Pogbagram

On the eve of the Beijing Auto Show, HarmonyOS unveiled its major new models for the year at its spring launch event.

In addition to the Shangjie Z7 & Z7T and Wenjie M6, which have already started pre-sales, they also brought the first MPV based on HarmonyOS—the Zhijie V9, as well as the brand-new Wenjie M9 and M9 UL versions.

The official launch price of the Shangjie Z7 & Z7T starts at 219,800 yuan, which is only 100 yuan lower than the new generation Xiaomi SU7.

We have already provided detailed introductions to the Shangjie Z7 and Z7T models, which you can read here. We will now focus on the new models of the Wenjie and Zhijie.

Questioning the World M6

The officially launched Wenjie M6 has a starting price of 259,800 yuan. Undoubtedly, with this price, the Wenjie M6 will become another blockbuster model, just like the Wenjie M7.

The Wenjie M6 is positioned as a five-seater model, with a length, width and height of 4960mm, 1985mm and 1736mm respectively, and a wheelbase of 2950mm. The lateral interior space is close to 2m, providing a relatively spacious riding experience.

The new car continues the family design language of the Wenjie series, with smoother lines than the M7 and a more youthful style. The pure electric version features a mesh structure on the lower front bumper, while the range-extended version uses a strip design to differentiate it.

The vehicle features semi-concealed door handles on the sides and is available in a variety of body colors and 20-inch or 21-inch wheels. Electric side steps are also available as an option for easier entry and exit. The rear is equipped with starlight taillights, illuminated badging, and driver assistance indicator lights.

Inside, the Wenjie M6 also leans towards a youthful aesthetic, adopting a bright two-tone color scheme of white, red, and orange. An "Inspiration Showcase" has been added to the area in front of the passenger seat, equipped with a magnetic tempered glass cover for convenient placement of personalized items, and includes three built-in e-ink screens that can sync photos from a mobile phone.

The center console features a retractable and rotatable central control screen, which can be adjusted by voice commands for the driver, front passenger, and rear passengers. The vehicle is equipped with the HarmonyOS Cockpit 5.0 system, a 26.1-inch W-HUD head-up display, and dual wireless charging pads.

The new car features integrated sports seats in the front row, which integrate heating, ventilation and massage functions and support one-touch folding. The rear seat backrests support 29 to 42 degrees of electric adjustment, also with heating and ventilation functions, and provide huge front and rear trunk space.

To cater to rear passengers, the vehicle also features small tables, an in-car refrigerator/heater, and a 17.3-inch ceiling-mounted entertainment screen. For audio-visual entertainment, the new car comes standard with a 21-speaker HUAWEI SOUND sound system, along with four microphones inside the vehicle and six microphones outside.

The Wenjie M6 comes standard with Huawei Qiankun Intelligent Driving ADS 4.1 High-End Edition, with a total of 36 sensors throughout the vehicle. The 896-line dual-optical-path image-level LiDAR boasts ultra-high definition, ultra-precision, and ultra-long-range recognition capabilities, supporting defensive assisted driving.

Furthermore, the new vehicle adopts a five-star safety architecture design, featuring a robust body structure with eleven horizontal and eleven vertical axes. High-strength body materials account for 83.62% of the total body structure, while hot-formed materials account for 44.76% of the passenger compartment. Sixteen locations throughout the vehicle utilize 2000MPa hot-formed steel. The front trunk, rear trunk, intelligent welcome pedals, power-adjustable seats, dynamic central control screen, and rear entertainment screen all feature a triple-redundant anti-pinch mechanism: sensor-based anti-pinch, hardware-based anti-pinch, and emergency stop.

The Wenjie M6 is available in both range-extended and pure electric versions, both featuring a dual-motor layout (front and rear motors) with maximum power outputs of 160 kW and 180 kW respectively. The range-extended version uses a 1.5T engine as a range extender and offers two battery pack options: 37 kWh or 53 kWh. The pure electric version comes standard with a 100 kWh battery pack.

Zhijie V9

The Zhijie V9 is the third model of the Zhijie brand and the first MPV of the HarmonyOS Intelligent Mobility brand. The pre-sale price given at today's launch event is 399,800 yuan.

Prior to the Zhijie V9, the Zhijie brand already had two models, the R7 and the S7. By 2025, the Zhijie brand's cumulative sales were less than 100,000 units.

Therefore, the Zhijie V9 shoulders the important responsibility of leading Zhijie's sales to break through and elevating the brand.

The Zhijie V9 measures 5359mm in length, 2009mm in width, and 1879mm in height, with a wheelbase of 3250mm. It is available in four body colors: Glowing Purple, Deep Sea Blue, Snow White, and Golden Black.

The new car features a closed grille and a continuous LED daytime running light at the front, with an illuminated brand logo in the center. The vehicle is also the first to be equipped with a new generation of DLP smart projection headlights, which support ground light and shadow projection, cornering assist, and customizable welcome light messages.

Meanwhile, the Zhijie V9 is equipped with an interactive side door light carpet. When the user approaches using millimeter-wave radar, it can automatically trigger welcome lighting effects, open the door, and rotate the second-row seats, improving the convenience of daily use.

Stepping inside, the interior design of the Zhijie V9 shares some similarities with the Zunjie S800, and offers color schemes such as "Velvet Purple." The front row features a debut of an integrated "Universal" triple-screen design, including two 17.2-inch displays with 3.4K resolution and a 21:10 widescreen aspect ratio, and built-in customizable card-style desktops. The driver's seat is equipped with a round three-spoke steering wheel, a streaming rearview mirror, and the family-style column shifter, while the center console features a star-ring diffuser.

The new vehicle features a 21.4-inch ceiling-mounted screen with 8-level tilt adjustment in the second row; a removable control screen is located at the rear of the center armrest, allowing rear passengers to adjust the air conditioning and multimedia. Below the rear of the center console is a 10.9-liter in-vehicle refrigerator that maintains cold for 12 hours after a power outage. Additionally, the vehicle is equipped with an in-vehicle oxygen bar function.

Seat materials and adjustment functions are key design elements of the Zhijie V9's cabin. The seats feature full-grain Nappa leather and semi-aniline leather, with a multi-layered structure and high-resilience padding. The second row is equipped with dual zero-gravity seats, supporting up to 123 degrees of adjustment and one-touch linkage for the backrest, leg rest, and footrest, as well as mechanical massage and active side support functions.

To facilitate entry and exit and daily interaction, the second-row seats also support multi-angle rotation of 45 degrees, 90 degrees and 180 degrees, taking into account both passenger comfort and space flexibility.

The Zhijie V9 is built on Huawei's Tuling platform and features an all-aluminum suspension structure with a front double wishbone and rear H-arm multi-link design. The new vehicle is equipped with dual-chamber air springs and CDC continuously variable damping shock absorbers, capable of adjusting chassis stiffness in real time according to road conditions.

Meanwhile, the vehicle is equipped with a rear-wheel steering system with a steering angle of up to 7 degrees in both directions, which reduces the turning radius of this large MPV, which is over 5.3 meters long, to 5.35 meters, improving driving agility on urban roads and narrow sections.

The Zhijie V9 is equipped with a 1.5T Huawei Snow Owl intelligent range extender system, with a maximum power of 115 kW. Depending on the configuration, the new car is equipped with battery packs of various capacities, with the highest configuration version having a capacity of 53 kWh. The CLTC combined driving range can reach more than 1250 kilometers to meet the needs of long-distance travel.

The new vehicle employs a four-laser radar system, consisting of an 896-line dual-beam image-level main radar located on the roof, and three high-precision solid-state lidars distributed on the sides and rear of the vehicle. Combined with the Huawei Qiankun Intelligent Driving Advanced System, the vehicle's perception accuracy and decision-making efficiency have been optimized, supporting advanced driver assistance functions such as NOA (Noise, Alert, and Assist) on highways and in urban areas.

The Zhijie V9 has been specifically enhanced to address safety concerns in MPVs regarding side impacts, rear passenger protection, and seats reclining at large angles.

The Zhijie V9 is equipped with a total of 13 airbags, incorporating the concept of "follow-up protection." Among them, the second-row seats integrate full-dimensional wrap-around airbags, which adjust their position in real time according to changes in seat angle and occupant posture.

In the event of a side impact or rollover, the airbag deploys to form a U-shaped wraparound structure, enveloping the occupant's head, neck, and chest. Especially when the seat is in a reclined position, in conjunction with technologies such as pre-collision quick return, seat pretensioners, and seat cushion airbags, the airbag can help secure the occupant's head and create a buffer between the left and right occupants.

According to 36kr, the Zhijie V9 has given its suppliers an annual production guidance of approximately 120,000 vehicles, a sales guidance that is close to the total sales of the Buick GL8 last year.

The all-new M9

Another highlight of the entire launch event was the all-new Wenjie M9, with a pre-sale price starting at 499,800 yuan.

The all-new M9 is 55mm longer, 27mm wider, 45mm taller, and has a 15mm longer wheelbase than the current model, bringing the overall dimensions to 5285/2026/1845mm, providing more interior space.

The front end continues the family-style closed grille design, with overall rounded lines. The smoked headlights are connected to the continuous daytime running lights, which stretches the horizontal visual width and echoes the raised lines on both sides of the hood.

In addition, vertically arranged air ducts have been added to the fenders on both sides of the grille to improve aerodynamic performance; the trapezoidal air intake at the bottom is complemented by high-gloss chrome trim, which increases the sense of layering of the front of the car.

The side profile remains largely unchanged, retaining the curved roof and roof rack. The continuous waistline, combined with the recessed door handles, multi-spoke wheels, and creases on the side skirts, maintains the overall sleek and stable style.

The overall configuration and design of the cabin are not much different from the old model, but many details have been improved.

The new car's cabin still features a triple-screen design, including a 12.3-inch LCD instrument panel, a 17.2-inch central control screen, and a 17.2-inch passenger entertainment screen, but the screen bezels are narrower and the display effect is clearer.

The door trim panels have been upgraded to more refined wood grain trim, and the wireless charging area for mobile phones in the center console supports lifting and lowering, making it better suited for foldable screen phones.

In terms of driver assistance hardware, the new Wenjie M9 is equipped with a total of 6 lidars, including one Huawei-developed 896-line forward main radar, one cabin radar, and four solid-state radars distributed throughout the vehicle.

To further enhance its integrated perception capabilities, the vehicle has also added a high-position side camera and a side perception unit at the C-pillar. This laser and vision-integrated solution aims to improve the vehicle's monitoring coverage of its surrounding environment.

The new hardware not only improves the accuracy and distance of identifying small objects in low blind spots, but also increases the ability to predict potholes on unpaved roads.

With this hardware redundancy, the vehicle's intelligent driving system has expanded its application scenarios to unpaved roads, and has also provided a hardware foundation for the further evolution of subsequent assisted driving capabilities.

The new Wenjie M9 is equipped with a fully active suspension system, which integrates dual-chamber air springs, CDC shock absorbers, and rear-wheel steering.

In actual driving, the fully active chassis can be deeply integrated with the vehicle's forward perception system, and can perceive the road conditions ahead in advance through anticipation and recognition technology.

Whether on smooth city roads or complex, bumpy sections, the system can quickly calculate and actively adjust the stiffness, damping, and support of the suspension, thereby ensuring driving safety while providing passengers with a smoother dynamic driving experience.

The new Wenjie M9 still offers both pure electric and range-extended versions, and has been fully equipped with the new generation of CATL ternary lithium battery packs.

The extended-range version continues to use a dual-motor four-wheel drive system, with the maximum power of the front motor increased to 220kW and the maximum power of the rear motor increased to 277kW, and is matched with battery packs of 60kWh and 75kWh capacity.

The all-electric version also uses a combination of front and rear dual motors with power outputs of 220kW and 277kW, and is equipped with a larger 120kWh battery pack.

At the end of the press conference, Yu Chengdong also teased "one more thing" – the Wenjie M9 Ultimate Extended Edition will be unveiled soon, with a pre-sale price of 669,800 yuan.

Its body length will exceed 5.4 meters, the D pillar of the body will have an exclusive "Ultimate" logo, and it will be equipped with an illuminated shield-shaped logo, offering flexible layouts of 4/5/6 seats.

Since its launch in 2023, the Wenjie M9 has maintained a good market performance for a considerable period of time, with cumulative deliveries exceeding 280,000 units.

However, newcomers are also catching up. After its launch at the end of last year, the sales of the Jike 9X grew rapidly and even surpassed the Wenjie M9 in the following months. At the same time, new models such as the NIO ES9 and the new generation Li Auto L9 Livis have been launched or are about to be released.

In this market situation, facing the increasing pressure from competitors in terms of space, performance and energy replenishment system, Wenjie has significantly upgraded its product strength by increasing the body size, comprehensively upgrading the intelligent driving hardware with the 896-line main radar as the core, and replacing it with a fully active suspension system with a pre-aiming function.

This continuous investment in core technologies and user experience has solidified its "technological luxury" brand image, providing it with a "point of reference" in a rapidly changing market.

#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.

This is the latest BMW, and also the most interesting BMW. Of course, not everyone may like this kind of BMW, and it even has contradictory attributes of being both trendy and anti-trend.

These new vehicles are the next-generation BMW iX3 long-wheelbase version and the next-generation BMW i3 long-wheelbase version, as well as the all-new BMW 7 Series.

That's right, after years of anticipation, BMW's next-generation model has finally shed its concept car and camouflage status and entered the consumer market just before the 2026 Beijing Auto Show.

The new generation BMW iX3 long wheelbase version: the biggest changes, but the handling remains the same.

Unlike the current BMW X3, the design of the new-generation BMW iX3 long-wheelbase version can be described as both retro and modern. Retro because its kidney grille resembles the small grille design of the BMW 1500 sedan from the last century; modern because its front fascia features the longest integrated headlight assembly in BMW history, embedding 322 LED bulbs to form a seamless, curved, one-piece design. The headlight assembly is covered with nine layers of coating material, concealing radar, cameras, and other sensing systems.

The integrated 3D light grid contains 144 LED beads, giving the retro design a touch of light. Through these light guides and light coupling strips, and with the help of an air cushion optical structure, the light from the LED beads is directed into depth, creating a profound and three-dimensional light effect.

The overall side profile retains the classic BMW X's robust and sporty proportions. The innovatively upgraded Hofmeister kink incorporates ultra-thin titanium oxide stripes using a screen-printed process, creating a "reflective on the outside, transparent on the inside" effect, evolving into a "mirror-like kink." The rear taillights feature an infinitely gradient visual depth, echoing the speed-inspired design of the headlights.

As the first new-generation model designed specifically for the Chinese market, the new-generation BMW iX3 is 4885 mm long, 1895 mm wide, and 1635 mm high, with a wheelbase of 3005 mm. Compared to the overseas standard wheelbase version, it is 108 mm longer, and all the extra space is added to the second row, making it even longer than the wheelbase of the standard wheelbase BMW X5.

The new "battery-body integration" provides greater vertical interior space, and the rear seats have been further enhanced for the Chinese market. Compared to the overseas standard wheelbase version, the long wheelbase version of the new generation BMW iX3 features softer foam in the seat cushions and thicker seat covers, with a total thickness increase of more than 20 mm; the seat cushion length is more than 40 mm longer than the standard wheelbase version; the rear seat backrests are adjustable, with a maximum horizontal angle of 121°, providing a more comfortable riding experience for the entire vehicle.

BMW also specifically mentioned the design of the exterior door handles: the new semi-concealed door handles, which comply with the new national standards, have a high-strength mechanical structure that can withstand full-force pulling operations. All door handles are equipped with a mechanical redundancy structure. If the vehicle loses power, the door can be opened through the mechanical structure of the door handle; if the power system fails, the door handle can still be activated mechanically from the outside; the 12-volt backup battery is located in the central area of ​​the vehicle and is fully protected.

In terms of handling, BMW defines the new generation BMW iX3 long wheelbase version as "the best-driving BMW pure electric vehicle to date." With the "Heart of Joy" driving system, shock absorbers with hydraulic rebound damping and high preload anti-roll bar bushings, an approximately 50:50 axle load distribution, and a rear-biased torque distribution logic, this car can achieve a braking distance reduction of 4 meters on bumpy roads at 100 km/h and a braking distance reduction of 15 meters at 200 km/h. Moreover, in 98% of driving scenarios, mechanical brake intervention is not required, and deceleration and stopping are achieved through kinetic energy recovery, resulting in imperceptible braking.

The Heart of Joy super brain can perform 1,000 dynamic calculations per second. By eliminating communication delays across control units, it achieves a seamless integration of motor and chassis control, thereby improving driving performance.

The new-generation BMW iX3 long-wheelbase version's electric drive system adopts a full-range 800V high-voltage architecture, capable of replenishing 400 kilometers of range in 10 minutes (CLTC cycle) and charging from 10% to 80% SoC in 21 minutes. With its impressive overall range, the new-generation BMW iX3 long-wheelbase version becomes the first luxury brand SUV with a range exceeding 900 kilometers (CLTC cycle).

To serve the "driving control" characteristics of the new generation BMW iX3 long wheelbase version, the BMW Panoramic iDrive interactive system, built with the concept of "visual cone", is also entirely geared towards driving behavior, in order to help the driver "keep their hands on the steering wheel and their eyes on the road" and maintain safe driving control at all times.

This BMW iDrive panoramic infotainment system consists of four brand-new interactive interfaces:

The industry-first eye-level panoramic display uses 3D panoramic projection technology to achieve a 1.2-meter-wide display, an approximately 43-inch projection screen, 4K retina-level ultra-high definition, and a 100,000:1 contrast ratio. Combined with specially supplied nano-coated windshield glass, the P-light reflectivity reaches twice the industry level, making the reflected image clearer and the colors more saturated. Even if the driver and passengers are wearing polarized sunglasses, they can still clearly see all the content of the eye-level panoramic display.

The 3D head-up display uses industry-leading oblique projection technology. In oblique projection mode, the virtual imaging distance ranges from 4 meters to 9 meters. The head-up display image presents a continuous focal plane effect within this distance, thereby achieving a blended image of the ground scene.

The industry's first longitudinally spoked sports steering wheel integrates physical buttons and capacitive touch feedback, supporting precise blind operation and reducing the risk of distraction.

The center console features a 17.9-inch LCD screen. Compared to the integrated curved screen used in previous BMW models, the distance between the super-sensing center console screen and the driver's line of sight is reduced by 15%, and the operating distance is reduced by 50 mm. The "freely cut" design, which is 17.5° biased towards the driver's side, is based on ergonomics and avoids the arm from obstructing any corner of the center console screen, making it easier and safer to operate while driving.

What's striking about this new generation BMW iX3 long-wheelbase version is its unconventional approach. While intelligent driving and smart technology are becoming the focus of automakers' launch events, BMW still emphasizes driving pleasure and safety. This is something a long-established automaker with over 140 years of history needs to uphold. Even if, many years from now, human driving is largely replaced by machines, BMW still hopes that when someone needs a driver's car, BMW will be the first brand that comes to mind.

New generation BMW i3 long wheelbase version: 1000 km range, 3-meter wheelbase

If the design of the new generation BMW i3 long wheelbase version is a sight to behold for some and a disappointment for others, then the new generation BMW i3 long wheelbase version is much more elegant and restrained.

Of course, this restraint is relative, because even within the BMW 3 Series family, the new generation BMW i3 long wheelbase version looks like a rather avant-garde product.

The pioneering aspect lies in the fusion of its iconic double-kidney grille with the reimagined classic quad headlights, forming a striking "digital angel wing" design. Furthermore, the hood's peak-and-valve design creates a shark-nose-like appearance, enhancing the sense of forward motion and highlighting BMW's signature sporty DNA.

At the rear, the wide body stance and horizontally arranged light blade taillights widen the visual width of the rear, making the driving experience more prominent.

To further emphasize this driving experience, BMW has also installed a unique illuminated M logo on the C-pillar of this car. In addition, the new generation BMW i3 long-wheelbase version is one of the few 3 Series models with a wheelbase exceeding 3 meters, which is obviously to meet the domestic demand for space and to compete with rivals in the same class.

As a model sharing the same origin as the new generation BMW iX3 long wheelbase version, we can see many similarities between the two, such as the all-domain 800V architecture and the ability to recharge 400 kilometers in 10 minutes. However, because sedans have a smaller overall vehicle weight and lower wind resistance, the new generation BMW i3 long wheelbase version can achieve a CLTC range of over 1,000 kilometers.

The BMW iDrive surround view system and the Heart of Joy driving super brain are also used in the new generation BMW i3 long wheelbase version to truly help drivers drive well and enjoy driving.

The all-new BMW 7 Series: Little Change, Yet Significant Changes

The design of the BMW 7 Series and BMW 5 Series has been quite controversial in recent years. The design language has gradually evolved from elegance to flamboyance and domineering. The design concept of the new BMW 7 Series continues to highlight the power and authority of the flagship aura, showing a very heavy feeling.

Specifically, the double kidney grille enhances the vertical expression, creating an upward sense of power and a forward-pushing momentum; the horizontally extended headlights and the main light module that has been moved down and integrated into the air intake structure widen the visual proportions of the front of the car, while making the visual center of gravity closer to the ground, creating a stable and substantial flagship posture.

In terms of overall design, the new BMW 7 Series establishes a sense of volume through proportion control, weakens unnecessary dividing lines and decorative elements, and the horizontally extended taillights are simple and orderly, which also emphasizes the overall unity and flagship feel.

In terms of details, the grille has evolved into a complete and continuous double-kidney contour light strip, and the front crystal daytime running lights are embedded with 24 crystal units that support dynamic lighting. When turned off, they blend into the overall structure, making the overall lighting effect more dazzling.

Building upon the BMW iDrive surround view system, the new BMW 7 Series introduces a passenger-side entertainment screen for the first time. Its layered, staggered layout minimizes distractions for the driver, ensuring focused attention. This creates a three-screen联动 (interconnected/coordinated) layout with the center console and rear floating screens, providing screen freedom for everyone.

In terms of sound and light, the new BMW 7 Series features a Diamond 4D sound system crafted by Bowers & Wilkins, supporting Dolby Atmos technology. It boasts 36 speakers (35 in the i7), including 4 roof-mounted 3D speakers, 8 headrest speakers, and 8 seat vibration units, delivering a net power output of 1925W. Regarding lighting effects, the light-reflecting trim blends seamlessly into the overall design when unlit. When illuminated, inspired by the penetration and flow of light, it releases a ripple-like 3D light and shadow effect, complementing the rear diffused light source and the panoramic sunroof to create a rhythmic and layered light and shadow experience.

In terms of driving experience, BMW hopes that people will forget "driving a BMW and riding in a Mercedes" and instead "driving a BMW and riding in a BMW." With the Magic Carpet intelligent air suspension system and integral active steering as the core, the vehicle maintains stable and consistent dynamic feedback in different road conditions and driving states.

The new generation of BMW 7 Series still offers both gasoline and pure electric powertrains, with the pure electric i7 boasting a range of nearly 800 kilometers under CLTC conditions.

Of course, while all three new models emphasize classic BMW characteristics like driving dynamics, safety, and style, they also incorporate intelligent driving features. All three models will be equipped with a Level 2 autonomous driving assistance system, developed in partnership with the experienced Momenta. This system is applicable nationwide, suitable for urban, highway, and parking scenarios. During its development, this system comprehensively considered the synergy between human-machine interaction, driving control, and driver assistance, ensuring the driver remains focused on the road. In other words, BMW maintains a restrained approach to intelligent driving, not wanting drivers to relinquish all decision-making and driving control to machines.

BMW, now officially entering a new generation, faces a competitive landscape that is vastly different from the past. The time it took for the new generation of design elements and the BMW iDrive infotainment system to go from unveiling to mass production is unimaginable in China. This difference in pace reflects a difference in thinking.

For example, BMW will never forget what "driving" means, and responding to the design of the 1961 BMW 1500 sedan also means that the car company is paying tribute to its history. In more than a hundred years of history, there is endless wealth, but also considerable burden.

#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.

Tim Cook appeared at an employee meeting: "I'm in good health," and plans to serve as Apple's executive chairman for the long term.

According to Bloomberg, Tim Cook, who will step down as CEO this September, appeared at an Apple all-hands meeting on April 21 local time. During the meeting, Cook addressed speculation about his health.

I am thrilled to continue serving at Apple as Executive Chairman… I am in good health and full of energy, and I plan to remain in this new role for the long term.

He emphasized that he would support Ternus, who will take over as CEO in September, "in any way necessary," providing knowledge and experience at any time, and serving as "someone with whom to exchange ideas."

Cook also stated that "there can only be one CEO at a time," and that the new role of executive chairman will allow him to focus on strengthening Apple's global relationships.

When asked why he chose to step down at this time, Cook said he wanted to achieve "the best transition in history" and listed three prerequisites: the company's business must be doing well, the product roadmap must be excellent, and Tenus must be ready to take over.

He pointed out that Apple currently has an "impressive" product line and achieved its best quarterly results ever during the recent holiday season.

These three things were intertwined before, and they are intertwined again now, so now is the right time.

Ternus stated at the meeting that he was particularly excited to take over as CEO at this moment because Apple is "about to change the world again." He pledged that AI will create virtually limitless potential for Apple products and services, while emphasizing that design will continue to be at the heart of everything the company does.

As a company, our essence will not change, and our mission will not change.

Ternus, 51, graduated from the University of Pennsylvania in 1997 and joined Apple in 2001. He has long been involved in hardware engineering, working on core product lines such as AirPods, Mac, iPad, and iPhone, and was promoted to Senior Vice President of Hardware Engineering in 2021.

Related reading: Apple's new CEO holds employee meeting: We want to build AI hardware, retain top talent, and change the world.

OpenAI has taken GPT-5.5, a cutting-edge AI agent programming model, offline due to an unexpected leak.

Yesterday, developer "DavidAGMM" posted on Reddit that when opening Codex, a large list of unreleased models unexpectedly appeared in the dropdown menu. Based on leaked screenshots and videos, the leak involves the following models:

• Both GPT-5.5 and oai-2.1 are labeled as "the most cutting-edge AI agent programming models";
• Arcanine is described as "a cutting-edge model of a legendary appetite for starchy foods";
• glacier-alpha is labeled as "the wisdom that shakes the continent," while glacier-alpha-block-cy3 is labeled as "cold wisdom."
• In addition, there is Heisenberg, described as a "latest-edge life science research model".

However, when submitting a request after selecting GPT-5.5, Codex returns a 400 error, indicating that it is not supported. OpenAI has quickly fixed the issue after discovering it, and the relevant model entry point has been removed from the Codex interface.

Related reading: Claude's account was suspended, traffic was limited, and benefits were cut; OpenAI seized the opportunity with Codex to catch up.

Reports indicate that Tencent and Alibaba are in talks to invest in DeepSeek, valuing the company at over $20 billion.

According to Reuters, Tencent Holdings and Alibaba Group are in talks to invest in AI startup DeepSeek, with a target valuation of over $20 billion in this round of funding.

The report, citing four sources familiar with the matter, stated that DeepSeek had previously launched its first round of external financing, with an initial target of raising at least $300 million and achieving a valuation of at least $10 billion. However, due to strong interest from potential investors, the valuation target has been significantly increased.

However, the report emphasizes that the financing negotiations are still ongoing, and the final financing amount and valuation are subject to change.

In addition, several developers reported yesterday in the DeepSeek official communication group and on social media that the context window of the model called by the DeepSeek official API has been expanded from 128k to 1 million tokens, and the knowledge expiration date has been updated from 2024 to May 2025.

However, the model currently does not support visual input, only text and speech processing, and is still a non-multimodal model.

Ultraman speaks for the first time about the attack on his home: He said the night before, "At least no one wanted to kill me."

Recently, OpenAI CEO Sam Altman publicly responded for the first time to the series of attacks on his home on the Core Memory podcast.

Altman revealed that the night before the attack, several colleagues had dinner at his home. During the meal, they discussed the recent barrage of negative media coverage, and someone comforted him, saying that although the pressure was immense, "it's a good thing in a way." Altman casually replied, "At least nobody wants to kill me."

Ironically, the very next day, Molotov cocktails were dropped on his doorstep. He called it "a point of reference"—that what people say online, as long as it's just talk, isn't a big deal. He didn't go into details of the attack, only saying "there will be more similar incidents," before turning the conversation back to work.

Altman also revealed for the first time on the show how negotiations with Musk broke down. Altman said that in the early days of OpenAI, he, Brockman, Ilya Sutskever, and Musk, the four core members, all agreed that the company needed to transform into a for-profit entity.

Negotiations were close to reaching an agreement: Musk demanded a majority stake, which they conceded; Musk also demanded to become CEO and "let everyone know he's in charge," which they also agreed to.

Ultimately, however, Musk demanded "absolute control" over OpenAI, and negotiations broke down. Altman stated, "No one person should control the entire future, no matter who that person is." The lawsuit was originally scheduled to go to trial on April 27, with Musk seeking between $79 billion and $134 billion in damages.

Tesla's Chinese in-vehicle infotainment system will be integrated with Doubao and DeepSeek.

According to CLS News Agency, Tesla China's in-vehicle voice service will soon be integrated with the Doubao big data model and the DeepSeek model.

Sources familiar with the matter revealed that both models are connected through the Volcano Engine. The Doubao model is responsible for voice command functions, covering navigation settings, media playback control, air conditioning temperature adjustment, and owner's manual lookup; while the DeepSeek model provides AI interactive services.

According to information disclosed by the Shanghai Cyberspace Administration, Tesla (Shanghai) Co., Ltd. completed the generative AI filing for its "Tesla In-Vehicle Voice Large Model Service" on April 20, with the filing number being Shanghai-TeslaCheJiYuYinDaMoXingFuWu-202603020128.

Zhao Lichen, a genius from Huawei's Class of 2019, left the company to start his own business, betting on an embodied intelligent operating system.

According to AI Technology Review, Zhao Lichen, a former Huawei genius and a 2019-level technical expert, left the company in March of this year to join Hangzhou Lagrange Embodied Technology Co., Ltd., where he is transitioning to entrepreneurship and focusing on the development of embodied intelligent operating systems.

Zhao Lichen received his undergraduate and master's degrees in Software Engineering from Beihang University. He has long been deeply involved in the fields of 3D vision and multimodal computing, and his academic achievements have been cited more than a thousand times. He is the only student from Beihang University to date to be selected for Huawei's Genius Youth Program.

After joining Huawei in 2023 as a graduate student, he led the continuous iteration of the AI ​​Smart Home Treasure AI intelligent agent edge-cloud architecture. The related project won the China Telecom Outstanding Innovation Award and is scheduled to enter mass production in the second half of this year.

With his outstanding technical and delivery capabilities, Zhao Lichen was promoted to level 19 in just over two years, becoming one of Huawei's youngest senior technical executives.

The U9 will be featured in Gran Turismo 7, marking the first time a Chinese supercar has appeared on the market.

PlayStation officially announced yesterday that the BYD Yangwang U9 will be officially launched in "Gran Turismo 7" at 14:00 today, becoming the first Chinese supercar to be featured in the game.

The model launched this time is rumored to be the high-performance version of the Yangwang U9 Xtreme – the track/extreme version of the Yangwang U9, equipped with a four-motor system, with a combined horsepower of approximately 3019 hp and a top speed of over 496 km/h. It once set a lap record for production electric vehicles at the Nürburgring Nordschleife with a time of 6:59.157 and was listed on the official Nürburgring Nordschleife lap time list.

Previously, the Xiaomi SU7 Ultra had already been launched in the game, becoming the first Chinese brand model to enter the GT series.

All 286 delivered vehicles have been recalled. Sales of Zhang Xue 820RR motorcycles have been suspended for rectification due to a potential oil leak hazard.

Yesterday, Zhangxue Motorcycle officially issued a recall notice for its 820RR model, announcing a comprehensive recall and rectification. The recall is reportedly due to a crankcase rupture incident that occurred in Yinchuan on April 16th.

The damaged vehicle was returned to the factory on April 21. After disassembly and inspection, the factory confirmed that the root cause of the failure was the incorrect assembly of the O-ring seal of the pressure relief valve. After the vehicle had been used for a period of time, the pressure relief valve became loose, which led to oil pressure leakage, resulting in insufficient lubrication and ultimately causing the connecting rod to break.

Currently, a total of 286 820RR motorcycles have been delivered. Zhang Xue Motorcycles has instructed all owners to immediately cease using this model and contact their authorized dealerships to arrange for towing, with the dealerships responsible for implementing the necessary rectification.

Based on the actual condition of the vehicle, Zhang Xue Locomotive provided two rectification plans:

• Owners whose vehicle pressure relief valve has become loose or has excessive play can choose to return or exchange their vehicle.
• Owners whose pressure relief valve is not loose and has no play, and who only need to replace the O-ring, will receive two free maintenance services as compensation.

Tesla reported first-quarter revenue of $22.4 billion, with six factories under construction simultaneously, and is expected to burn through over $25 billion this year.

According to CNBC, Tesla released its Q1 2026 earnings report after the market closed yesterday:

• Revenue was $22.4 billion, slightly below the average analyst estimate of $22.6 billion from LSEG;
• The overall gross profit margin rose to 21.1%, an increase of approximately 4.78 percentage points compared to the same period last year;
• The gross margin of the automotive business (excluding carbon credits) rebounded to 19.2%, the highest in recent quarters;
• In the first quarter, 358,023 vehicles were delivered, an increase of about 6.3% year-on-year, but lower than Wall Street's expectations, and inventory backlog exceeded 50,000 vehicles;
• The energy storage deployment volume in the energy business decreased by 38% quarter-on-quarter to 8.8 GWh, which was also lower than analysts' expectations.

On the capital expenditure front, CFO Vaibhav Taneja raised the full-year capital expenditure forecast for this year from $20 billion to over $25 billion , a significant increase from $8.6 billion in 2025, and explicitly stated that there will be negative free cash flow for the remainder of this year.

Tesla is simultaneously advancing the construction of six factories and making large-scale investments in projects such as AI infrastructure, Cybercab, Optimus, and Megapack 3.

CEO Elon Musk stated that Cybercab has begun production, Semi is about to go into production, and the Optimus humanoid robot is expected to start mass production at the Fremont factory later this year, but all three products are in the early stages of ramping up.

In addition, he revealed that the AI5 chip has completed tape-out and will be prioritized for deployment in Optimus and data centers. The company also announced plans to build a research-oriented wafer fab in Giga Texas, with an initial investment of approximately $3 billion, in collaboration with Intel using the 14A process. The initial phase of the large-scale Terafab project will be handled by SpaceX.

In terms of autonomous driving, the number of paid FSD users has reached approximately 1.3 million, and Robotaxi services have expanded from Austin to Dallas and Houston, with plans to enter more than ten states this year. The opening of unsupervised FSD to consumer fleets is expected to proceed gradually in the fourth quarter.

Claude Pro subscription benefits have been cut; official explanation: it was a small-scale test.

Yesterday, Anthropic quietly removed Claude Code from the Claude Pro subscription benefits page without issuing any official announcement, sparking strong dissatisfaction from the developer community.

The latest pricing page on Claude's official website shows that the Claude Code item in the Pro package has been marked with a red cross.

After the incident was exposed, Amol Avasare, Head of Growth at Anthropic, responded on X, stating that the move was merely a "small-scale test" targeting approximately 2% of new users, aimed at exploring "more sustainable service models," and emphasized that the benefits of existing Pro and Max subscribers were completely unaffected .

Currently, Anthropic offers the Claude Pro plan for $20 per month and the Claude Max plan starting at $100 per month. If Claude Code is eventually removed from the Pro plan, developers will have to upgrade to Max 5x or a higher tier, increasing the monthly fee to at least $100.

GLM Coding Plan's unlimited weekly plan is being phased out this month; existing users will receive two months' compensation.

Yesterday, the Zhipu Open Platform officially announced that the old GLM Coding Plan (without weekly limit) will stop automatically renewing from 10:00 on April 30th. The price and benefits within the current subscription period will not be affected.

The official statement said that as the usage of GLM Coding Plan continues to expand, the original supply method of the old plan can no longer support long-term stable service, and this adjustment is inevitable. The affected users are: users who currently subscribe to the old plan (without weekly limit) and have enabled automatic renewal.

As compensation, GLM Coding Plan will automatically issue two months of the same tier of benefits to affected users on April 30th. Users who wish to continue using the GLM Coding Plan can subscribe to the latest online plan available after their current plan expires.

Neta Auto loses 18.3 billion yuan in 3 years of car manufacturing.

According to CCTV's "Focus Interview" report, Hozon New Energy, the parent company of Nezha Auto, accumulated a net loss of 18.3 billion yuan from 2021 to 2023, with an average loss of more than 80,000 yuan for every car sold.

In 2022, Hozon Auto's annual sales reached 152,100 vehicles, making it the sales champion among emerging electric vehicle manufacturers. However, misaligned product positioning led to a continuous decline in sales, and the company officially ceased production in November 2024. In the second half of 2025, Hozon New Energy was petitioned for "bankruptcy reorganization" by its creditors.

Meanwhile, state-owned enterprises in many places have become embroiled in these issues. One state-owned company in Yichun has invested a total of 1.42 billion yuan, with at least 800 million yuan unrecoverable; Nanning also funded the purchase of land and construction of factories for it. CCTV pointed out that some local preferential policies for attracting investment violate or indirectly violate relevant policies and regulations.

In March of this year, founder Fang Yunzhou and former CEO Zhang Yong were both listed as dishonest persons subject to enforcement, with Fang Yunzhou being restricted from high-level consumption hundreds of times. Currently, Hozon New Energy has begun soliciting potential investors from the public, and the company's official seal, licenses, and bank U-shields have been transferred to the provisional administrator.

Li Feifei's team's latest research reveals a fatal flaw in multimodal AI: it can still "see" things clearly even without images.

A recent paper published by Fei-Fei Li's team at Stanford University reveals a systemic flaw in current mainstream multimodal AI: even without receiving any images, cutting-edge models such as GPT-5, Gemini 3 Pro, and Claude Opus 4.5 still "confidently" describe image details and provide diagnostic conclusions.

Researchers have named this phenomenon "Mirage Reasoning".

The team built a test set called Phantom-0, removing all the images from the 200 questions that required visual representation to answer, without informing the model of this. The results showed that all tested models "described" an image that didn't exist in over 60% of the cases .

If common evaluation prompts are included, this rate even soars to 90%-100%. On six major multimodal benchmark tests, the model can still retain 70%-80% of the original score on average in "image-free mode", which means that **the actual contribution of the image itself to the final score may only be 20%-30%**.

Even more impressively, the team trained a small, purely text-based model with only 3 billion parameters using Qwen-2.5, which had never seen any images. On the chest X-ray question-answering benchmark, it not only outperformed all multimodal large models but also surpassed the average level of human radiologists by more than 10 percentage points.

This flaw is particularly dangerous in medical settings: when an image upload fails, the model does not report an error, but instead directly outputs a professionally worded diagnostic report, with the content systematically biased towards severe illnesses requiring emergency treatment, such as myocardial infarction and melanoma.

To address this vulnerability, the team proposed the B-Clean cleaning framework, which identified 74% to 77% of the questions in the three authoritative benchmarks as "answerable without looking at the picture" and removed them. After cleaning, the scores of each model dropped significantly, and two-thirds of the benchmarks saw a reversal in ranking.

Full text of the paper: arxiv.org/abs/2603.21687

HarmonyOS-enabled vehicles launch five models simultaneously, with the all-new Wenjie M9 starting at a pre-sale price of 499,800 yuan.

HarmonyOS held its spring new product launch event last night at the Greater Bay Area Cultural and Sports Center in Guangzhou, showcasing five new vehicles covering various models including coupes, shooting brakes, SUVs, and MPVs.

• The Shangjie Z7 and Z7T are priced from RMB 219,800 to RMB 299,800 and RMB 229,800 to RMB 309,800 respectively. Both models come standard with a new generation 896-line dual-beam-path image-level LiDAR and Huawei Qiankun Intelligent Driving ADS 4.1. SAIC Group President Jia Jianxu revealed at the launch event that pre-orders for the two new models have exceeded 80,000 units.
• The Wenjie M6 is positioned as a mid-to-large SUV, with the pure electric version starting at 279,800 yuan and the range-extended version starting at 259,800 yuan. A research report from Huachuang Securities shows that the Wenjie M6 received over 60,000 orders within 24 hours of opening pre-orders, and is expected to achieve stable monthly sales of 15,000 units.
• Zhijie V9: The first MPV under HarmonyOS, and also the first new model launched by the Zhijie brand after entering the 2.0 era. It is equipped with Huawei Turing MPV customized platform and pre-orders opened last night.
• The all-new M9 and M9 Ultimate Extended Edition: the standard version starts at 499,800 yuan, and the Ultimate Extended Edition starts at 669,800 yuan; equipped with the "Universe" triple screen (dual 17.2-inch + dual 3.4K resolution), HUAWEI SOUND Ultimate sound system (43 speakers, 2920W amplifier); and featuring an evolved HarmonyOS cockpit.

At the press conference, Yu Chengdong, Executive Director of Huawei and Chairman of the Consumer Business Group, announced that the total delivery volume of HarmonyOS has exceeded 1.35 million vehicles, with 112,700 vehicles delivered in the first quarter of this year, representing a year-on-year increase of 41.9%. He also revealed that HarmonyOS currently has 1,835 sales outlets, which is expected to expand to more than 2,459 by the end of the year.

It's worth mentioning that a minor incident occurred during the press conference: when the presentation of the Shangjie Z7/Z7T was completed and the segment switched to the Wenjie M6, a Shangjie vehicle remained stationary on the stage, sparking speculation among viewers. HarmonyOS later responded, stating that this was to allow the audience to experience multiple products simultaneously and was not a vehicle malfunction; they asked everyone "not to over-interpret or spread misinformation."

Xiaomi MiMo-V2.5 series begins public beta testing: flagship version targets Claude Opus 4.6

Xiaomi officially announced yesterday that the Xiaomi MiMo-V2.5 series of large models has started public beta testing, covering four products: MiMo-V2.5, MiMo-V2.5-Pro, MiMo-V2.5-TTS Series, and MiMo-V2.5-ASR. It also announced that two main models will soon be open sourced globally.

MiMo-V2.5-Pro ​​(Flagship, Long-Range AI Agent):

• Compared to Claude Opus 4.6 and GPT-5.4, it can reliably complete long-term tasks with nearly a thousand tool calls in a single session;
• Completed the SysY compiler project for the Peking University "Compiler Principles" course in 4.3 hours, achieving a perfect score of 233/233 on the hidden test set;
• Build a video editor web application with features such as multi-track timeline and audio mixing in 11.5 hours (8192 lines of code);
• Compared to Kimi K2.6, it saves 42% of tokens under the same ClawEval benchmark.

MiMo-V2.5 (General-purpose, native multimodal AI agent):

• It natively supports multimodal input of images, audio, and video, and its Agent capabilities surpass those of the previous generation MiMo-V2-Pro.
• The API cost is reduced by about 50% compared to the previous generation, and it approaches the top closed-source models in evaluations such as VideoMME, CharXiv, and MMMU-Pro.
• Compared to Muse Spark, it saves 50% of tokens on the same ClawEval benchmark.

OpenAI releases enterprise-grade AI agent

Early this morning, OpenAI released ChatGPT workspace agents, designed for enterprise teams. Powered by Codex and running in the cloud, these agents can handle complex tasks and long-cycle workflows, and continue to run even when the user is offline.

The workspace AI agent is positioned as an evolved version of GPTs. Teams can simply describe their workflow in ChatGPT and then build a usable AI agent under system guidance.

The AI ​​agent supports cross-system context retrieval and can perform operations such as updating work orders, creating documents, and sending messages after authorization. It has been integrated with ChatGPT and Slack.

Currently, the Workspace AI Agent is available in a research preview version to ChatGPT Business, Enterprise, Edu, and Teachers program users. It is free until May 6, after which it will switch to a credit-based billing model.

Volcano Ark Coding Plan launched GLM-5.1, enabling continuous operation for 8 hours on long-duration missions.

Volcano Engine announced yesterday that the Volcano Ark Coding Plan has officially launched GLM-5.1, matching the original manufacturer's full capabilities. GLM-5.1 performs comparably to Claude Opus 4.6 in multiple software development benchmark tests, and can continuously execute up to 8 hours of work at a time for long-duration agent tasks.

At the same time, ArkClaw launched the Hermes Agent plugin, which allows users to obtain an AI agent with self-evolution capabilities simply by clicking "Activate Now" on the welcome page, without the need for manual environment configuration.

In terms of architecture, ArkClaw acts as the "control center" responsible for planning the process, while Hermes acts as the "executor" in an independent container to launch the terminal, control the browser, and run scripts. The execution results are synchronously stored as ArkClaw's Memory and Skills.

Starting at 215,900 yuan, the Avita 06T is launched.

Yesterday, Avita officially launched its new 06T model, with an official price ranging from 219,900 yuan to 279,900 yuan. Five models are available, offering both pure electric and range-extended powertrains, and deliveries will begin immediately upon launch.

• All models come standard with Huawei Qiankun ADS assisted driving, equipped with an 896-line image-level LiDAR, and will be among the first models to be equipped with Huawei Qiankun ADS 5; the Elite version is equipped with a Limera in-cabin laser vision sensor and supports subscription to urban NCA function;
• It features the HarmonySpace 5 cockpit powered by HarmonyOS, incorporates the MoLA hybrid large-scale AI intelligent agent architecture, and is equipped with a 35.4-inch 4K panoramic star ring screen;
• All models come standard with Taihang distributed electric drive, with the fastest 0-100 km/h acceleration in 2.78 seconds, supports active and independent adjustment of left and right torque, and has chassis lift passage function;
• Both pure electric and range-extended electric powertrains support 5C supercharging, with a wheelbase of 2940mm;
• Equipped with Huawei Qiankun CAS 4.0 all-dimensional collision avoidance system, all models come standard with CATL batteries, with a cell defect rate as low as one in a billion.

Roewe and Volcano Engine jointly developed the AI-native automotive architecture "CPP".

Recently, at its 20th anniversary product launch event, Roewe officially released the world's first AI-native car series, "Jiayue," in deep collaboration with Volcano Engine, with three concept cars making their debut.

Unlike the traditional "car + AI" model, the "Jiayue" incorporates AI into the vehicle architecture from the very beginning of its design. Based on Roewe's SOA atomic scheduling capabilities, the Doubao large model is deeply integrated into the entire vehicle, forming a complete closed loop from understanding requirements to execution.

Roewe simultaneously released the industry's first "CPP" AI native technology architecture, covering three levels: information acquisition and memory management (Context), intelligent planning and decision-making (Planner), and atomic-level precise execution (Pixel).

The three concept cars are Jiayue 07, Jiayue 06 and Jiayue 09. The first production car, Jiayue 07, will be unveiled in June this year.

Reports indicate that Ferrari's 1000-horsepower all-electric supercar will debut next month, priced at approximately 4.4 million yuan.

According to CLS News Agency, Ferrari's first all-electric model, the Luce, has been initially priced at €550,000 (approximately RMB 4.4 million) and will officially debut in Rome, Italy next month. The final price is still subject to a fluctuation of about 10%.

In terms of performance, the vehicle adopts a four-motor layout with a combined power of over 1,000 horsepower, a 0-100 km/h acceleration time of 2.5 seconds, a 122 kWh battery pack with a range of 530 kilometers, and an 800V architecture that supports fast charging to 80% in 15 minutes. The battery cells are supplied by SK On and have an energy density of 195 Wh/kg.

To compensate for the inherent lack of sound experience in pure electric vehicles, Ferrari has applied for two patents: an acoustic amplification device and a "virtual gearbox," to recreate the auditory and shifting feedback of an internal combustion engine.

The interior was designed by former Apple chief design officer Jony Ive, who explicitly opposed the design concept of large touchscreens, emphasizing that "everything is built on functionality."

Alibaba releases AI digital human "Qianwen Xiaojiuwo" (Thousand Questions Little Dimples)

Yesterday, Alibaba officially launched "Qianwen Xiaojiuwo," an ecosystem-level AI assistant digital avatar. Within the Qianwen app, users can interact with her to complete everyday tasks such as ordering food, buying tickets, hailing a ride, and planning trips, covering multiple high-frequency life scenarios.

Alibaba stated that "Qianwen Xiaojiuwo" will gradually appear in various applications within the Alibaba ecosystem, such as Taobao and Fliggy, becoming an AI life assistant that permeates both work and life scenarios.

Lark Multidimensional Tables launches AI Agent, enabling data analysis to be completed in a single sentence.

Yesterday, Lark Multisheet underwent a comprehensive upgrade to its AI capabilities, launching the AI ​​Agent feature, which covers multiple core scenarios such as data analysis, visualization, and form generation.

• AI-powered data analysis: Users can directly ask questions to the table in natural language, such as "What was the sales volume in East China in Q1?" or "Predict next month's performance." The system will return the analysis results instantly, without the need to manually filter data or write formulas.
• Custom page generation: AI Agent supports converting table data into a custom page with just one sentence, automatically generating charts and laying out information to make the data presentation more intuitive and clear;
• AI-generated questionnaires: Users only need to describe their needs, and the system can automatically generate forms and corresponding data tables, which are suitable for scenarios such as event registration, user surveys, and opinion collection.

Ant Financial's Ling-2.6-flash has been officially released, focusing on token efficiency.

Yesterday, Ant Financial officially released Ling-2.6-flash, with a total of 104B parameters and 7.4B activation parameters. It focuses on "Token Efficiency" and is positioned for efficient application scenarios.

It is worth mentioning that the anonymous test version of the model, "Elephant Alpha", was launched on OpenRouter a week ago. It has been ranked first on the Trending list for several consecutive days, with an average daily token call volume of 100B and a weekly growth of over 5000%.

• Token Efficiency: In the full Artificial Analysis test, the total consumption was only 15M tokens, which is about 1/10 of Nemotron-3-Super, while achieving an Intelligence Index of 26.
• Inference speed: Up to 340 tokens/s under 4-card H2O conditions; Output Speed: 215 tokens/s, ranking in the top tier among similar parameters.
• AI agent capabilities: Achieve state-of-the-art (SOTA) performance in the same size category on mainstream AI agent benchmarks such as BFCL-V4 and SWE-bench Verified;
• Pricing: $0.1 per million tokens input to the API, $0.3 per million output. A one-week free trial is available now.

Odyssey releases the most powerful world model 2 Max

Odyssey released its latest general-purpose world model, Odyssey-2 Max, yesterday. It has three times the number of parameters of its predecessor, Odyssey-2 Pro, and its training computation is increased tenfold.

Unlike video generation models such as Sora and Veo, Odyssey-2 Max generates frames based on real-time user actions, rather than pre-generating fixed videos. Users can interact with the simulated world in real-time, like playing a game, and the video can be generated continuously for over 120 seconds without crashing. In terms of physics simulation accuracy, Odyssey-2 Max outperforms its competitors.

• VBench 2 Physics score of 58.52, higher than Odyssey-2 Pro (49.67) and NVIDIA Cosmos-Predict2.5-14B (44.92).
• PAI-Bench Physics score of 93.02, also outperforming major competitors.

Odyssey was co-founded by Oliver Cameron (CEO) and Jeff Hawke (CTO), both with backgrounds in the autonomous driving industry, and had previously raised a total of $27 million in funding.

Alipay AI Pay integrates with the "Lobster" platform

Alipay announced yesterday that its AI Pay feature now supports "Lobster"-type AI agents, allowing users to directly complete payments, purchase tokens, and shop when using these AI agents.

Currently, Alibaba Cloud's JVS Claw platform has been the first to pre-install Alipay AI Pay and launch it online, while Ant Financial's DTClaw has also been launched simultaneously; other "lobster"-related AI agents such as Claude Code and Hermes Agent also support one-click installation and use of Alipay AI Pay.

After completing the installation according to the instructions on the official website, users only need to say "Enable Alipay payment function" to "Lobster" and follow the instructions to complete the authorization to enable it.

In terms of security mechanisms, Alipay AI Pay requires users to manually verify their identity when activating the service, and each payment requires the user's authorization and confirmation. It is also equipped with a 24/7 intelligent risk control system. Alipay also promises "You dare to pay, I dare to compensate."

The UK has passed a new law: those born after 2009 will be banned from buying cigarettes for life.

According to The Paper, the UK Parliament passed the Tobacco and E-cigarettes Bill yesterday, which stipulates that it is illegal for businesses to sell tobacco products to people born on or after January 1, 2009. This means that teenagers aged 17 and under will be banned from buying cigarettes for life.

British Health Secretary Wes Straying called the legislation a "historic moment for national health," saying the bill aims to create the "first generation of smoke-free people," protecting them from lifelong addiction and the harms of tobacco.

According to data from the UK's National Health Service, approximately 75,000 people die each year in England from smoking-related diseases, accounting for about a quarter of all deaths.

Once the bill comes into effect, the UK government will be authorized to extend the indoor smoking ban to public areas such as children's playgrounds, schools, and outside hospitals, while also restricting the flavors and packaging of e-cigarettes and banning e-cigarette use in places where smoking bans have already been implemented.

Xibei's "rebranding" to save itself: Opening 11 braised noodle restaurants in two months, with some receiving over 20% negative reviews.

According to Sina Finance, Jia Guolong, the founder of Xibei, is accelerating the transformation of his stores into a new brand, "Tianbian Casserole Braised Noodles".

Yesterday, Xibei's first store in Shanghai was renovated and rebranded. On the same day, two other stores in Shanghai also opened, bringing the brand's total number of stores nationwide to 11 within two months. Most of these stores are renovated from the original Xibei locations, with an average spending of about 50 yuan per person.

In terms of reputation, the first store to open in Beijing's 798 Art Zone only scored 3.9 on Dianping, with over 20% of consumers giving it a medium or lower rating, and most stores scoring below 4. User complaints focused on "not tasting like braised noodles" and "slow service," with some even joking that "the pot is bigger than your face, and the noodles are smaller than your hands."

At the beginning of this year, Jia Guolong disclosed that he expected to lose more than 600 million yuan and announced the closure of 102 stores nationwide, accounting for about 30% of the total. "Tianbian Casserole Noodles" is the brand that was "transferred" in the strategy of "closing some, transforming some, and retaining the core group".

The final trailer for "Cold War 1994," slated for release during the May Day holiday, has been released.

The movie "Cold War 1994" released its final trailer, titled "Open and Covert Struggles," yesterday and simultaneously launched nationwide pre-sales.

The film is set against the backdrop of the pivotal historical moment of Hong Kong's handover and Chief Executive election in 1994. The plot revolves around a kidnapping case of a wealthy businessman, as young men Cai Yuanqi and Li Wenbin, caught in the turmoil, take sides and test each other, gradually shifting from allies to adversaries. The film will be released nationwide on May 1st.

Top Secret Mission is set to release on June 19th.

According to Weibo Movies, the movie "Top Secret Mission" has officially been scheduled for release on June 19th, and a "Volunteer for Battle" trailer has been released.

The film focuses on an all-female counter-terrorism special forces team, telling the story of their top-secret mission to rescue hostages and eliminate terrorist threats, and their high-difficulty operations in a dangerous environment.

Michael Jackson: Road to Stardom opens with a 34% rating on Rotten Tomatoes, with critics offering mixed reviews.

According to blogger "Hollywood Watch," the embargo on reviews for the Michael Jackson biopic "Michael Jackson: Road to Stardom" has been lifted, with initial opinions showing a clear divide. The film will officially premiere this Friday.

As of yesterday afternoon, the film had a 34% freshness rating on Rotten Tomatoes (96 reviews, 33 positive, 63 negative) and a Metacritic score of 38 (35 reviews, 6 good, 17 average, 12 bad), indicating a generally low level of critical acclaim.

Variety magazine gave a relatively positive review, saying that although the film did not touch on the dark side of Michael Jackson, it was "well-acted, engaging, and a biographical film that most people can accept," and compared it to "an 80s TV movie with more vivid performances and better cinematography."

Screen Rant gave a more positive assessment, believing that the film "lives up to the legendary status of Michael's name, from casting to costumes to concert stage performances."

#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 the past few years, the narrative of the Chinese automotive market has been almost entirely dominated by new energy vehicles. Lower operating costs, enhanced intelligent experiences, and continuously evolving electric platforms have rendered traditional gasoline-powered vehicles somewhat obscured. However, the real needs of the market are never a simple choice. In terms of charging convenience, long-distance travel efficiency, and adaptability to various scenarios, gasoline-powered vehicles still maintain a stable user base.

On April 22nd, Chery Automobile held a "New Fuel Night" event in Beijing and officially launched the all-new Tiggo 9. The new car is available in four models, with official guide prices ranging from RMB 159,900 to RMB 179,900, and limited-time special offers ranging from RMB 146,900 to RMB 166,900. As the flagship product of the Tiggo family, the all-new Tiggo 9 continues its mid-size SUV positioning while undergoing systematic upgrades in core dimensions such as power, intelligence, comfort, and safety.

If we rewind five years, when buyers of a gasoline-powered SUV, their focus was mostly on the powertrain, interior space, chassis quality, and brand reputation. Today, however, market standards have been redefined by new energy vehicles. Consumers are accustomed to larger screens, faster infotainment systems, and more advanced driver assistance features, and they've become accustomed to considering comfort features as standard rather than optional. For gasoline-powered vehicles, relying solely on traditional mechanical advantages is no longer sufficient to create a competitive edge.

This is also the product logic behind the all-new Tiggo 9: retaining the mature advantages of gasoline vehicles in terms of efficiency, range, and usage scenarios, while making up for the shortcomings in intelligent experience and configuration experience as much as possible. It attempts to become a flagship gasoline SUV without a significant sense of being outdated.

For Chery, the powertrain has always been one of the brand's most important technological hallmarks, and this is also the focus of the Tiggo 9's upgrade.

The all-new Tiggo 9 is equipped with a Kunpeng 2.0T engine, delivering a maximum power of 192kW and a peak torque of 400N·m, paired with an 8-speed automatic transmission. Official information indicates that this 8-speed automatic transmission is a Chery-developed product, achieving a maximum mechanical efficiency of 96.6%.

Judging from the specifications, this powertrain is already above the mainstream level in its class. For a mid-size SUV primarily designed for family and long-distance travel, more abundant mid-to-high-range power reserves, high-speed cruising capabilities, and a more mature and stable mechanical structure remain the real value that many users care about.

In terms of chassis, the new car comes standard with the CDC suspension system across the entire range, and features a chassis control solution officially called the "Flying Fish Chassis." The value of CDC is not new, but making it standard across the entire range in this price range means that users do not have to be forced to upgrade to a higher-spec version for a better driving experience.

This is also an increasingly obvious trend among Chinese brands: features that were once considered high-end are being rapidly made available to the mainstream market.

If the powertrain and chassis are the foundation of the Tiggo 9 as a gasoline-powered vehicle, then its intelligent features are the part that it most needs to prove itself in.

The all-new Tiggo 9 features a 30-inch integrated screen and is equipped with a Qualcomm Snapdragon 8255 chip. According to official statements, the new car supports faster voice response, multitasking, and a smoother in-vehicle infotainment experience.

In today's market environment, a large screen is no longer a selling point, but the hardware platform and interactive experience still determine the user's daily experience. For many consumers, the ease of use of the in-car infotainment system can influence their car-buying decisions more frequently than a few seconds faster 0-100 km/h acceleration.

In terms of driver assistance, the new car offers the Falcon 500 driver assistance system, equipped with 26 sensors including cameras, millimeter-wave radar and ultrasonic radar, supporting functions such as highway navigation, lane centering, and parking assistance.

It should be noted that the actual experience of assisted driving capabilities still depends on software maturity, scenario adaptation, and long-term OTA performance; hardware parameters alone cannot tell the whole story. However, at least at the configuration level, the significant generational gap between gasoline vehicles and new energy vehicles that was once prevalent is gradually narrowing.

Despite the industry's enthusiasm for discussing computing power, models, and intelligent driving, for family users, the core value of a mid-size SUV remains the riding experience and sense of security.

The new vehicle is available in 5-seater and 7-seater versions, featuring a 13-layer seat structure. The front seats offer heating, ventilation, and massage functions, and include a "Queen's Seat" for the front passenger. Comfort features such as an audio system, ambient lighting, and healthy cabin materials are also included across the entire product line.

These types of configurations may not be the biggest talking point in marketing, but they often have the greatest impact on long-term user satisfaction. After all, most of the time family cars are used, not in high-speed NOA or extreme acceleration scenarios, but for commuting, picking up and dropping off passengers, weekend trips, and long-distance trips home.

In terms of safety, the all-new Tiggo 9 adopts a high-strength steel body structure, with high-strength steel accounting for 85% of the body. It offers 5-seat configurations with 8 airbags and 7-seat configurations with 10 airbags, and is also equipped with a 360-degree four-channel driving recorder.

In an increasingly competitive mainstream market, enhanced safety features are no longer just a bonus, but a prerequisite for success. Consumers are also becoming more aware that what truly determines a car's basic capabilities is often not screen size, but rather the engineering investments that are not readily apparent in everyday life but whose value only becomes apparent in critical moments.

With a limited-time offer price ranging from 146,900 yuan to 166,900 yuan, the all-new Tiggo 9 is targeting the most competitive and realistic price range for family SUVs.

This is a typical market segment where users want both: they want ample space and comprehensive features, as well as reliable power and controllable fuel consumption; they want a smart experience but are unwilling to bear excessively high purchase costs.

In this segment, the traditional advantages of joint venture brands are weakening, while Chinese brands continue to expand their market share with higher configuration density and faster product iteration speed. The Tiggo 9, as Chery's flagship model, enters this arena not only with sales targets but also with the responsibility of elevating the brand.

The automotive market is large enough, and user needs are diverse enough. Not all consumers are ready to enter the pure electric era, nor are all families willing to completely relinquish their driving habits to the new energy system. For these users, a gasoline-powered SUV with mature powertrain, complete features, and up-to-date intelligent experience still holds significant appeal.

#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.

With the official launch of GPT Image 2, the field of AI image generation now only has two tiers: GPT Image 2 and others.

In the large-scale model arena, GPT Image 2 leads by a wide margin with an absolute advantage of 241 points, but the development list is actually more interesting than the benchmark scores.

The core OpenAI team that developed this model consists of only 13 people.

Furthermore, Chinese people make up half of the team. If you look closely at their resumes, you'll find that many of them had already met at a university, a laboratory, or even a research summer camp in China before joining OpenAI.

The AI ​​community is basically a huge circle of acquaintances.

Silicon Valley tech giants come and go, but the mentor-apprentice relationship remains constant.

Chen Boyuan is an absolute core member of GPT Image 2, and his growth is an epitome of the "mentorship" model in Chinese academia.

During high school, Chen Boyuan participated in a research summer camp in Wuxi. At that time, he had not learned programming, and there he met Xia Fei, a Chinese scholar who later became a senior researcher at Google DeepMind. Xia Fei introduced him to deep learning and became his guide into the field of AI.

The two have remained in contact ever since. During his undergraduate studies, Chen Boyuan majored in Computer Science and Mathematics at UC Berkeley, entering the EECS Honors program with a GPA of 3.96. He studied under Pieter Abbeel and also founded a robotics education company in 2017, which operated until 2020.

During his first year as a PhD student at MIT, Chen Boyuan faced difficulties due to a lack of publications. Xia Fei provided crucial support, assisting him in publishing his first influential paper, NLMap. Xia Fei also invited him to intern at DeepMind twice. During his internship in 2023, he led the construction of a multimodal large language model data synthesis pipeline, and the instruction fine-tuning techniques he summarized were used in the development of Gemini 2.0.

With this background, Chen Boyuan joined OpenAI in June 2025. In addition, he is also a member of the Sora video generation team, holding multiple positions.

During his time at MIT, Chen Boyuan studied under Assistant Professor Vincent Sitzmann at the Computer Science and Artificial Intelligence Laboratory (CSAIL), focusing on world modeling. Kiwhan Song was a fellow student in the same laboratory, mentored by the same professor.

The core research focus of the Sitzmann Lab is "world modeling," which, simply put, involves enabling AI to predict changes in the physical world through mental simulators, rather than simply mimicking pixels. This research approach may have directly influenced the technological direction of GPT Image 2.

The two collaborated multiple times during their doctoral studies and jointly published two papers, "History-Guided Video Diffusion" and "Large Video Planner," which mainly explored how to combine diffusion models and sequence generation, allowing the model to clarify the temporal and spatial causal logic before generating content.

It's worth mentioning that Kiwhan Song is the creator of the "long-necked" sticker-style cartoon avatar.

Besides these two fellow students, the team also has two Chinese members who come from a network of Chinese researchers that has been built up over many years in the industry.

Jianfeng Wang worked at Microsoft for nearly nine years, focusing on large-scale multimodal representation learning as a principal researcher. During the development of DALL-E 3, he had in-depth collaborations with the OpenAI team. After joining OpenAI, he was primarily responsible for improving the model's instruction compliance capabilities and understanding of the world.

Bing Liang spent more than five years at Google, where he participated as a senior software engineer in the core R&D of Imagen 3, the Veo video model, and the Gemini multimodal series. He joined OpenAI last August to lead research on image generation.

They brought not only their own personal abilities, but also the engineering experience accumulated by their competitors over many years, as well as the pitfalls they had encountered, saving the team a lot of detours.

Provide a stage for geniuses

Weixin Liang and Yuguang Yang are another noteworthy pair in the team. Both graduated from the Chu Kochen Honors College of Zhejiang University and have the same undergraduate background.

Yuguang Yang's resume spans a wide range. He studied engineering at the Chu Kochen Honors College for his undergraduate degree, and then went to Johns Hopkins University for his doctoral studies in computational chemistry physics and machine learning. After obtaining his degree, he worked at Amazon Alexa on deep learning research in speech recognition, and then went to Microsoft Bing to be responsible for query understanding and large-scale retrieval.

He also conducted visiting research at Tsinghua University, focusing on reinforcement learning algorithms for nanorobot navigation in human blood vessels, during which he published seven peer-reviewed journal papers. This interdisciplinary background is directly evident in the GPT Image 2 release demonstration.

Compared to Yuguang Yang, Weixin Liang's career path is more academically oriented. He pursued his PhD at the Stanford AI Lab (SAIL), during which time he collaborated with several renowned professors, including Christopher Manning, Li Fei-Fei, and James Zou.

During his internship at Meta, he published the paper "Mixture-of-Transformers (MoT)," which introduced a hybrid expert model architecture with modal decoupling. It implemented modal-aware sparsity processing on every non-embedded parameter of the Transformer, including the feedforward network, attention matrix, and layer normalization, ultimately reducing the computational cost of multimodal pre-training by 66% and completing pre-training validation on a 30-byte parameter scale.

Multimodal models, which need to process both text and high-resolution images simultaneously, can easily experience exponentially increasing computational costs. MoT, however, effectively addresses this problem by efficiently allocating weights between different modalities during the pre-training phase through its modal decoupling attention mechanism. This research, later hailed as a "foundational contribution to unifying multimodal understanding and generation," has garnered significant attention in the field of multimodal modeling.

In recent years, graduates from Tsinghua University's Yao Class, Zhejiang University's Chu Kochen Honors College, the University of Science and Technology of China's Junior Class, Shanghai Jiao Tong University, and other institutions have become the core force of overseas AI labs such as OpenAI, Anthropic, DeepMind, and Meta.

In addition to the aforementioned members, the team also has several core researchers, each playing an important role:

Kenji Hata: Master of Computer Science from Stanford University, formerly worked at Google Research. After joining OpenAI, he participated in the development of multiple models, including 4o image generation (GPT-Image-1) and Sora 2, and is one of the team members with the most complete model iteration experience.

Ayaan Haque: Prior to joining OpenAI, he was a researcher at Luma AI, where he participated in training the video generation model Dream Machine. This experience gave him the ability to process high-dimensional temporal data. At OpenAI, he was primarily responsible for the development of GPT Image 2 and thinking patterns.

Dibya Bhattacharjee: Bachelor's and Master's degrees in Computer Science from Yale University, and nearly 5 years of experience at Google. She joined OpenAI in February 2024 to lead image generation research, demonstrating the model's multi-format generation capabilities at the launch event, and is a key member in making the model's output format "out of the box".

Mengchao Z.: Bachelor's degree from Shanghai Jiao Tong University, Master's degree from Texas A&M University, with a solid engineering background. Before joining OpenAI, he led the architecture design of large-scale recommendation systems, and is currently responsible for transforming model technology capabilities into usable product forms.

In addition, the identities of several other team members are currently unavailable.

The mentorship network fosters research taste, and the university background lays the foundation for basic understanding. This talent network woven from mentors, classmates, and former colleagues naturally forms a very high density of trust. Due to the high degree of unity in values ​​and engineering language, the cost of innovation is almost zero.

Simply poaching one or two Chen Boyuans with hefty sums is merely a stopgap measure. Without partners, it's difficult to replicate this spontaneous innovation. The organizational structure of large companies, with its clear hierarchy and objectives, is beneficial for commercialization, but it often creates constraints for basic research that requires free exploration.

The competition in AI is ultimately a story about "people." Rather than searching for the next Chen Boyuan, we need to build an ecosystem where people like Chen Boyuan can meet naturally in their local context and achieve mutual success.

#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.