Pogbagram

For many ordinary Chinese families, a car is not just for one person's needs, but for the whole family. Therefore, a large six-seater has become the focus of the market as a product with the greatest common denominator. On weekdays, it is used for commuting, picking up and dropping off children, and occasional shopping; on weekends, it becomes a mobile living room for the family, accommodating the elderly, children, camping equipment, and a short trip that can be taken at any time.

The reason why large six-seater SUVs have become increasingly popular in the past few years is not essentially because the cars have become bigger, but because family travel scenarios have become more complex.

On May 23, the Dongfeng Yipai M8 made its global debut in Wuhan. Officially defined as the "golden standard large six-seater for Chinese families," this new model is a key product for the Dongfeng Yipai brand's upward advancement, and it integrates resources from Dongfeng Motor, Huawei Qiankun, and CATL. Rather than simply emphasizing size, power, or features, the Yipai M8 aims to answer another question: where should the standard be set for a large six-seater SUV targeting young Chinese families?

From a product definition perspective, the key word for the Yipai M8 is "uncompromising." Dongfeng Yipai stated that in the early stages of product development, they conducted hundreds of user co-creation experiences and communicated with thousands of users, condensing the high-frequency car usage needs of young Chinese families into "gold standards" in several dimensions: appearance, comfort, safety, and intelligence.

This is also a more realistic assessment after the competition in the large six-seater SUV market entered a new stage: simply making the car bigger is no longer enough.

First, let's talk about the design. The Yipai M8's exterior doesn't take an overly aggressive approach, instead emphasizing a sense of elegance, stability, and technology. It combines Eastern aesthetics with modern technological language, resulting in a relatively full body posture, while the smart blue light on the roof further enhances the vehicle's intelligent identity. For a family SUV, design isn't just about attracting attention at first glance; more importantly, it's about maintaining a pleasing appearance, providing peace of mind, and offering a certain sense of occasion during repeated daily use.

What truly determines the family's user experience is the interior of the car.

The Yipai M8 boasts a length of 5 meters and a wheelbase exceeding 3 meters, with a wheelbase-to-length ratio of 60.3%. For users, these figures translate into a more direct experience: the second row is no longer just a "boss seat," and the third row is no longer just an emergency seat. The elderly can sit more comfortably, children won't feel cramped on long journeys, and even with a full load of passengers, everyone won't have to compromise on legroom.

In terms of configuration, the Yipai M8 offers features unique in its class, including dual first-class seats, dual-view large screens, dual-door dual-way intelligent cooling and heating compartments, and the largest 50-inch AR-HUD in its class for real-view navigation. In practical terms, this means it can provide a more comfortable resting position on long journeys, and also make in-car entertainment, storage, and navigation information more convenient. The cooling and heating compartments are no longer just a bonus for camping; they could also be a cold drink when picking up the kids from school in the summer, or a hot meal prepared for the family on a long trip.

Safety is the bottom line that cannot be ignored when using family cars.

The Yipai M8 boasts a comprehensive safety system covering driving and parking safety, active safety, electric powertrain safety, passive safety, health safety, and quality safety. The new vehicle will be equipped with full-link NCA assisted driving, the CAS 5.0 all-dimensional collision avoidance system, and features a cage-type high-strength body and a baby-friendly environmentally friendly cabin. Furthermore, the official warranty includes a lifetime guarantee for the entire vehicle and its three core electric components.

These features, if presented in a product launch presentation, would easily become a string of parameters; but in daily life, they correspond to a more concrete sense of security. For example, on long highway journeys, assisted driving can reduce driver fatigue; in urban areas, collision avoidance systems can cover more unexpected scenarios; and the environmentally friendly cabin and the safety of the three core electric components (battery, motor, and electronic control system) are related to the long-term experience of the elderly and children sitting in the car every day.

In terms of power, the Yipai M8 offers both pure electric and range-extended versions, which is a practical choice for family users.

The pure electric version features an 800V high-voltage platform and a 4C fast-charging architecture, with a peak motor power of 230kW and a CLTC range of 600km. For families with fixed charging facilities and who primarily use the vehicle between cities and intercity routes, the pure electric version offers lower operating costs and a quieter driving experience. The range-extended version boasts a CLTC-rated pure electric range of 300km, which, combined with the range-extending system, covers most daily commutes and alleviates charging anxiety during long-distance travel.

This dual-power strategy actually reflects the current reality for families: not everyone is ready to jump straight into a pure electric lifestyle, nor are all families willing to continue being limited by the operating costs and experience of gasoline vehicles. Promoting both pure electric and range-extended electric vehicles essentially preserves the choice for users.

Intelligent features are another key aspect of the Yipai M8, and their importance is comparable to its mechanical qualities.

The new car is equipped with Huawei's Qiankun technology and what the company calls the "Smart 6-Piece Suite," including the first batch of features such as Qiankun Intelligent Driving ADS 5 Pro, HarmonySpace 5.2 cockpit, Qiankun Vehicle Cloud, Qiankun Vehicle Control, Whale Fin Communication, and HUAWEI SOUND 21 speakers. For family users, the significance of intelligence lies in "the car driving itself a little," and in whether navigation, cockpit, voice, entertainment, communication, and sound experience can be truly integrated.

As a car carries passengers for longer periods, the cabin becomes more than just a transitional space from point A to point B; it becomes an integral part of family life. Children watch cartoons in the back, parents listen to music in the front, navigation information is projected onto an AR-HUD, and the refrigerator is stocked with drinks and fruit. These details combine to create what we call a "mobile home."

From this perspective, the debut of the Dongfeng Yipai M8 reflects a shift in standards in the large six-seater SUV market. In the past, the market focused more on whether the car was big enough, had enough features, and was attractive enough; now, users are starting to ask whether these features truly serve family life, whether the intelligent features are convenient enough, and whether the sense of security covers every journey.

Competition in the large six-seater SUV market will continue to intensify, but the truly viable product won't necessarily be the one with the most impressive specifications on the list. Instead, it will be the one that transforms the complex needs of a family into a more stable, comfortable, and reassuring experience. The Yipai M8 chooses to enter this market by focusing on space, comfort, safety, dual powertrains, and Huawei's Qiankun intelligent technology. Whether it can become a key model for Dongfeng Yipai's brand to break through to a higher level remains to be seen, pending price, delivery, and real-world user experience. But at least from the information released at its debut, this car has clearly stated its purpose: today's family SUVs are not just about selling six seats, but about providing a complete lifestyle solution for the whole family.

#Welcome to follow iFanr's official WeChat account: iFanr (WeChat ID: ifanr), where more exciting content will be presented to you as soon as possible.

In 2007, Steve Jobs used a 3.5-inch screen to fold human information interaction into a glowing two-dimensional plane.

Nearly twenty years have passed in the blink of an eye. The challengers of yesteryear have become the targets of today's revolution. The wave of AI hardware is attacking smartphones, attempting to shatter this glass barrier. New forms of hardware, represented by AiPin, have emerged one after another, rising and falling like tides. In contrast, traditional PCs are more stable, ushering in a second spring through agents.

Since reinventing the form is difficult, wouldn't it be a better option to put AI into mature devices that are already integrated into people's daily lives?

Recently, iFanr exclusively reported that Apple's AirPods project with a camera had actually been delayed. In this lull between giants, a Chinese startup—Guangfan Technology—founded by an early team at Xiaomi and comprised of employees from giants like ByteDance, Alibaba, Huawei, and Tencent, has preemptively delivered: launching the industry's first fully-feeling wearable device with a camera, powered by native AIOS—presented in the form of headphones.

Last year, we reported on the launch of the LightSail headphones. Last week, the product officially went on sale, and we raised a question: why does a pair of headphones need to have a camera attached?

After 72 hours of in-depth testing, iFanr is convinced that "headphones with cameras" represent a direction for future AI hardware, but there is still a long way to go before reaching the finish line.

One watch, one box, one earphone—all serving AI.

When we talk about the LightSail AI All-Sense Wearable Device, we are actually talking about a large and well-defined hardware suite.

Lay it out on the table, it consists of an earphone case, a pair of ear-hook headphones, and a smartwatch. In this system, each component plays a specific role, and the flow of AI data is carried out collaboratively among them.

At the heart of all three is the earphone that hangs on your ears.

Guangfan has kept the weight of each earphone to under 11 grams, and the black and white "panda" color scheme effectively reduces the sense of bulkiness, while the C-shaped bridge ear hooks extend naturally backward along the ear canal.

In comparison, a single AirPods 4 weighs 4.3 grams, while a single Ola Friend earphone with an ear hook design, also launched by Doubao, weighs 6.6 grams.

The reason the LightSail AI Earphones are slightly heavier is because they have an 88° wide-angle, 2-megapixel camera embedded in the front, allowing the user to look directly at the world while wearing them. To alleviate privacy concerns, they do not offer users any regular photo or video recording functions; all images captured by the camera are sent to the cloud for AI interaction.

Overall, Guangfan chose the OWS open-back headphone design, with metallic speakers floating on the outside of the tragus. Combined with the battery at the end of the ear hook, the body cleverly achieves a 5:5 front-to-back weight distribution, so there is no fatigue even after wearing it for a long time.

When used with the charging case, these AI earbuds offer up to 90 hours of battery life. Beneath the skin-friendly shell, Guangfan has packed in a dedicated eSIM module and dual-band GPS. Commands captured by the earbud microphones and images captured by the camera are transmitted back here, then packaged and sent to the cloud via a separate network for analysis by a large data model.

Finally, there's the smartwatch with a 1.97-inch AMOLED screen. When auditory information is overloaded, it takes over and displays the key data that needs to be scanned for confirmation, completing the final link in the "watch-ear linkage".

The three components each serve their own purpose: the headphones take center stage, providing visual and auditory input and serving as the primary interactive device; the case is tucked into your pocket, acting as the central hub for computing power and networking; and the screen on your wrist serves as a supplementary interactive element, providing a home for frequently accessed information.

I'm definitely looking at my phone less now, but the interaction could be better.

According to the official specifications, this device is characterized by "full perception, all-weather, and proactive AI".

Based on several days of heavy use, I have summarized my experience with its core functions into three main categories:

Category 1: More proactive and intelligent schedule managers

In the mobile internet era, the most important information is often scattered across fragmented conversations on apps like WeChat and Lark. Guangfan's approach is to use AI to act as an information funnel.

When WeChat, Lark, or DingTalk receive a flood of messages, the AI ​​automatically filters out irrelevant chatter and focuses on broadcasting high-priority information. The accompanying Hi Light software supports personalized settings for the frequency of summary broadcasts for each software, offering three options: real-time broadcast, quick summary, and concise summary. In my personal experience, the quick summary is the best choice that combines timeliness and efficiency.

Based on intelligent message notifications from three software programs, when specific times, locations, and events appear in chat content, the AI ​​will proactively pop up a message asking:

This message contains a meeting schedule. Would you like me to add it to your calendar?

On the day of the meeting, the AI ​​will act as a dispatcher. When the system recognizes that I have a topic selection meeting at 8:30 in the morning, the AI ​​in my earphones will remind me at 8:00 that I live not far from the company and that I can leave home, based on the current real-time traffic conditions and commute distance.

After adopting the Guangfan voice broadcast system, I have indeed picked up my phone less often to check messages and add memos. However, the current limitations of this function are also obvious. To reply to Lark via voice, you have to authorize the cloud system separately in the app. To hail a ride, you need to have already linked your Didi account or logged into the Shouqi platform, which only covers some cities, using your Guangfan account.

However, this is a last resort. Last year, Doubao Mobile Assistant proved that breaking down the barriers of super apps is incredibly difficult. Guangfan's access to services through the cloud system is more like trying to dig a tunnel under a high wall.

Although there are occasional risks of interface changes or license expiration, at least the possibility of cross-application scheduling has been touched upon.

Category 2: Slightly dulled visual perception

The biggest selling point of Guangfan is undoubtedly its camera, which can cover the area of ​​everyday line of sight. To summarize Guangfan's expectations for it in one sentence: to bring the action of acquiring information forward.

While shopping on the weekend, I saw a restaurant, glanced at the sign, asked a question, and it told me the ratings and queue status. When I spotted the black cat clock on my colleague's desk, the AI ​​earphones used visual recognition to add it directly to my shopping cart.

Of course, the prerequisite is similar to taking a taxi—I need to log in to my JD.com account in Hi Light first.

But this part of the functionality is actually the part where I felt the biggest disappointment during my experience.

Human vision has always been instantaneous and continuous. However, the visual perception of the LightSail AI Headphones is point-like and delayed.

When a store catches my eye, I double-tap to wake up my headphones. What follows is a long period of waiting. You need to stand still for 8 to 10 seconds before you hear a simulated shutter sound; then, there's another 10 seconds of data transmission and recognition before the cloud-based processing center provides feedback on whether to add the item to your cart.

After careful consideration, I believe the problem mainly lies in two aspects: from a psychological perspective, if the camera is used for a long time, it will create pressure for the user and those around them to feel that their lives are being spied on; from a product perspective, the weak battery and compact body of the earphone cannot support the sensor to work for a long time.

The two constraints are unlikely to be overcome in the short term, but solutions are not impossible.

Currently, the earphone case, which serves as the data hub, supports 4G eSIM, but the uplink and downlink speeds of data transmission are severely limited. Furthermore, based on the current performance, the LightSail AI Earphone should follow the concept of "continuous computing experience," where the lightweight device handles millisecond-level basic data capture and dialogue, while the computationally intensive visual analysis is seamlessly handed over to the cloud.

If, with further optimization of the local model, more and more simple interactions can be processed directly locally without having to go through the cloud, the feedback speed may be able to reach a new level.

Of course, it would be much better if we used more energy-efficient wearable chips and switched to 5G eSIM.

Category 3: Conditionally triggered and usable AI memos

Compared to complex visual recognition, conditional triggering based on location and physiological data offers a significantly more impressive user experience.

If I casually remind my earphones in advance that I need to buy shampoo, dish soap, and laundry detergent next time I go to the supermarket, then when the Guangfan AI earphones recognize my location and pass by or enter the supermarket, they will proactively provide the reminder information to avoid forgetting anything.

This type of triggering logic also applies to physiological monitoring—I first set a reasonable heart rate on the watch based on my physical condition. Once the watch detects that my current vital signs exceed the threshold, it will remind me through the earphone to slow down or take a break.

In actual use, these two functions are stable and restrained, with timely and accurate prompts, making them the most practical highlights in my opinion.

However, aside from its functionality, there are still some minor flaws—

The LightSail AI Earphones rely entirely on the charging case as their computing and transmission hub. If you take the earphones out but leave the charging case at home or your workstation, the earphones will only go silent after the prompt sound, or prompt you to put them back in the case and try again. To use the AI ​​services, users must carry the charging case with them at all times. During the few days I've been using the product, I've kind of "adapted" to it.

This creates a difference from the mainstream user experience, where people are used to wearing AirPods all day long and even briefly leaving the case unattended.

Can AI headphones replace mobile phones?

After 72 hours of testing, let's return to the question we started with: Why add a camera to headphones? And what is the actual experience like?

When we examine this product priced around 2,000 yuan, we must understand its underlying ambition—this product has essentially become independent of the mobile phone, with the only connection being Bluetooth.

This is precisely the biggest difference between Guangfan's and Apple's thinking—in Apple's logic, the iPhone is always the computing power hub, and AirPods are just an extension of the senses; while what Guangfan sells to you is a complete set of independent AI hardware hubs that attempt to completely break away from or even replace the mobile phone.

However, to realize this grand narrative, we must objectively distinguish which of the many flaws exposed under high-intensity use are secondary contradictions caused by the limitations of the times, and which are naturally insurmountable gaps in the short term.

Some obstacles are destined to be overcome with technological and ecological development.

Firstly, there are physical limitations in the operational mechanism. The timing of camera wake-up, the 20-second delay, and the operational architecture of this "edge-side acquisition – cloud computing" still have ample room for improvement in user experience with continuous optimization of the local model.

The current cumbersome third-party authorization barriers are not a terminal illness.

The market itself needs time to educate, but once this type of wearable AI product has enough influence, integration will become a piece of cake—the emergence of intelligent agent tools like Lobster has already set an example for everyone. As long as the influence is large enough, even national apps like WeChat and the giants behind them will flock to it.

When all applications start proactively providing APIs and CLI interfaces for others to access, breaking down ecosystem silos is only a matter of time.

However, some obstacles are difficult to overcome in the short term, or are even inherent.

This resistance is the "first principle" of headphones today.

In today's world where internet social interaction has become mainstream, the most crucial task of headphones is to establish and maintain a private environment for listening to music and watching videos.

But once it becomes a proactive AI assistant, it will inevitably interrupt you constantly with its voice—after all, this is the only way for it to demonstrate its existence and prove its value.

The equipment originally intended for immersion has become a source of interference. This sense of disconnect in the experience is an inherent flaw caused by requiring the auditory channel to bear high-density information.

This is not only a core experience issue that Guangfan, but also that Apple, OpenAI, and any product teams aspiring to develop AI hardware for headphones must clearly understand.

However, this does not prevent us from having high hopes for the picture it depicts.

Looking back at the Guangfan AI All-Sense Wearable Device, although there are still some frictions in terms of experience and interaction, it's not hard to get a glimpse of the bigger picture: the real purpose of adding a camera to headphones is to explore the ultimate form of the next generation of personal terminals.

Functionally, it will "replace" mobile phones, and even completely "eliminate" the presence of mobile phones in future life scenarios.

Imagine this era: when tools like "lobsters" allow people to control productivity even in mobile scenarios; in the future, you might be wearing a pair of headphones, or a camera headset from Apple or OpenAI, walking down the street, and with just a voice command, the headset can remotely wake up an agent on your computer to automatically perform complex tasks.

In this sense, Guangfan, as a pioneer, has indeed revealed a very attractive future for us. After all, the movie "Her" together constituted our generation's aspirations for an audio AI companion.

But when I pulled myself back from my reverie, I saw one thing more clearly: the evolution of hardware is always a dance in shackles.

This 2,000 yuan wearable AI device is certainly not enough to completely retire mobile phones, but it has indeed opened a crack in the door to the future.

All that's left is to open the door wider and pave the path behind it smoother.

#Welcome to follow iFanr's official WeChat account: iFanr (WeChat ID: ifanr), where more exciting content will be presented to you as soon as possible.

Xiaomi MiMo-V2.5 series API permanent price reduction, with a maximum reduction of 99%.

Early this morning, Xiaomi Technology announced a permanent price reduction for its MiMo-V2.5 series APIs, with a maximum reduction of 99% compared to the original price, and the price will no longer be differentiated based on the context window length.

The MiMo-V2.5-Pro ​​input cache hit price has decreased from ¥2.80/million tokens in the ≤256k window to ¥0.025, a decrease of approximately 99%; the input cache miss price has decreased from ¥7.00 to ¥3.00, a decrease of approximately 57%; and the output price has decreased from ¥21.00 to ¥6.00, a decrease of approximately 71%.

At the same time, the Token Plan billing system has been optimized, with the pricing remaining unchanged, but the available tokens have increased to 5 to 8 times. The MiMo-V2.5 Standard Edition (¥99) plan has upgraded its available tokens from 200M to 1300M, and the Pro plan (¥329) has upgraded from 700M to 4700M.

In addition to the price reduction, Xiaomi also announced the early conclusion of its "Trillion Token Creator Incentive Program." Launched on April 28th, the program had distributed all 100 trillion tokens as of 16:08 Beijing time on May 26th.

QuestMobile: AI-native apps reach 461 million monthly active users.

Research firm QuestMobile released a report yesterday on AI platform adoption logic and source preference. The report shows that as of April 2026, the monthly active user base of AI-native apps reached 461 million, with an average of 91 uses and 180 minutes of usage per user per month.

The report also mentioned that the average monthly usage time per user for Doubao and DeepSeek was 144.6 minutes and 109.5 minutes, respectively, representing year-on-year increases of 80.6% and 106.9%.

Related reading: Stop mindlessly filling up your pockets with stuff! A real-world test of five domestic AI apps, and the strongest one is…

Xiaomi reported revenue of 99.1 billion yuan and adjusted net profit of 6.072 billion yuan in the first quarter.

Xiaomi Group released its financial report for the first fiscal quarter of 2026 yesterday. Total revenue for the quarter was 99.1 billion yuan, a year-on-year decrease of 10.9%; net profit attributable to shareholders was 4.723 billion yuan, a year-on-year decrease of 56.76%; adjusted net profit was 6.072 billion yuan, a year-on-year decrease of 43.1%. Breaking it down by segment:

• Revenue from the mobile phone × AIoT segment was RMB 79.3 billion, a year-on-year decrease of 14.5%, with a gross profit margin of 22.5%.
• Revenue from innovative businesses such as smart electric vehicles and AI reached RMB 19.9 billion, a year-on-year increase of 6.9%, with a gross profit margin of 20.1% and an operating loss of RMB 3.1 billion.
• The overall gross profit margin in the first quarter was 22.0%, a decrease of 0.8 percentage points year-on-year and an increase of 1.2 percentage points quarter-on-quarter;
• Xiaomi MiMo-V2.5-Pro ​​ranked first globally in both the comprehensive intelligence index and the agent index of the Artificial Analysis list, and will invest 60 billion yuan in the field of AI over the next three years.

During the earnings call in the evening, Lu Weibing, president of Xiaomi Group, officially announced that a large new car built on a brand-new platform will be launched in the second half of this year, with multiple models planned, calling it "very innovative" and "very competitive".

At the meeting, Lu Weibing explained the reasons for the decline in YU7 sales: on the one hand, the long delivery cycle caused some users to switch to competing products; on the other hand, there were deficiencies in the initial product line planning, and the absence of an entry-level standard version resulted in insufficient appeal to commuters in central cities.

To this end, Xiaomi launched the YU7 standard version and the high-performance YU7 GT version, using a "low-end and high-end" strategy to complete its product matrix. He stated that current feedback from store test drives and pre-orders is positive, and he remains confident in achieving the annual delivery target of 550,000 vehicles.

Regarding its overseas expansion plans, Lu Weibing revealed that Xiaomi Auto plans to launch its overseas expansion in the third or fourth quarter of next year, starting with Europe. The strategy will be phased, prioritizing developed countries over developing countries, high-end products over mid-range products, and right-hand drive vehicles over left-hand drive vehicles. Preparatory work is currently underway.

Didi experiences system malfunctions in multiple locations across the country

According to Bianews, around 5:40 PM yesterday, multiple users reported system malfunctions on social media platforms. Issues included inability to start a trip, app screen flickering and freezing, abnormal location tracking, inability to display driver location, and inability for drivers to locate passengers.

Screenshots circulating online show that some users encountered messages such as "Sorry, there's a network problem, please try again later" and "Didi has too many orders to handle, please try again later" after hailing a ride. The report also mentioned that users in Guangdong, Jiangsu, Beijing, Jiangxi, Xinjiang, and other regions reported abnormalities, and Didi's services were also affected.

That evening, Didi issued an official apology:

We sincerely apologize that due to a network outage by the cloud provider, some services of the Didi App experienced a brief interruption around 5 PM today. Services have now been fully restored. We are urgently addressing any abnormal charges or other issues that occurred during the outage and will resolve them as soon as possible.

Google CEO explains the reason for lagging AI programming: lack of user entry points like Cursor.

In a recent interview on The New York Times podcast Hard Fork, Google CEO Sundar Pichai publicly acknowledged for the first time that Google is currently lagging behind the industry in the field of AI-assisted programming.

He attributed the problem to the lack of an editor entry point like Cursor or Claude Code that could be directly embedded into developers' daily workflows—a deficiency that caused Google to lose the most critical dynamic workflow data, thus slowing down the iteration speed of models in agent programming scenarios.

In the interview, Pichai focused on showcasing closed-loop test data for Antigravity, Google's internal AI programming tool, including the doubling of internal token consumption every week and the ability to build an operating system from scratch in 12 hours during extreme performance tests.

He stated that Google's underlying technology foundation remains complete, and its current passivity in the open market stems primarily from the limited breadth of application scenarios and user feedback loops, rather than a fundamental lack of underlying capabilities.

Xiaohongshu launches audio podcast service

Yesterday, Xiaohongshu announced the launch of its audio podcast service at the New Generation Creators Festival, allowing direct audio uploads on both PC and mobile devices, and supporting the simultaneous import of podcast content via RSS on PC.

The new feature will showcase podcasts on creators' homepages. Xiaohongshu also plans to launch a dedicated podcast channel in its Discover section, supporting audio listening features such as picture-in-picture playback, resume playback, screen-off playback, timed shutdown, and 15-second fast forward/rewind.

Tesla executive: Feasibility study underway for a tri-motor version of the Model 3

Recently, Lars Moravi, Tesla's Vice President of Engineering, revealed in a special episode of the "Ride the Lightning" podcast that he is studying the feasibility of equipping the Model 3 with a three-motor system.

When Moravi was directly pressed by host Ryan McCaffrey on the show about whether the Model 3 would be equipped with a tri-motor system, he did not deny this direction, saying that he had been "seriously thinking about it."

He specifically mentioned his desire to introduce carbon fiber sleeve motors to the Model 3 platform—the core hardware that enables the Model S Plaid to achieve sustained and powerful acceleration.

Currently, the Model 3 Performance uses a dual-motor layout, front and rear. Adding a third motor within the rear subframe would present extremely limited engineering space. Moravi admitted that this is a "high-investment, cost-benefit-balancing" plan, which does not align with Tesla's overall development goals at this stage.

He further explained that the company's current high-performance R&D resources are all focused on the next-generation Roadster. This flagship sports car will be equipped with Tesla's latest generation of electric motor technology, which is currently a priority for the engineering team. Moravi stated that once the Roadster officially enters production, the related high-performance motor technology is expected to be extended to other models in the future.

Qijing Auto announced the gradual opening of its first batch of 300 related stores nationwide, covering 70 cities.

Qijing Auto announced yesterday that the first batch of 300 related stores across the country are gradually opening, covering first- and second-tier cities such as Beijing, Shanghai, Guangzhou, Shenzhen, Hangzhou, Wuhan, and Zhengzhou, totaling 70 cities. The stores opening this time are of two types.

• The "User Center" is operated by Qijing's directly operated or authorized dealers and mainly provides services throughout the entire process, including sales consultation, in-store car viewing, test drive appointment, vehicle delivery and after-sales service.
• The "Experience Center" is jointly created by Yinwang, Qijing and authorized dealers. It is positioned as an immersive experience space, with the display and experience of Huawei Qiankun Intelligent Driving Technology and the vehicles equipped with it as the core. It is located in the core business district of the city.

Dongfeng led the drafting and approval of mandatory national standards for passenger vehicle chassis.

Dongfeng Motor recently announced that its first national mandatory standard in the field of passenger vehicle chassis, which it spearheaded, has passed review unanimously. This standard focuses on brake assist systems and has been under preliminary research for over three years.

The brake assist system recognizes the driver's intention when the driver brakes suddenly but the pedal force is insufficient, and instantly increases the braking force to the maximum to shorten the braking distance.

Dongfeng stated that the BAS (Battery Access System) equipment rate among newly launched M1 and N1 models in my country has been increasing year by year, with over 90% of new passenger vehicles now equipped with BAS as standard; exported models are now 100% required to be equipped with BAS in accordance with UN Regulation UN R139.

Jensen Huang: In the AI ​​era, what you learn is not important; storytelling and creativity are what matter.

According to Business Insider, Nvidia CEO Jensen Huang recently discussed educational choices in the AI ​​era during an interview broadcast on Channel 4 in the UK. He believes that what's most valuable to cultivate today is storytelling, creativity, and judgment, rather than betting on a specific subject in advance.

Whatever you choose as your passion, the only thing you need to do is ask yourself: How can artificial intelligence enhance my learning, my skills, and my mission?

Huang Renxun stated that in the past, people would easily recommend that children study physics, engineering, or computer science; however, with AI being able to quickly supplement tool capabilities, choosing a single major is no longer the only answer.

His core judgment is that storytelling, creativity, and judgment will continue to be valuable. AI can help students enter different fields like mentors and coaches, but learners still need to ask good questions, organize information, evaluate results, and explain complex content clearly.

Jensen Huang views AI tools as a way to expand educational feedback: whether students are studying art, science, or engineering, they can use AI to gain faster practice and explanations. He doesn't assign majors to students, but rather uses "how to use AI to amplify one's own judgment" as a new learning framework.

vivo's first wireless noise-canceling headphones are scheduled for release on May 29th.

vivo officially announced its first wireless over-ear noise-canceling headphones yesterday, which will be released on May 29th at 19:30. The teaser poster shows that the headphones will be available in two colors: Cloud White and Breeze Purple, and will weigh approximately 238g.

In terms of functionality, the headphones support highly immersive active noise cancellation, seamless cross-ecosystem connectivity, and multi-device switching.

Starting at 1699 yuan, XREAL launches its younger sub-brand xbx, and unveils its first AR glasses, the a01.

XREAL announced yesterday the launch of its younger sub-brand xbx (X by XREAL), with its first product being the xbx a01 series AR glasses. The new product focuses on portable large-screen entertainment, gaming, and commuting scenarios, with the brand slogan "Just Play".

The XBX A01 series features a translucent lens and a replaceable "versatile front frame" design, allowing users to replace the front frame or 3D print custom accessories. The entire unit weighs only 62g and comes equipped with three sizes of nose pads, three-position adjustable flexible temples, and adaptive flexible hinges.

In terms of display, the xbx a01 supports 1600 nits of perceived brightness, 14 levels of brightness adjustment, HDR10, 1.07 billion colors, AI SDR to HDR conversion, and a 50° field of view, corresponding to a display effect of approximately 147 inches from 4 meters away.

The new product also features ultra-clear spatial image stabilization, which reduces image shake in moving scenarios such as subways, high-speed trains, and airplanes through AI posture prediction, motion capture, and image pre-rendering.

In terms of price, the xbx a01 is priced at 1699 yuan, and the xbx a01+ (which adds a sun hood and surround mode) is priced at 1799 yuan.

Xiaomi Auto releases world model framework, integrating reconstruction and generation.

Xiaomi Technology yesterday released a new framework for Xiaomi Auto World Model, designed for assisted driving world models. It merges the reconstruction module WorldRec and the generation module WorldGen into a single system. According to the introduction, this solution avoids the sequential path of first reconstructing the scene and then simply feeding it to the generated model, instead allowing the two sides to constrain each other.

WorldRec maintains a 4D Gaussian global representation that expands with observations, providing 3D geometric priors for the generative model; WorldGen completes future frames, unobserved viewpoints, and occluded regions, enabling the model to generate low-frequency but high-risk long-tail driving scenarios.

Xiaomi claims that this framework complements each other in terms of stability, consistency, and realism. The generation side also employs a two-stage training process: first, pre-training with full bidirectional temporal attention, followed by causal fine-tuning, and then reducing the denoising steps from 50 to 4 through ODE distillation, thus speeding up the process by 12 times.

Alibaba's Qwen 3.7-Max tops the list of domestic smartphones, second only to Claude Opus 4.7 Thinking.

Yesterday, Alibaba Cloud launched a new AI product website, Qwen Cloud, and the Agent product MuleRun for overseas markets, and updated its intelligent agent programming platform Qoder and general desktop intelligent agent QoderWork.

Qwen Cloud is positioned as an AI product portal for overseas developers, providing three types of entry points: website, Skills, and CLI. The website is used to browse, try out, and compare models and access OpenAI-compatible APIs; Skills encapsulates platform capabilities into standardized commands that are readable by agents; and CLI provides a reusable command-line operation layer for developers and agents.

Alibaba Cloud also revealed that Qwen3.7-Max scored 1541 points on the LMARana Code Arena WebDev leaderboard, ranking second only to Claude Opus 4.7 thinking, surpassing Claude Opus 4.6, and ranking first among domestic products.

This model is designed for agent-based scenarios and can work with frameworks such as Claude Code, OpenClaw, and Hermes Agent to complete long-range tool invocation tasks.

Alibaba Cloud has also transformed over 60 cloud products into Skill, MCP, and CLI-based systems, enabling agents to call upon these products using standard capability modules. Furthermore, Qwen 3.7-Max has automatic implicit caching enabled by default on the Bailian platform, which can reduce input costs in certain scenarios.

Kunlun Tech releases SkyClaw-v1.0, a model with one million context agents.

Kunlun Tech AI released SkyClaw-v1.0 yesterday, positioning it as a million-context agent model for real-world workflows. The official statement claims that the model supports millions of token contexts and focuses on optimizing complex tool calls, multi-round task execution, code generation, file editing, interactive application building, and research-oriented data analysis.

SkyClaw-v1.0 has undergone large-scale mid-training, high-quality synthetic task SFT, and end-to-end reinforcement learning optimizations. It can run in mainstream agent environments such as OpenClaw, Hermes, and Nanobot, and is also compatible with code agent frameworks such as Claude Code and Codex.

The official statement also claims that the model demonstrates stable multi-step execution capabilities in mainstream agent benchmarks and Skywork's internal Claw tasks, outperforming the Minimax 2.7, DeepSeek V4 Flash, Qwen 3.6 35B A3B, and 27B models, while being priced at less than half the cost of the Minimax 2.7 and Qwen 3.6 series models.

Keenon Robotics releases the small-sized humanoid robot XMAN-L1, integrating with large models from companies like Doubao and Tencent.

According to Jiemian News, Keenon Robotics announced yesterday the launch of its new small-sized humanoid robot, XMAN-L1, which is positioned for lightweight interactive job scenarios.

The XMAN-L1 robot stands 136cm tall, boasts 42 bionic degrees of freedom, a peak knee joint torque of 132 N·m, and a single leg power exceeding 2000W. Its edge computing power is 100 TOPS, supporting integration with large models from platforms like Doubao and Tencent, and enabling natural language dialogue functionality.

The official statement said that the robot can be put into use immediately after its release, and is suitable for scenarios including interaction, customer flow guidance, and light performances.

Facewall Intelligence releases MiniCPM5-1B, primarily targeting edge text models for businesses below 2B.

Wallfacer Intelligence and OpenBMB officially released and open-sourced MiniCPM5-1B yesterday. The official statement claims that this 1B parameter edge-side text pedestal model scored 17.9 points on the Artificial Analysis Intelligence Index's small-size model leaderboard, surpassing Qwen3.5-2B's 16.3 points.

MiniCPM5-1B focuses on making text model capabilities more accessible for deployment. Official materials state that the model weights after INT4 quantization are approximately 0.5GB, allowing it to run on mobile phones, browsers, and low-resource devices; its capabilities cover knowledge, mathematical reasoning, code reasoning, and tool invocation.

The report also revealed that the Base Model version of MiniCPM5-1B was pre-trained using ForgeTrain. The official description states that ForgeTrain is a production-grade pre-training framework for large models written in AI, with a training speed 10% faster than NVIDIA's Megatron.

Alipay AI-powered payment transactions reached 300 million.

Alipay officially announced yesterday that the number of smart agent payment transactions it has processed has reached 300 million, and released its AI Wallet, Token Pay, and ACT Protocol 2.0. The official statement indicated that this suite of capabilities covers leading development platforms, OpenClaw-like agents, smart glasses, smart headphones, smart cockpits, and some AI tool platforms.

Alipay AI Wallet is geared towards individual users, and its core function is to manage AI agent authorization and payment tasks. Users can view AI agent tasks before and during payments, and check their bills after payments; Token Pay is for AI services that are billed in tokens.

Alipay also stated that its self-developed "AI Payment Intelligent Security System" has passed two security certifications from the China Academy of Information and Communications Technology (CAICT) Telecommunication Technology Labs, and has launched token subsidies for AI developers, zero payment fees for individual developers, and fee discounts for enterprises.

Shanghai launches online electronic license verification application for catering businesses.

The Shanghai Municipal Administration for Market Regulation launched the nation's first online catering electronic license verification application yesterday, aiming to crack down on "ghost food delivery" services. This application relies on the electronic business license system to connect regulatory departments, platforms, and merchants, forming a closed loop from merchant application, license comparison, personal authorization to platform display of licenses.

In terms of process, when merchants join the platform, the information they submit will be compared with the official database in real time. After the verification is passed, the legal representative of the merchant needs to complete real-name authentication and precise authorization through the "Electronic Business License" mini-program. The platform can then retrieve the official electronic license with the platform name and time added, and display it on the store's homepage.

The regulatory authorities stated that this mechanism shifts the platform from manual comparison and post-event processing to pre-event verification, and can trace the platform's records of obtaining merchant authorization to retrieve licenses and permits.

Douyin (TikTok) announced the results of its AI-powered rumor-mongering efforts: related content saw an average decrease of 62% in views.

Douyin disclosed yesterday that in the past year, the platform has introduced large-scale model technology in its efforts to combat rumors, resulting in a 62% decrease in the average number of views for rumors that have been dealt with. The platform also compiled a list of the "Top Ten Rumor Debunking Cases" from the past year, stating that these rumors were all identified or dealt with through the AI-powered truth-seeking model, and that authoritative source information was aggregated and debunked through "truth cards."

Douyin had previously launched the "AI Douyin Truth Seeker" feature, which allows users to use a large model to help identify rumors and view the true information about events. In this disclosure, the platform positioned AI Truth Seeker as part of its rumor control tools, designed to reduce the number of views before rumors spread.

Jay Chou's "Son of the Sun" is now available on Apple Music in China.

Jay Chou's new album "Son of the Sun" ended its approximately two-month paid exclusivity period yesterday and was released on Apple Music in China.

The album was released on global digital platforms on March 25th of this year. Previously, it required separate purchase on platforms such as QQ Music, with a single album costing approximately 40 yuan. Now, Apple Music users in China can listen to it directly through a subscription.

JVR Music's official Weibo account has previously released seven music videos since the release of "Son of the Sun".

The trailer for "Love is Anger" titled "A Warm Home" has been released, and the film is scheduled for release on June 19th.

According to Weibo Movies, the film "Love is Anger" released a trailer titled "A Warm Home" yesterday. The film is directed by Park Song-il, written by Ri Zelin, and stars Wang Anyu as Liu Hao and Wang Yuwen as Chen Caicai.

The trailer revolves around Liu Hao and Chen Caicai's meeting, falling in love, and starting a family. The couple initially hoped to build a warm and loving home, but their peaceful life is unexpectedly disrupted. The film is scheduled for release on June 19th, during the Dragon Boat Festival holiday.

The Long March-themed film "Four Crossings" is scheduled for release on June 26.

The major revolutionary historical film "Four Crossings" has been officially announced to be released nationwide on June 26 this year, as a tribute to the 90th anniversary of the victory of the Red Army's Long March.

The film is produced by Andrew Lau, written by Liu Yi, and stars Liu Ye, Wang Lei, Wang Zhifei, Yu Shi, and Wang Yaoqing. Directed by Xu Zhanxiong, the film focuses on the historical event following the Zunyi Conference, where 30,000 Red Army soldiers, under the leadership of Comrade Mao Zedong, crossed the Chishui River four times amidst an encirclement by 400,000 enemy troops, achieving a victory against overwhelming odds.

#Welcome to follow iFanr's official WeChat account: iFanr (WeChat ID: ifanr), where more exciting content will be presented to you as soon as possible.

XREAL reserved its first press conference of the year for a brand it had never heard of before: xbx.

The full name inside is x, by XREAL.

Considering its price of 1699, xbx's first product, the a01, offers excellent value for money: a 50° field of view, a tandem OLED display module with a brightness of up to 1600 nits, equivalent to a 147-inch screen at approximately 4 meters, and support for HDR10 and spatial image stabilization at up to 120Hz refresh rates.

But specifications are far less important than aesthetics and comfort. Weighing only 62g, with a semi-transparent, futuristic body and multiple interchangeable personalized frames, CEO Xu Chi says, "Aesthetics are everything; just have fun!"

This is the first time XREAL, which has been striving to "go up" for the past ten years, has "gone down".

Over the years, Xu Chi and his company XREAL have never taken an easy path. Although China's consumer electronics supply chain resources are so good that integration capabilities have become an essential quality for successful hardware entrepreneurs, XREAL does not engage in pure "supply chain integration."

On the contrary, XREAL has always been moving upmarket, making the most expensive, most difficult, and most "geeky" products. To this end, XREAL is willing to do an extremely high proportion of self-developed products, which is also the highest proportion in the industry, and even at the cost of losing a considerable portion of its profit margin due to the fluctuations in the international semiconductor market over the past two or three years.

This is why XREAL was able to collaborate with Silicon Valley giant Google last year to launch Project Aura, a prototype that impressed industry insiders and shattered consumers' stereotypes about existing "smart glasses" (which will officially launch this year). Xu Chi unabashedly called Project Aura the "ceiling" of experience in his industry.

Creating such a product couldn't possibly be achieved by simply integrating the supply chain. Why is XREAL the only one capable of this? Why did Google choose it, and why did LG, ROG, and others also choose XREAL?

Xu Chi said that the answer is XREAL's A side: introverted, austere, long-term-oriented, and betting on absolute technological innovation.

So, what is the B-side of XREAL?

While Xu Chi led the company relentlessly towards performance optimization and extreme lightweighting of head-mounted display technology, he faced numerous perilous situations behind the scenes:

After testing the waters with Apple Vision Pro and making a mistake, Apple has launched more lightweight AI/AR glasses products, which are expected to be gradually released throughout the year from the second half of 2026 to 2027; Xiaomi, Alibaba's Qianwen, Rokid, VITURE and others have entered the market.

Among them, some are using AR display glasses to besiege XREAL's long-held stronghold, while many more are using AI glasses (screen display/screenless) to preemptively seize new spaces that industry leaders have not yet clearly defined—regardless of the product definition, prices are being driven lower and lower.

Xu Chi wasn't worried about this. After a deep conversation with him, iFanr came to this conclusion: XREAL has been showing its A side for so long that its peers seem to mistakenly believe that it doesn't have or disdains to show its other side.

"Well, if side A isn't established, side B won't exist. We've arrived now, though late."

With the momentum of the XREAL main brand already in place, Xu Chi and his product team are finally able to free up resources to create a different style of product—a younger, more outgoing, and more affordable brand called xbx.

This is the B-side of XREAL, which mirrors the ever-innovative A-side.

He said that since starting his business, he has witnessed the VR/AR bubble burst; then the metaverse came and burst. Looking back, this entrepreneurial journey has increasingly resembled a 10,000-meter marathon—a format he has firmly believed in from the beginning. "Getting a head start is not important at all; running in the right direction is what matters."

Therefore, Xu Chi doesn't seem worried about these new competitors. When asked if he's afraid of large companies and other startups rushing in, he replied, "What we fear most is that we're the only ones in this industry. If there's no one else, it might mean we're going in the wrong direction. So a lively atmosphere is good."

In 2016, Xu Chi left Magic Leap, a pioneering company in mixed reality, and returned to China to start his own business, making a pair of glasses that no one believed in at the time.

Nearly ten years have passed, and he is grateful that XREAL has survived to this day, with luck playing a significant role.

"I am grateful for the ups and downs of the first ten years of this track, which gave me the opportunity to keep leveling up… so that when it comes time to really compete with the big companies, I won't be facing the finals right away."

iFanr and others conducted an exclusive interview with Xu Chi and Liu Zongkai, the product manager of XREAL. They started by talking about the new sub-brand xbx and the first AR glasses product, a01, priced at 1699 yuan. The conversation went on to discuss his views on competition, how he compares his own capital efficiency with that of his peers, whether AI glasses will eventually replace mobile phones, and how a first-time entrepreneur has managed to survive to this day.

"The best thing about young people is that they don't blindly follow tradition."

ifanr: XREAL's defining characteristic in recent years has been its premiumization strategy. Why launch the xbx brand at this particular time?

Xu Chi: We've always said that the smart glasses industry today is very similar to the mobile phone industry in 2005 and 2006—very fragmented, with no unified system, application ecosystem, or interaction paradigm. In this relatively early stage, no single brand can cover all price ranges. So we wondered if it was possible to create two products with different styles, like side A and side B.

This industry is notoriously difficult and fraught with challenges. For a long time, everyone was figuring things out, and we ourselves didn't have a particularly clear positioning for a period of time. But in my mind, "XREAL" was that brand of ultimate innovation, more ruthless and more classic. However, before a brand could be established, I couldn't create something more mainstream or macro-level.

Gradually, XREAL became the more understated brand I wanted, and at that point, a more vibrant brand could complement it. This is what I mean by the A-side and the B-side. Without a solid A-side, there is no B-side.

This may be late, but it has come at a time. From this day forward, we are not just a company called XREAL, but a company that is x by XREAL.

ifanr: What kind of AR glasses do young people want? Don't young people want cutting-edge products?

Liu Zongkai: For young people, individuality and self-expression are desires in everyone's heart. Whether it's AR glasses, AI glasses, or XR helmets on the market, many people subconsciously feel that these things should be neither good-looking nor lightweight. But we insist on doing the opposite. Why not first make a pair of glasses that are good-looking, lightweight, and that young people are willing to wear? A pair of glasses that they are willing to wear out is the first step in everything.

Xu Chi: The best thing about young people is that they don't blindly believe in those traditional grand principles. Looks are everything; as long as it looks good and works well, that's all that matters. We hope to use the a01 glasses to show everyone that you can get a two-thousand-yuan experience for a price of just over a thousand yuan. We will continue to develop this product long-term.

ifanr: It must be worn outdoors to reach a wider potential audience.

Xu Chi: That's right. We hope that more young people can wear this product on the subway, in cafes, on airplanes, and in various places without feeling embarrassed. So we made it extremely lightweight and personalized.

ifanr: There are always trade-offs in lightweight design. If a product wants to achieve higher resolution, frame rate, and field of view, the module will become larger; a larger module will result in greater weight and an imbalance in weight distribution.

Liu Zongkai: Making a lightweight device while maintaining a good user experience is incredibly difficult. Could we reduce the thickness of the lens and casing even further, while still preserving their strength? We meticulously considered every detail to ensure that each component was both lightweight and performed optimally.

This road has no end; it's just one night after another, one struggle after another, one argument after another. Of course, we feel we can do better.

ifanr: What is the physical limit of the field of view (FoV) that AR glasses can achieve? And without considering the limit, what is the optimal solution for ergonomic comfort for humans wearing them in different scenarios?

Xu Chi: I'll give you the most direct answer. The best field of view should be around 85°, but that's assuming we don't care about the cost of materials or the weight.

On Project Aura, we achieved 70°, which is already quite good for this product form factor, but there's still a gap. When we can achieve 85°, while still maintaining a lightweight design, then we'll feel we've at least reached the pinnacle of display technology.

Liu Zongkai: It depends on the scenario. For example, if you wear a VR headset to Mars, the main subject is a spaceship, and the background is the starry sky—you need to see both the subject and the background simultaneously to achieve maximum immersion. However, for AR glasses, the best background is actually the real world. If you're playing games or watching a sports game, you might not need a wide viewing angle; but if you're watching a movie, or an AR display attached to a real environment, then a wide viewing angle is more important for immersion. So ultimately, it depends on whether the content is immersive.

As for the human eye's focusing angle, from an ophthalmological perspective, there is indeed a limit, generally around 50° horizontally and 30-40° vertically.

Field of view is not the only key factor; there are also electrochromic properties, performance, battery life, and so on. When defining different products, we have countless trade-offs to make, and we can make choices in different directions.

"What we fear most is that we're the only ones left in this industry."

ifanr: Apple has entered the market, and competing products from major domestic manufacturers have also been launched, driving prices lower and lower. What's your opinion?

Xu Chi: I think it's a good thing that everyone is coming in. What we fear most is that we're the only ones in this industry—that means no one is paying attention to this sector, and no one is optimistic about it.

We firmly believe that glasses are the next-generation computing terminal with the greatest potential to replace mobile phones. Although we have been in business for ten years, we are just getting started. Our penetration rate is probably less than 1%, and there is still room for growth of hundreds of times or even more. So it's a good thing for everyone to work together to make the market bigger.

There is a bubble in our industry, but a bubble isn't necessarily a bad thing; it means people have high expectations for the industry. In the past, at every stage, there have been people who wanted to make quick money, but they left when they found it wasn't profitable. When the bubble bursts, it's actually the consumers who suffer. What truly drives the industry forward are those who gradually narrow the gap between "user expectations" and "product experience."

To put it simply, today's AI glasses are like a five-year-old child, while our defined all-weather AI glasses are like Jarvis. This gap needs to be bridged step by step through fundamental innovation. These innovations don't happen out of thin air; there are always people working tirelessly behind the scenes.

Q: How far are you from Meta?

Xu Chi: To give a somewhat inappropriate comparison: In 2025, Meta's Reality Labs business will generate $2.2 billion in revenue and incur losses of nearly $20 billion. This year, we achieved $200 million in revenue, roughly one-tenth of that, but our losses are less than $20 million.

With one-tenth of our revenue and one-thousandth of our losses, I think our capital utilization efficiency is quite good, which is also our advantage.

ifanr: You have your own fully self-developed chips and optics, but Project Aura still uses Snapdragon for some of its computing power. What is the relationship between these two? Will XREAL increase its autonomy in core computing power in the future?

Xu Chi: The X1S is a complete SoC. On Aura, all latency- and bandwidth-sensitive calculations are performed on our X1S chip, while the rest are handled by Snapdragon.

Our chip is purely for edge computing; the Snapdragon chip is mounted on a puck (an external computing unit). The relationship between these two isn't that of a processor and a coprocessor, but rather more like that of "edge" and "cloud." Some computations need to occur closer to you, more promptly.

We've been saying that glasses will replace cell phones. In the foreseeable future, pucks will disappear, and you can simply replace them with your cell phone; in the longer term, if glasses really do replace cell phones, they need to be able to handle all the calculations themselves. That's why we're betting on domestically developed chips.

The US recently issued a ban, prohibiting the direct import of advanced process wafers into mainland China. This was quite shady. Our chips, which should have been packaged in mainland China, now had to be packaged in Taiwan before being shipped back. At the time, a large number of domestic chip manufacturers were vying for packaging resources in Taiwan, causing a one-time shortage, similar to today's memory shortage. As a result, our revenue suffered a significant loss; otherwise, the growth in Q3 and Q4 of last year would have been substantial.

But in the long run, this actually drives us to keep moving forward. Fortunately, our sales volume isn't very large today, which is better than selling millions of units and then suddenly being held back (laughs). We hope that China will have more and more advanced manufacturing processes in its own hands, so that no one can hold it back.

ifanr: Will Project Aura be released in China? Which domestic model manufacturers will you choose to explore this project with?

Xu Chi: Because Android XR and Gemini are strongly bundled, and Gemini can't be used in China. So unfortunately, you might as well buy it from overseas (laughs).

We won't give up on the domestic market. If Android XR can be decoupled from Gemini and connected to domestic AI, that's when Project Aura will enter the domestic market, but not today. Just like the iPhone didn't enter China immediately after its release. I think we can accept this outcome. It's okay.

For us, Alibaba is a shareholder, and we've maintained communication with ByteDance. Regarding the model, we won't exclude any company. Our ultimate ideal is that AI can be used interchangeably, like search engines. Future large-scale models will become infrastructure; we'll use whichever token performs best, allowing for seamless switching.

"Why should glasses replace cell phones?"

ifanr: You yourself said that AR glasses have existed for many years, but the penetration rate is still very low. What kind of product would be the "entry-level poison" that gets more people to accept it?

Xu Chi: Most likely, there will still be two main categories: AI glasses that are more all-weather and AR glasses that have displays but are not all-weather enough.

The term "always-on" has two meanings: first, it means wearing it all day; second, it means using it all day. The problem today is that the primary use case for AI glasses isn't AI itself, but rather listening to music and taking photos; you open the camera and take pictures for 30 minutes, and the product runs out of power. If you consider the glasses as your personal assistant, but they can only be open for 30 minutes a day… then they are not an all-day assistant.

At some point in the future, there will be a product weighing less than 35g with all-day battery life, serving as a platform for AI interaction. I believe such a product is achievable. If it does, it will definitely become a ubiquitous device.

Another path is AR glasses, which aim for higher resolution and more content. This approach currently uses a separate device form factor, achieving a weight of 60g, but the ultimate form may be an all-in-one device.

These two types of products are like the iPhone, which everyone has one of, and the entire category may have a shipment volume of more than one billion units per year; the other is like our current devices, and the ultimate form may be the combined volume of tablets and laptops, with a volume of 150 to 250 million units per year, which is also very good; and traditional headsets may have a volume similar to desktop computers—these three will coexist for a long time.

As for the product that will completely ignite the category and truly propel glasses onto the path of "replacing mobile phones," I believe we will see a clearer answer by 2027 or 2028.

ifanr: Even if you achieve the ultimate in lightweight design, how do you convince customers who still find it too heavy?

Xu Chi: I think people today are too easily influenced by stereotypes, such as "I absolutely won't wear one without 35G." In today's industry, excluding subsidized products, there isn't a single product that has sold over a million units without government subsidies. If it truly achieved 35G, it would already be at the level of 1.5 billion units sold.

We need to take it one step at a time: first, take a single product to a million units, then to tens of millions, then to a hundred million, then 1.5 billion. There are many steps in between. I believe that today, a pair of glasses with a sufficiently good user experience, even weighing only 50g, can still sell a hundred million units. What affects acceptance and hinders sales is simply that the user experience hasn't been refined enough.

ifanr: Mobile phone manufacturers believe that mobile phones will remain the dominant technology for the next 5-10 years. However, they are also making eyewear. What do you envision the future competitive landscape to be like?

Xu Chi: Indeed, what exists today will continue to exist for a long time. But the key is who can stand at the top of the value chain. For example, there was a time when we thought the super apps of major internet companies were amazing, but today they certainly don't have the same prominence as AI companies. The same goes for mobile phones. With technological advancements, there will always be new fields and companies that will rise to higher positions in the value chain.

We believe a consensus will emerge within the next two years: glasses are the best native terminal for AI, and may be the closest thing to multimodal AI. This is why we are working with Google to envision a completely new interaction paradigm for the future, and what terminals under this new paradigm will look like.

This task excites me greatly, firstly because it is difficult, and secondly because it is very rewarding if done correctly.

ifanr: Are other forms of AI hardware, such as pins and headphones with cameras, inferior to glasses?

Xu Chi: It's not just me who thinks this way; Demis Hassabis also said that glasses are absolutely the most central device in all of AI. Because only glasses can obtain the crucial contextual information of a person's attention.

You wear a PIN, which can see a bunch of people in front of you, but future glasses will have eye-tracking capabilities. They'll know exactly what you're looking at at that moment, making surrounding information less important. Only glasses can provide an end-to-end closed-loop data link; other devices don't have this capability. Of course, other forms can assist, but glasses will definitely be the most crucial entry point.

"You need to rely on luck, but you also need to fight your way through challenges."

ifanr: Entrepreneurs and business owners have different sources of fear. It might be internal organizational inefficiency lagging behind the times, competition from peers, or disruption from other industries. What is the fear that could wake you from your sleep?

Xu Chi: Running a business is like being a person. People also experience confusion, and there are people who can guide you and help you find benchmarks. But I think that ultimately, all troubles are self-inflicted.

I believe that all great companies are value-driven. The core is to find a comfortable organizational state where the entire company accepts your set of values—whether they leave or stay, they will continue to work under the guidance of these values. As long as this is achieved, competition and other factors will not matter much.

I personally sleep quite well, which I think is a quality that entrepreneurs should have (laughs).

If there's anything that truly worries me, it's whether the values ​​I uphold can truly be fully implemented. I fear that as XREAL grows larger, its culture will be diluted. I need everyone to genuinely believe one thing: we must be innovators and leaders. This isn't easy, especially in China. In China, people are used to obeying hierarchical systems, believing "whatever the boss says is right." But I still hope that we can form a flat and efficient mechanism, both from the bottom up and from the top down.

ifanr: Just like you said, XREAL has survived several rounds of bubbles and is still alive today.

Xu Chi: I returned to China from Magic Leap in 2016, and it's been exactly ten years since then. Back then, it was truly a grassroots startup; I just wanted to make a pair of glasses. Looking back, I'm incredibly lucky to have survived this long. This was my first entrepreneurial venture, and I'm grateful to the investors (and other partners) along the way for helping me gradually understand how to operate a company, an organization, and a business.

To be honest, if this industry had developed any faster or gained momentum any stronger, and we hadn't had the chance to hone our skills to cope with the intense competition when giants entered the market, we might not have survived.

Every startup probably goes through this phase: you have to level up before you can reach a bigger stage. If you start with a top-tier boss, you're outmatched by competitors like Alibaba and ByteDance. So I'm actually quite grateful for the ups and downs of the first ten years of this industry, which allowed me to finally compete with the big companies.

The AR industry is notoriously difficult, and I've been in it for quite a while, so I've become less concerned about it now. As long as everyone's still at the table, it's a long-term game.

I believe AR is a marathon, and running in the right direction is more important than getting a head start. If the industry is still in its early stages but everyone is rushing in the same direction, that so-called consensus is likely just a bubble. Conversely, things that are not considered consensus in the early stages are often proven correct by time. History has shown us this countless times.

By Du Chen

Interview by Du Chen

#Welcome to follow iFanr's official WeChat account: iFanr (WeChat ID: ifanr), where more exciting content will be presented to you as soon as possible.

Those who bet on a surge in token prices in 2026 have been proven wrong twice in just one week.

On May 22, DeepSeek announced a permanent price reduction for the DeepSeek V4 Pro; early this morning, Xiaomi's MiMo-V2.5 series followed suit with a price reduction of up to 99%.

At the same time, Xiaomi's Token Plan billing system has been optimized, with the pricing remaining unchanged, but the available quantity has increased to 5 to 8 times the original amount.

Unsurprisingly, discussions about the price reduction of the Xiaomi MiMo model on overseas Reddit, the X platform, and various developer forums also surged rapidly.

However, at a time when the entire industry is lamenting the unbearable costs of tokenization, why is Xiaomi daring to go against the tide and lower prices? More importantly, where will this price reduction push the AI ​​industry?

Token prices plummet, the AI ​​industry welcomes its strictest father yet.

Xiaomi's announcement indicates that its MiMo-V2.5 series AI large-scale model API will undergo a permanent price reduction, with a maximum reduction of 99%, and the price will no longer differentiate based on input length. The new price took effect globally at 00:00 Beijing time on May 27th.

However, a 99% reduction does not mean that every call will be charged at the lowest price; the key variable is whether the input cache is hit.

Taking MiMo-V2.5-Pro ​​as an example, once the cache is hit, the input price drops to approximately 0.025 yuan per million tokens. However, if the input cache is not hit, the price remains at 3 yuan per million tokens, and the output price is 6 yuan per million tokens.

In other words, the prerequisite for this extremely low price is that requests must hit the cache a large number of times.

This price is very attractive for tasks with high repetition of contexts, high frequency of agents, multi-round code tasks, and batch inference tasks. However, if your application scenario has a poor cache hit rate, the actual cost will obviously not reach the lowest point.

The Token Plan operates on a similar logic.

Xiaomi emphasized that pricing will remain unchanged, but credits will be significantly increased: the monthly fees for the four tiers of Lite, Standard, Pro, and Max will remain at 39 yuan, 99 yuan, 329 yuan, and 659 yuan, respectively. The credit limits will also be increased from 60 million, 200 million, 700 million, and 1.6 billion to 4.1 billion, 11 billion, 38 billion, and 82 billion, respectively.

According to the new conversion, MiMo-V2.5-Pro ​​only requires 2.5 Credits/token to hit the cache, and 300 Credits/token to miss the cache, with an output of 600 Credits/token.

This is exactly the same strategy used by DeepSeek.

Here's a brief timeline: DeepSeek V4 preview version was released on April 24; the next day, V4-Pro was offered at a 25% discount; on April 26, the price of cache hits plummeted to one-tenth of the initial price; and by May 22, the temporary discount had become a permanent price reduction, with V4-Pro permanently reduced to one-quarter of its original price.

After some adjustments, the input cache hit price of DeepSeek-V4-Pro dropped from 0.1 yuan to 0.025 yuan. With Xiaomi MiMo-V2.5-Pro ​​quickly following suit, the input cache hit price of domestic models has been completely locked at this benchmark.

Both DeepSeek and Xiaomi have focused their most impactful pricing strategies on cache hit rates and scenarios, and the reason is not complicated. The larger model is shifting from chat to actual work, and the Agent is where token consumption truly amplifies.

In chat scenarios, the user asks a question and the model answers, making the cost relatively easy to estimate.

However, in an agent-based scenario, a task may involve long contexts, multiple rounds of inference, code generation, tool calls, webpage reading, file analysis, and result verification. The user only sees the final output, while the backend may have already processed multiple requests and a large number of context reads.

This is where cache hits are important.

Agents, code helpers, and long-context applications share a common characteristic: much of the content appears repeatedly. This includes system prompts, project code, API documentation, tool descriptions, historical conversations, and dependency files. Recalculating this content every time would be very costly; however, if it can be cached, and billed only based on cache hits the next time it's used, the inference cost will significantly decrease.

In other words, the lower the cache hit price, the more suitable it is for real-world work scenarios involving high frequency, multiple rounds, and long contexts. The low prices offered by DeepSeek and Xiaomi are actually aimed at attracting developers and high-frequency applications, encouraging more agents, code assistants, and office automation applications to run on their models.

Xiaomi previously used initiatives such as MiMo Orbit and the Trillion Token Creator Incentive Program to allow more people to experience MiMo and solve real-world problems. This Trillion Token Incentive Program, launched on April 28th, saw all 100T tokens distributed ahead of schedule by 16:08 on May 26th.

From the platform's perspective, the low-priced tokens and free quotas result in a massive amount of real-world usage. These real-world usages bring complex tasks, failure samples, user feedback, agent workflows, code scenarios, and long-term contextual data, all of which in turn help the model and inference system iterate.

The phenomenon of "shrimp farming" in the community can also be understood within this logic. While maximizing their spending limits, users are also helping the platform create pressure, expose problems, and accumulate data.

Therefore, this cannot be analyzed solely based on the gross profit per inference. While short-term revenue is suppressed, the gains come from developer migration, increased call volume, and genuine feedback. For model vendors aiming for a position in the Agent ecosystem, this represents a very worthwhile platform investment.

Luo Fuli's "True Fragrance Law" is rooted in engineering violence.

However, having the will is not enough; the key is being able to afford to lower the price. What makes Xiaomi's price reduction this time special is that it contrasts with the previous public statements made by Luo Fuli, head of MiMo's large-scale model.

A month ago, Luo Fuli publicly opposed the token price war. Her assessment at the time was that low-priced tokens combined with an open third-party agent framework could easily lead to uncontrolled costs for the platform.

She noted that third-party agent frameworks often have poor context management. A single user query can trigger multiple rounds of low-value tool calls, with each request carrying an excessively long context containing over 100,000 tokens. If the platform cannot constrain this waste, the actual API cost could be dozens of times the subscription price.

She also believes that global computing power supply can no longer keep up with the growing demand for tokens driven by agents. Large-scale companies, without clarifying the cost structure of programming and agent scenarios, engaging in blind price wars will lead to throttling, resource degradation, and decreased stability, ultimately harming the user experience.

However, Xiaomi's price cut this time did not overturn previous judgments, but rather changed the premise for a price war. Luo Fuli previously opposed low prices without a supporting cost structure. What Xiaomi is now showcasing is a theoretical engineering solution that it believes can support low prices.

According to Xiaomi's announcement, its technical team, based on SGLang HiCache, fully supports SWA, which stands for Sliding Window Attention. This reduces the amount of data movement between multiple levels of storage such as GPU memory, CPU memory, and SSD in KV Cache to nearly one-seventh of what it was before optimization, and increases the number of cacheable tokens to nearly five times that before optimization.

At the same time, Xiaomi also optimized its expert parallel processing solution and input length bucketing strategy to improve the cluster's input throughput. Without this level of engineering capability, low prices can easily become unsustainable subsidies. Only with a sufficiently robust infrastructure system can low prices be transformed into a long-term advantage.

Price wars test engineering capabilities, as well as the strength of the support system.

Unlike pure AI model companies, Xiaomi's smartphone, automotive, IoT, and consumer electronics businesses provide it with a longer investment cycle and greater strategic patience. It can view its large-scale model services as an entry point into the AI ​​ecosystem, avoiding the pitfall of focusing solely on short-term API revenue.

This is not friendly to small and medium-sized model companies. Without a core business to support them, without strong infrastructure capabilities, and without players with sufficient scale to dilute costs, they are destined to be unable to keep up with this price in the long run.

DeepSeek's low prices have directly threatened the market positioning of many domestic model providers. With Xiaomi MiMo following suit, more manufacturers with significant scale will be forced to adjust their prices or redefine their product value. Smaller model service providers may be pushed into narrower vertical markets.

This round of price cuts is, to some extent, a market selection process for efficiency-oriented model vendors. Companies with engineering capabilities, computing power scheduling capabilities, and ecosystem entry points can withstand the pressure from lower prices. Companies that only have model capabilities but cannot reduce inference costs will become increasingly passive.

Furthermore, as the room for further price reductions gradually narrows, the closer the price gets to the physical cost, the less valuable simple price cuts become. In the next stage, model quality, agent adaptation, developer tools, ecosystem integration, service stability, and enterprise delivery capabilities will all face a new round of intense competition.

Model capabilities determine the upper limit of AI development, while inference costs determine the scale of AI adoption. Only when truly affordable tokens flood the application layer will we truly see what the next era of AI explosion will look like.

#Welcome to follow iFanr's official WeChat account: iFanr (WeChat ID: ifanr), where more exciting content will be presented to you as soon as possible.