fragments of everyday life and transform them into a comic‑style format, the tech team also admitted that currently the team is prioritizing polishing this high‑tolerance, emotion‑value‑oriented life‑recording scenario.
Enabling AI to achieve flawless, continuous “full‑scene documentary understanding” under a few milliwatts of power is still a challenge; the industry remains in a steep climb phase regarding the generalization of algorithms for long‑tail scenarios.
Being able to sense the environment only solves half the problem. The other half is whether users are willing to wear it continuously, and whether the experience is natural enough when AI attempts to provide services.
In the on‑site experience, the most immediate change is the N1 series has fully moved to a screen‑less design. Compared with the previous S1, which featured a dual‑eye display module, the new generation finally returns to the restrained form of ordinary glasses. For users who have worn smart glasses,the pressure on the nose bridge often determines a product’s fate more than any technical specification.
N1 Pro squeezes the temple width down to 8 mm and, through a front‑back 1:1 weight balance, keeps the weight of the whole device without lenses at about 44 g. Combined with hot‑swap temple batteries,it achieves a total battery life of 8 to 9 hours. These engineering trade‑offs are less about hitting specs and more about using subtraction to clear the most basic physiological barriers for “all‑day wear”.
This subtraction brings a direct change: it greatly reduces the interruption of recording everyday life. As travel blogger Fang Qi shared on site, when using a camera or phone in the past, the first reaction to light was to find a shooting spot; stepping into an alley would trigger script ideas, and the person became disconnected from reality behind the viewfinder. The core value of first‑person recording has never been about competing with phone camera image quality,but about allowing people to stay immersed in the scene in front of them.
However, behind the seemingly silky experience there are also real challenges. In on-site testing,Compared with the straightforwardness of physical button photography, interactions that rely on voice wake‑up and edge‑cloud multimodal understanding often come with a one‑ or two‑second delay and are heavily dependent on the on‑site network. This also confirms the real threshold for current AI hardware: it’s not about packing in more features,but about making the perception‑to‑understanding‑to‑execution pipeline fast enough that computation becomes completely invisible to the user’s consciousness.
If the N1 glasses are responsible for capturing visual context, then Qianwen’s first AI ear‑clip headphones extend their reach to the auditory and action layers, tuned by Bose, preserving low‑frequency response and clarity in an open‑form factor, the true embodiment of the Personal Agent attribute is the “asynchronous task delegation” capability.
The reason previous voice assistants quickly became mere ornaments is that they trapped interaction in a mechanical “question‑answer, waiting in place” mode. Qianwen tries to change this logic: during the evening rush‑hour commute, the wearer doesn’t need to stare at a screen, just casually tell the earphones “help me follow up on today’s Qianwen launch event, gather industry viewpoints, make a brief and send it to me,”and the large model will autonomously run the long chain of retrieval, reading, and organization in the background.
After the task is completed, the earphones give a soft cue at key moments to indicate the task’s status. The earphones are no longer just a simple playback device; they become a lightweight, on‑the‑go workstation.
“The sense of exclusivity isn’t that AI needs to know more; it’s that it needs to know what relates to you.” This remark by Qianwen hardware lead Wu Jianjun on site exposed the core of Personal Agent hardware and directly hits the common pain point of today’s large‑model vendors.
If you are always trapped behind a phone screen and a PC browser, even a massive model with an enormous number of parameters only captures data that is reported, second‑hand. It may know that you bought a jacket on an e‑commerce platform, know that you edited the wording in an input box three times, but it won’t know that you lingered for three seconds at a storefront in an alley, and it certainly won’t know the fluctuations of your heart rate and body temperature under high‑pressure work.
Without the native senses of the physical world, a large model can never piece together a truly complete picture of intent; it can only become a passive answering machine waiting for queries in a data center.
At this point, asserting what the ideal “physical body” of a large model should look like in the real world may be premature. In the on‑site experience, the reliance on network stability, the coarse feel of on‑device model parsing in complex real‑time scenarios, and the fact that advanced biometric monitoring and iris authentication are locked behind the Pro version due to cost constraints, all indicate that hardware evolution is still in its early, hard‑fought stage.
But the exploration toward the real world is timely. The answer delivered by Qianwen no longer tries to force a screen that blocks reality in front of you, nor does it trap the product in the narrow narrative of a handheld camera and a translation pen. By stripping away redundant excess, weighing only 44.8 g, and tolerating just 5 mW at the hardware level, it steps the hardware back and pushes the sensory experience forward.
Disclaimer: This article is original content from AI Frontline, does not represent the platform’s views, does not constitute investment advice, and unauthorized reproduction is prohibited.
Conference Recommendations
QCon Global Software Development Conference·2026 (Shanghai) will be held on October 22–24. This edition focuses on engineering practices in the Harness AI era, from “building AI” to “harnessing AI”, covering hot technical directions such as AI Native architecture, Agent Runtime, AI Infra, Agent security and observability, Loop Engineering, embodied intelligence and world models. We invite the global tech community and frontline industry practitioners to share the most valuable engineering experiences of the AI Native era. For more details, scan the QR code or contact ticket manager 18514549229 for inquiries.
Are you also “watching”? 👇
Get MirrorSpeed