Pogbagram

Early this morning, Ferrari finally unveiled the long-covered red cloth, revealing the interior design of their first all-electric GT – the Ferrari Luce.

At first glance, you might doubt whether this is a product of 2026. In current car design, "futuristic" usually means a continuous screen spanning the left and right sides, touch or pressure-sensitive buttons, and ubiquitous RGB ambient lighting.

But Ferrari's interior is so old-fashioned, it looks like a vintage car from the 1970s.

The three-spoke steering wheel has a similar feel to the Ferrari 308 GTB, and the three gauges in front still retain mechanical needles. Then look at the restrained size of the center console screen, which is only slightly larger than an iPad mini. It stands there all alone, without any intention of extending it to the passenger side. There are even a few levers below it.

But when you get closer, a strong "apple smell" hits you.

Those retro lines were all machined from a single block of aluminum alloy using modern CNC machine tools. The metal surface has undergone a delicate anodizing treatment, presenting a cool, matte texture. The button on the upper right of the central control screen is somewhat like the digital crown on an Apple Watch; each time it is pressed, the watch face will move.

From afar, it pays homage to the Enzo Ferrari era; up close, however, it is a showcase of industrial design prowess—this is the first work Ferrari has produced after its alliance with Jony Ive, former design director of Apple.

The person who eliminated the buttons brought them back.

If the name Jony Ive is unfamiliar, take a look at the equipment on your desk. This Briton has defined "the aesthetics of modern technology" over the past thirty years.

Starting with the colorful iMac G3, he freed computers from their beige plastic casings; then came the ingenious Click Wheel of the iPod, which revolutionized the way people listen to music; followed by the unibody MacBook, and the iPhone that completely eliminated the physical keyboard and taught the world how to "slide on glass," and even the current flat interface of iOS is his creation.

Simply put, his career at Apple is a history of constant subtraction: cutting out interfaces, hiding gaps, eliminating buttons, until the device becomes a pure, seamless black mirror.

But after leaving Apple, founding LoveFrom, and allying with Ferrari, Jony Ive, who once dedicated himself to eliminating buttons, has now become the most steadfast defender of physical objects.

Jony Ive believes that the appropriate interaction should be provided for the specific scenario.

On smartphones and tablets, functionality is limitless and applications are abundant, making a screen that can transform into any shape the most efficient form of interaction. However, in a cockpit, the best interaction is not navigating through secondary menus on the screen, but rather relying on muscle memory.

Ferrari CEO Benedetto Vigna is well aware of this as well. The executive, who has spent over 30 years in the tech industry, stated frankly:

Technology should stem from human emotion. Ferrari doesn't need touchscreens to emphasize its technological sophistication; the damping and tactile feedback provided by its mechanical structure is the greatest respect for the driver.

Focus your gaze on that three-spoke steering wheel, and you can intuitively feel the weight of that respect.

It's more than just a steering wheel; it's a modern sculpture suspended in the cockpit. The entire steering wheel is precision-machined from a single piece of aluminum, showcasing the unibody design technology of the MacBook to its fullest extent.

There are no unnecessary seams, no cheap plastic covers, only a cool metallic luster and a pure form.

Ferrari has incorporated interactive units known as "Analogue Control Modules" onto this metal frame.

Instead of those cool-looking but easily mis-tapped capacitive touchpads, they stuck with mechanical buttons, dials, and switches—just like other modern Ferraris, your thumb can easily find them during spirited driving.

The same logic extends to the startup process.

Most electric cars these days are "get on and go," efficient, but boring. Ferrari, on the other hand, wants to rediscover that sense of ritual, like "awakening a beast."

The Luce retains Ferrari's iconic key holder on the center console. When you insert it into the holder, the mechanical structure makes a crisp locking sound, the gear shift mechanism unlocks, and the vehicle officially enters standby mode.

Interestingly, Ferrari not only placed a small gear lever on the Luce's center console, but also installed two paddle shifters behind the steering wheel.

Actually, the Luce doesn't have a traditional multi-speed gearbox. The "gear shifting" here refers to a completely new manual torque control system:

When you pull the left paddle, the regenerative braking effect is enhanced, and the vehicle produces a dragging sensation similar to downshifting, shifting the center of gravity forward and making it easier to enter corners; pulling the right paddle releases the motor torque in stages, bringing a push-back feeling similar to upshifting.

To help you pinpoint the right time to shift gears, Ferrari has also designed a prominent torque indicator above the speedometer. When an upshift is needed, this light strip will flash continuously, just like the shift indicator light on an F1 car steering wheel.

This series of actions transforms your mindset from simply starting a household appliance to awakening a machine with a soul. This psychological suggestion is crucial for a Ferrari without the roar of an engine.

When Ferrari meets California Highway 1

As your gaze passes over the classic-inspired steering wheel, the dashboard design ahead is equally full of intriguing details.

Instead of using a simple, single LCD panel, a unique layered structure is employed. You'll see the superposition of a physical metal frame and a high-resolution screen, and the interplay of mechanical pointers and digital markings.

This design approach bears a strong personal touch from Marc Newson.

As another key figure at LoveFrom, Marc Newson's status in the industrial design world is no less than Jony Ive's. From the Lockheed Lounge chair that fetched a record price at auction to the Apple Watch he collaborated with Ive on, Newson has always been a master of "soft minimalism".

He is also a passionate watch enthusiast. The Ikepod watches he founded, with their rounded flying saucer shape and easy-to-read dials, overturned the stereotype of Swiss watchmaking.

Luce's dashboard does have a bit of an Ikepod feel to it, with rounded, clear fonts and no unnecessary decorations. Combined with old-fashioned physical pointers, scales, and an outer glass layer, you wouldn't even recognize it as a screen if it weren't for the interface changing.

It's worth mentioning the independent circular module on the far right, which Ferrari has named "Multigraph".

It looks like a standalone mechanical instrument, encased in a precision anodized aluminum housing and glass lens.

In everyday driving, it quietly serves as a clock or compass. But when you activate launch control via the physical lever above your head, this module instantly switches roles, becoming a dedicated 5-second countdown stopwatch.

Watching the physical pointer count down on the mechanical dial creates a sense of pressure and realism that is far more exhilarating than watching pixelated numbers jump on a screen.

Shift your gaze to the right and take a look at Luce's central control screen.

The screen is designed with restraint; it doesn't take up the entire dashboard, but rather floats in the center like a separate panel.

The screen is covered with high-strength glass and surrounded by an aluminum alloy frame. It even features a special handle that allows the driver and passenger to manually adjust the screen's tilt angle according to the lighting conditions.

If you want to operate the screen while the vehicle is moving, this handle can also act as a palm rest to prevent your fingers from missing the screen buttons due to bumps.

The most touching details are at the bottom of the screen.

In an era where even adjusting airflow direction requires navigating through three menus, the Luce surprisingly retains several levers for controlling basic functions such as temperature, fan speed, and media volume. The direct, mechanical feedback—without waiting for system animations—is indeed a costly luxury in today's automotive market.

While carefully studying the official demonstration video, we also discovered a little Easter egg on the car's infotainment screen:

On the media playback interface, the song being played is called "LoveFrom," and the artist is listed as "California." The album cover, however, is a California state flag with the words removed—the famous grizzly bear.

Shifting my gaze to the navigation map beside me, the route showed that the vehicle was traveling on the famous California Highway 1 (CA-1), approximately near Point Reyes, heading south towards San Francisco.

These are the personal signatures of Jony Ive and Marc Newson. These two designers, who have lived in California for thirty years, defined the soul of Ferrari Luce in this extremely subtle way.

These details piece together a true profile of the Luce. It's not the kind of hardcore lap-setting machine that makes drivers nauseous at Monza; it's a true GT (Grand Tourer).

It belongs to California's Sunshine Coast and is a long-haul cruise.

#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.

ifanr | Original Link · View Comments · Sina Weibo

This video aims to reveal the underlying logic behind Apple's user-friendly interface. We'll start with three unseen capabilities—telepathy, empathy, and the ability to create multiple devices—to break down why the Apple ecosystem becomes increasingly smooth with use. It's not a victory of specifications, but rather a prediction of real-life situations: assuming you'll experience frantic activity, motion sickness, lack of free hands, and even moments when you lose the ability to move.
From iOS's motion sickness prevention technology to the seamless interaction of AirPods and Apple Watch, and even overlooked hidden features like head-controlled pointers and Continuity, Apple is trying not to create stronger tools, but rather more natural extensions of the body.
In this video, we'll use a 60,000 RMB Apple ecosystem to talk about what true "magic" is—an experience you barely feel, yet can never go back to once you leave it.

#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.

ifanr | Original Link · View Comments · Sina Weibo

Market Supervision Bureau: Opposes manufacturers engaging in "in-depth competition" and selling cars at a loss.

On February 12, the State Administration for Market Regulation issued the "Guidelines for Compliance of Pricing Behavior in the Automobile Industry" to further regulate pricing behavior in the automobile industry and promote the healthy and orderly development of the automobile market.

According to Xinhua News Agency, the automotive industry is currently plagued by some irregularities. The guidelines, based on the current realities of the industry's development, further clarify the boundaries of acceptable behavior, unify regulatory rules, guide automotive manufacturers and sellers to operate legally and compliantly, and promote a market order characterized by high quality, fair pricing, and healthy competition.

This guide mainly includes:

• The guidelines clarify the overall principles and requirements: they define the purpose, basis, and scope of application of the guidelines, stipulate the basic principles that pricing behavior should follow, and encourage industry associations to strengthen self-regulation and promote the construction of industry price compliance.
• The regulations refine pricing practices for automobile manufacturers: clarifying price compliance requirements at every stage, from vehicle and parts production to pricing strategies and sales activities. Implementing end-to-end price management, strengthening fair pricing constraints, standardizing promotional and pricing practices, and legally combating unfair pricing practices.
• Clarify pricing behavior requirements for car dealerships: Focusing on the new car sales process, efforts will be made to regulate prominent issues such as failure to clearly display prices as required and false promotions. A risk warning mechanism will be established, encouraging platforms to provide both operational and consumer risk warnings for significantly low-price practices.
• Guide enterprises to strengthen internal compliance: Encourage automobile manufacturers and sales companies to establish internal price compliance management mechanisms, including six mechanisms: price decision-making, sales contract management, internal supervision, price emergency response, risk prevention and control, and price compliance training, to achieve full-process control of pricing behavior.

The guide specifically mentions:

Besides legally reducing prices to clear stockpiled inventory, automobile manufacturers face significant legal risks if they engage in the following pricing practices aimed at squeezing out competitors or monopolizing the market:

• By using high-specification, high-grade products to pass off as low-specification, low-grade products, etc., the price is reduced in disguise, making the actual ex-factory price lower than the production cost.
• By offering larger quantities and bulk discounts, prices are reduced in disguise, making the actual ex-factory price lower than the production cost.
• Other methods are used to make the actual ex-factory price lower than its production cost.

Automobile manufacturers should regulate their charging practices for "paid unlocking" functions:

• For features with a free period, the salesperson should inform the consumer of the free period, subsequent charges, and other details, and remind the consumer again before the free period ends.
• For differentiated value-added features that require payment, this should be clearly stated during the sales process; differentiated value-added features that are not clearly stated should not be charged.

Car dealerships that use the following false or misleading pricing tactics to deceive consumers face significant legal risks:

• Using deceptive or misleading language, text, numbers, images, or videos to indicate prices and other pricing information;
• Using false comparison prices such as "market price", "manufacturer's suggested retail price" and "market reference price" in advertising;
• If a purchase of a car or insurance is used as a price-addressing condition, and this is not disclosed or clearly indicated before the transaction;
• Refusing to fulfill or not fully fulfilling price commitments without justifiable reasons, making commitments regarding car prices, the value of gifts, after-sales service fees, etc., but failing to deliver;
• They attract consumers with low-price advertising, but actually do not sell any related products.

OpenAI Releases New Real-Time Programming Model

Early this morning, OpenAI announced the release of a new real-time programming model, "GPT-5.3-Codex-Spark".

According to the official statement, GPT-5.3-Codex-Spark is a smaller version of GPT-5.3-Codex, and the launch of this model marks the first milestone in the collaboration between OpenAI and AI chip startup Cerebras.

It's worth noting that Cerebras is considered a challenger to NVIDIA, specifically designed to address the issues of AI inference speed and latency.

According to OpenAI's official description, Codex-Spark is optimized to provide near-instantaneous responses when running on ultra-low latency hardware— generating more than 1,000 tokens per second —while maintaining robust capabilities in real-world programming tasks.

At launch, Codex-Spark has a 128k context window and currently only supports text input . During the research preview, Codex-Spark will have a separate rate limit, and its usage will not be counted towards the standard rate limit.

In terms of performance, in the SWE-Bench Pro benchmark test, Codex-Spark is slightly lower in accuracy than the complete GPT-5.3-Codex, but it significantly reduces task time (several times faster) and is significantly better than GPT-5.1-Codex-mini.

Notably, this GPT-5.3-Codex-Spark runs on Cerebras' Wafer Scale Engine 3 (WSE-3) . The WSE-3 reportedly boasts 4 trillion transistors and 900,000 AI cores, providing extremely low latency for Codex-Spark.

GPT-5.3-Codex-Spark is now available as a research preview for ChatGPT Pro users.

Musk's assessment of Seedance 2.0: Its development is too rapid.

ByteDance recently launched its latest video generation model, Seedream 2.0, which has garnered significant attention for its excellent image quality consistency and smooth camera movement.

Yesterday, Musk retweeted a video generated by Seedance 2.0 and exclaimed, "It's happening fast."

Also yesterday, director Jia Zhangke posted that "Seedance 2.0 is really amazing, I'm going to use it to make a short film." Luo Yonghao then reposted it and exclaimed, "Making…making a movie only requires one person, the director."

As early as February 9th, Feng Ji, producer of *Black Myth: Wukong* and founder of Game Science, also published an article reviewing Seedance 2.0:

This is currently the most powerful video generation model available, bar none. AI's ability to understand and integrate multimodal information (text, images, video, and audio) has achieved a remarkable leap forward. The production cost of general videos will no longer follow the traditional logic of the film and television industry, gradually approaching the marginal cost of computing power. The content industry will inevitably experience unprecedented inflation, and traditional organizational structures and production processes will be completely restructured.

Feng Ji also mentioned that AI videos will become more accessible to everyone, and the videos themselves will gradually evolve towards customization, real-time interaction, and gamification, "which may become a completely new form of entertainment in the future."

He also emphasized that with increasingly sophisticated video models, the "proliferation of fake videos and the crisis of trust" will worsen. He stated that realistic fake videos will become virtually effortless, and existing intellectual property and censorship systems will face unprecedented challenges.

Seedance 2.0 is reportedly capable of creating cinematic-quality videos from text or images. It employs a dual-branch diffusion transformer architecture, generating both video and audio simultaneously. With just detailed prompts or a single image uploaded, Seedance 2.0 can generate a multi-shot video sequence with native audio within 60 seconds.

It's worth mentioning that this model's unique multi-camera storytelling feature can automatically generate multiple interconnected scenes based on a single prompt. The AI ​​automatically maintains consistency in characters, visual style, and atmosphere across all scene transitions, eliminating the need for manual editing.

Songdian cameras join the Micro Four Thirds camp

Yesterday, Olympus Olympus Digital Technology issued a press release announcing:

Shenzhen Sonida Digital Technology Co., Ltd. has officially joined the Micro Four Thirds (M4/3) standard camp.

Subsequently, Songdian Camera stated:

We extend our sincerest respect to industry pioneers such as OM Digital Solutions and Panasonic, who initiated and have continuously promoted the M43 (Micro Four Thirds) standard, as well as to all partners who have contributed to this open ecosystem.

We are honored to join the M43 Alliance and know it is a great encouragement. We are willing to always learn from our colleagues with humility, move forward side by side, encourage each other and make progress together to build a more vibrant future for imaging.

According to previous official announcements from Matsunori, it is highly likely that they will launch a digital camera with an M43 sensor.

Many netizens have started to come up with ideas for Matsunori: directly challenge Panasonic by making a GX10 – since they have the GX9, why not make a GX10?

Multiple automakers disclose their all-solid-state battery industry plans

According to CLS News Agency, following the announcement that the first part of the national standard for solid-state batteries for automobiles is scheduled to be released in July this year, several vehicle and battery companies, including Geely, Chery, BYD, and Sunwoda, have recently disclosed their respective technical paths and industrial plans for all-solid-state batteries.

Geely Automobile:

In the field of all-solid-state batteries, three major technical routes have been developed, employing polymer (organic), sulfide, and halide (inorganic) composite solutions to meet the needs of different application markets.

Meanwhile, Geely Automobile has developed high-nickel ternary cathode materials for all-solid-state batteries and composite electrolytes with flame-retardant and self-extinguishing functions, and invented in-situ lithium dendrite re-intercalation repair technology, etc.

Regarding application planning, according to Shen Yuan, Senior Vice President and CTO of Geely Holding, Geely Automobile's short-term goal is to complete the debut of the prototype vehicle in 2026; to achieve small-batch industrialization of all-solid-state batteries and 1,000 vehicles for demonstration operation in 2027; and to complete the industrialization layout of all-solid-state batteries by 2030 and launch them in batches on high-end models.

At that time, Geely's all-solid-state battery cell energy density is expected to exceed 500Wh/kg, and the BOM cost will be controlled within 0.6 yuan/Wh.

Chery Automobile:

Gu Chunshan, Vice President of Chery Automobile, recently stated that the company plans to put a 0.5GWh pilot production line into operation and roll out PACK sample packs in 2026, complete the continuous production of 60Ah-level all-solid-state battery cells, and simultaneously strengthen the construction of the supply chain system. In 2027, the company will officially launch the demonstration work of all-solid-state battery installation in vehicles, promote the technology from the production line to real vehicle verification, and gradually realize the large-scale application.

China FAW:

Earlier reports indicated that the first prototype of Hongqi's independently developed all-solid-state battery officially rolled off the production line in January of this year, achieving a series of breakthroughs in key areas such as sulfide electrolyte, 10Ah cell performance, and 60Ah cell technology.

The Hongqi all-solid-state battery cell, which has been verified for vehicle use, has successfully passed the 200℃ extreme thermal abuse test, and the ionic conductivity of the sulfide electrolyte has exceeded 10mS/cm.

BYD:

The Investor Relations Department recently disclosed that the company is focusing on sulfide solid-state batteries as an important technology direction. Solid-state batteries have achieved breakthroughs in areas such as battery life and fast charging, and are expected to achieve small-batch production in 2027.

On February 2, Sunwoda responded to questions about the company's battery progress on the investor interaction platform, stating that the company's first-generation and second-generation semi-solid-state batteries have achieved mass production, and all-solid-state batteries are expected to achieve mass production in 2027.

Reports indicate that Meizu's mobile phone business is about to be dissolved.

According to Sina Tech, some netizens recently revealed that the Meizu 23 project has been halted and Meizu's mobile phone business is about to be dissolved.

Meizu has not yet responded to the above situation; Meizu customer service stated that they have not received any relevant notification.

Previously, at the Meizu Fans' Spring Festival event in 2026, Meizu announced that it would release the Meizu 23 smartphone in 2026, claiming that the model would have an even better feel and the narrowest bezel in Meizu's history, but would no longer feature a white front panel. Meizu had previously revealed that the Meizu 23 was planned for release in the middle of the year.

In addition, Meizu announced at the time that the Meizu 22 Air launch plan was canceled, citing the impact of a sharp rise in memory prices as the reason.

At its annual meeting, InStone gave away five apartments; the CEO responded.

Recently, Insta360 has attracted much attention for giving away houses at its annual meeting.

It is reported that at this year's annual meeting, Instone gave away 5 apartments to winners, all of whom were born in the 1990s and had an average age of less than 32. In addition, 6 cars of their choice, including popular models such as Porsche, BMW and Tesla, were also given away. A total of 33 cars were given away, with a total value of over 10 million yuan.

Yesterday, InStone CEO Liu Jingkang posted an article stating:

Often, motivating top talent relies on a sense of accomplishment, honor, or shared values, but direct material incentives are also important.

It's worth mentioning that, according to The Paper, Liu Jingkang also showcased the dual-camera design of InSight's first handheld gimbal camera at the InSight annual meeting, and officially named the product "Luna"; the new camera will be equipped with a telephoto function.

"This is definitely not another Pocket (DJI's gimbal camera product)," Liu Jingkang said. "This is a product with many innovative definitions and innovations. The core of innovation is to solve problems that others have not yet solved, rather than simply making a new cost-effective product."

Former Honor CEO Zhao Ming joins Qianli Technology

Yesterday, Zhao Ming, former CEO of Honor Terminal Co., Ltd., announced that he has officially joined AI technology company Qianli Technology.

Zhao Ming said, "It is a great honor to have the opportunity to join Qianli Technology, a career I can dedicate myself to for ten years. I look forward to working with Yin Qi (Chairman of Qianli Technology) and his brother to create an AI business loop and help Qianli take off." Yin Qi's side also said, "Welcome Zhao Ming."

It is reported that Zhao Ming announced his official resignation on Weibo on January 17, 2025. At the time, he stated in the Honor internal community that he chose to resign due to poor health caused by the high intensity of work.

According to Jiemian News, after joining Qianli Technology, Zhao Ming will serve as co-chairman and has been nominated as a candidate for non-independent director of the company's sixth board of directors.

It is understood that Qianli Technology is an AI technology company transformed from a manufacturing enterprise. In November 2024, Yin Qi, a leading figure in China's artificial intelligence field, became the chairman of Lifan Technology. In January 2025, Yin Qi proposed a comprehensive embrace of the "AI + Car" strategy, and in February, the company officially changed its name to Qianli Technology.

Before joining Qianyu Technology, Zhao Ming had already stepped down as CEO of Honor and taken a year off, during which time there were many rumors in the industry about Zhao Ming's future plans.

In September 2025, he revealed at a university event that he was still paying attention to the industry, and there were things that excited him even more, including "things that are worth investing another ten years or even longer in compared to what I did with Honor."

Sources familiar with the matter revealed that, regarding the future division of labor at Qianli Technology, Yin Qi will focus more on grasping the development strategy of AI technology, while Zhao Ming will focus on promoting the closed-loop strategy of AI business models.

Lenovo's latest financial report shows that AI revenue increased by 72% year-on-year.

Yesterday, Lenovo Group announced its third fiscal quarter results for the fiscal year 2025/26, ending December 31, 2025:

• The group's revenue reached RMB 157.5 billion, representing a year-on-year increase of over 18%, significantly exceeding market consensus expectations and setting a new historical record.
• Adjusted net profit increased by 36% year-on-year, far exceeding market expectations. The profit growth rate was twice that of revenue growth, demonstrating Lenovo Group's resilience and certainty in delivering on its commitments in a complex environment.

Official sources stated that in the past quarter, the global AI industry chain faced multiple challenges, including rapidly rising prices of key components such as memory chips, increased supply and demand volatility, and questions about the sustainability of the AI ​​cycle.

Lenovo not only delivered on its promise to the market of "double-digit revenue growth and profitability," but also responded to external concerns with a "good track record" by leveraging its comprehensive advantages in supply chain management, operational capabilities, and overall resilience.

It is worth mentioning that AI-related businesses have become an important growth engine for Lenovo Group:

In the third fiscal quarter, Lenovo's AI-related revenue increased by 72% year-on-year, accounting for 32% of the group's total revenue.

Among them, AI PC revenue achieved high double-digit growth, AI smartphone revenue achieved triple-digit growth, AI server revenue achieved high double-digit growth, and AI service revenue achieved triple-digit growth, driving the Group's business structure to continue evolving towards higher value.

DeepMind CEO: AI will solve humanity's intractable problems in the next 15 years.

In a recent interview with Fortune magazine, Google DeepMind CEO Demis Hassabis mentioned:

Humanity is standing on the edge of a "new golden age of scientific discovery." Although the next 10 to 15 years will bring about a dramatic industry reshuffle and growing pains, it will ultimately usher in a technological revolution comparable to the Renaissance.

In the interview, Hassabis introduced the concept of "radical abundance." He predicted that AI will solve humanity's most intractable problems through the deep internalization of scientific methods.

• Medical Revolution: In the next 15 years, AI will make personalized medicine the norm and conquer major diseases.
• Energy Breakthrough: AI will accelerate the research and development of nuclear fusion and new solar energy materials, completely solving the energy crisis.
• Space exploration: Breakthroughs in computing power will ultimately enable humanity to "travel between stars and explore the Milky Way".

In the interview, Hassabis also mentioned some of the risks and challenges Google faces in today's AI community.

Faced with the rise of competitors like OpenAI, Hassabi frankly admitted that Google must confront the "innovator's dilemma." He emphasized, " If we don't disrupt ourselves, someone else will. You'd better do it your own way. "

It is reported that with the release of the Gemini series models and the Nano Banana image generation model, Alphabet's (Google's parent company) stock price soared by about 65% last year. Hassabis believes that the company has crossed the "watershed" of AI assistants assisting in advanced research.

Doubao 2.0 is rumored to be released on February 14th.

According to CLS News Agency, ByteDance's Volcano Engine has tentatively scheduled a series of important upgrades for the Doubao Big Model to be released on February 14, 2026.

The report states that the models released this time include Doubao Big Model 2.0, Audio and Video Creation Model Seedance 2.0, and Image Creation Model Seedream 5.0 Lite.

With this upgrade, Doubao Big Model 2.0 will be officially released, significantly improving its basic model capabilities and enterprise-level agent capabilities.

According to a previous report by LatePost, the Doubao 2.0 model is the most significant output of Wu Yonghui (head of basic research at ByteDance Seed) in the year since he took over Seed. It is a multimodal model similar to Gemini, with 1 trillion parameters, and is the largest model trained since Seed's inception.

According to reports from Seed, the model encountered infrastructure challenges during training. They analyzed that during Seed's two-year pursuit of excellence, it relatively neglected building fundamental capabilities, leading to instability when scaling parameters during the training of Beanbag 2.0, which at one point hindered progress.

It is reported that after encountering problems during the training of Doubao 2.0, multiple teams worked together for three months to solve the problems, mainly focusing on model architecture and training data, to ensure that the model could be launched before the Spring Festival.

Xiaomi open-sources its first-generation robot, VLA, large-scale model.

Yesterday, Xiaomi officially released its first-generation robot VLA large-scale model, Xiaomi-Robotics-0. Founder Lei Jun subsequently reposted the announcement and commented, "…"

Xiaomi started its robotics business very early on.

According to reports, Xiaomi-Robotics-0 is an open-source VLA model with 4.7 billion parameters, combining visual language understanding with high-performance real-time execution capabilities.

According to the official statement, Xiaomi-Robotics-0 not only achieved excellent results in the three major simulation tests, but also realized the generalization of physical intelligence in real-world tasks—with smooth movements, sensitive reactions, and the ability to perform real-time inference on consumer-grade graphics cards.

The model adopts the mainstream Mixture-of-Transformers (MoT) architecture, equipped with a combination of "Visual Language Brain (VLM) + Action Expert". According to the official statement, the model can both understand instructions and maintain a high degree of physical flexibility when performing actions, just like a human.

In terms of performance, Xiaomi-Robotics-0 sets several new state-of-the-art (SOTA) records:

• In the LIBERO, CALVIN, and SimplerEnv tests, the model achieved state-of-the-art results across all benchmarks and comparisons with 30 other models.
• The model retains VLM's inherent multimodal understanding capabilities, especially performing well in benchmarks that are more relevant to embodied concepts.

Technical homepage: https://xiaomi-robotics-0.github.io

Open source code: https://github.com/XiaomiRobotics/Xiaomi-Robotics-0

Model weights: https://huggingface.co/XiaomiRobotics

MiniMax launches its new flagship programming model, the M2.5.

MiniMax recently launched its new model, M2.5, in its MiniMax Agent. The MiniMax M2.5 is reportedly a flagship programming model.

According to reports, the MiniMax M2.5 has reached or surpassed the industry's state-of-the-art (SOTA) standards in productivity scenarios such as programming, tool access and search, and office work. For example, it has achieved SWE-Bench Verified (80.2%), Multi-SWE-Bench (51.3%), and BrowseComp (76.3%).

It's worth mentioning that in the core programming tests, M2.5 showed a significant improvement over its predecessor (M2.1), reaching a level similar to the Claude Opus series. Furthermore, in the multi-language related task Multi-SWE-Bench, M2.5 achieved first place.

In terms of performance, the M2.5 model has only 10B of activated parameters, which gives it a clear advantage in terms of memory usage and inference energy efficiency. It supports an ultra-high throughput of 100 TPS and its inference speed surpasses that of top international models.

In terms of cost, the official statement claims that M2.5 is the first cutting-edge model that can be used indefinitely without considering usage costs. When outputting 100 tokens per second, M2.5 only costs $1 to operate continuously for one hour; while when outputting 50 tokens per second, it only costs $0.3.

It's worth mentioning that MiniMax's stock price rose by more than 20% at one point during yesterday's trading session, with a total market capitalization exceeding HK$180 billion.

Several third-party train ticket sales platforms were summoned for talks.

According to Jiemian News, the Beijing Municipal Administration for Market Regulation recently organized an administrative meeting with 12 mainstream platforms involved in online train ticket sales, including Ctrip, Qunar, Fliggy, Tongcheng, Meituan, JD.com, TravelSky, High-Speed ​​Rail Butler, Didi, Gaode Map, Baidu Map, and Tencent Map, to focus on prominent issues in online train ticket sales that have drawn strong public criticism.

At the meeting, the Beijing Municipal Administration for Market Regulation clearly outlined four compliance requirements for each platform:

• Strictly implement the main responsibility and social responsibility, establish the correct business philosophy, and "help passengers travel more and cause less congestion";
• A comprehensive review of business models and service processes will be conducted to prohibit explicitly or implicitly suggest that consumers can obtain priority ticket purchase privileges through paid services. Misleading advertising such as "acceleration packages," "dual channels," and "remaining ticket monitoring" after tickets are sold out will be rectified in a timely manner, and social supervision will be accepted voluntarily.
• A comprehensive investigation and rectification of the platform's pages was conducted, products suspected of misleading advertising were removed, page promotional content was adjusted, and the use of 12306 images, text, trademarks, etc., in promotions that could mislead consumers into believing that the platform had a specific business partnership with 12306.
• We must conscientiously ensure clear pricing, prominently display the content and prices of value-added services, promptly rectify any discrepancies between the displayed price and the actual payment for train tickets caused by inconspicuous display of value-added services, and effectively protect consumers' right to know.

12306 adds a "lower-class seat" reminder

According to Density Finance, a netizen recently posted that when purchasing high-speed rail tickets, the 12306 app displayed a friendly reminder:

The only seats available for this ticket collection are first-class seats located in the back row of the Business Class 1 (the backrests are not adjustable, and there is no armrest separating the two seats).

Regarding the above situation, the 12306 customer service hotline stated that because some first-class seats are located behind business class, which is a special location, an additional reminder has been added.

Meanwhile, the customer service representative explained that if passengers see the notification and are concerned about purchasing this seat, they can cancel the order and repurchase it.

Cyberspace Administration of China is cracking down on the spread of false and misleading information without AI tags.

According to the official WeChat account of "Cyberspace Administration of China," recently, some online accounts have been publishing AI-generated and synthesized information without adding AI labels, which constitutes a problem of deceiving and misleading the public with false and misleading content, damaging the online ecosystem and causing a negative impact.

In response to the above situation, the Cyberspace Administration urged websites and platforms to conduct in-depth investigations and rectifications, and disposed of 13,421 accounts in accordance with laws and contracts, and cleaned up more than 543,000 pieces of illegal and irregular information.

It is understood that the accounts and content involved include maliciously gaining traffic by fabricating touching stories and sensational events, and selling AI-generated videos of public figures "sending blessings" and "paying New Year's visits" without authorization (for improper profit).

The Cyberspace Administration of China (CAC) calls for:

The majority of online content creators are proactively adding AI identifiers in accordance with relevant regulations to avoid misleading the public and jointly create a clean and healthy online environment.

Zootopia 2 enters the top 8 in Chinese film history at the box office.

According to Lighthouse Pro, as of 10:00 AM on February 12, "Zootopia 2" had grossed 4.523 billion yuan, surpassing "Detective Chinatown 3" to enter the top 8 of the all-time Chinese box office chart.

The list shows that the top three are *Ne Zha* (15.446 billion yuan), *The Battle at Lake Changjin* (5.775 billion yuan), and *Wolf Warrior 2* (5.694 billion yuan). *Zootopia 2* currently surpasses *Detective Chinatown 3* and *Avengers: Endgame*.

Zootopia 2 was released on November 26, 2025, and became one of the top three highest-grossing films in China in December of the same year.

A sequel to F1: Racing has been confirmed and production has begun.

According to IT Home, citing a BBC report, Jerry Bruckheimer, producer of "F1: Racing," confirmed during the Oscars luncheon that production of the sequel has officially begun.

It is understood that the production team is currently working on a sequel. However, Bruckheimer did not reveal a specific timeline or confirm whether Pitt would reprise his role, but emphasized that he would definitely be involved in casting.

Released last summer, the film grossed $630 million worldwide, or about £467 million, making it the most successful theatrical release of any Apple original series to date.

The film was shot extensively on location during the 2023 and 2024 F1 seasons, with many scenes filmed directly at actual race weekends, including the British Grand Prix at Silverstone.

Bruckheimer said he was pleasantly surprised by the film's recognition with multiple awards, including a Best Picture nomination, and that working with Brad Pitt for the first time was an exciting experience.

A new trailer for "Star River Dream" has been released.

Yesterday, the sci-fi film "Star River Dream" released a trailer titled "Dream Crisis".

The film tells the story of a near-future virtual dream system called "Good Dream," where people can freely create their own dreams, but this unexpectedly triggers a crisis that threatens reality. Xu Tianbiao, the administrator of "Good Dream," tries to rescue the hibernating astronaut Li Simeng by uncovering the truth that her world is a virtual program. The two travel through more than ten distinct dream worlds to carry out the rescue.

The film is executive produced by Fu Ruoqing, directed by Han Yan, and written by Qian Ninghuang, Han Yan, Gao Lu, Liu Dalin, Xu Chen, and Ma Peisi. It stars Wang Hedi and Song Qian, with Zu Feng and Luo Haiqiong in supporting roles and Wang Duo making a special appearance. It will be released on February 17.

#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.

ifanr | Original Link · View Comments · Sina Weibo

Twenty years ago, Stephen Chow used a bruised-faced character named Sing in "Kung Fu Hustle" to tell us that true masters often hide among ordinary people, waiting for the moment when their meridians are opened.

Twenty years later, this once-in-a-million martial arts prodigy reappeared. However, it was not human.

Just now, Zhiyuan Robot released a video titled "Peerless Master, Coming Down the Mountain Soon" on its video channel. The official account specifically noted that the entire video was filmed live, without any CG effects or AI creation.

At the beginning of the video, a humanoid robot named Zhiyuan Yuanzheng A3 performs a clean and decisive flying kick in the training room, not only shattering the glass and kicking open the door to the Year of the Horse, but also probably waking up many people's stereotype that domestic robots are "not nimble".

In the video, A3 starts with a rising dragon punch, followed by a series of flying kicks, an aerial walk, and even an extremely difficult "ground whirlwind".

Anyone who has worked with robots knows that making a humanoid robot walk steadily is a major engineering challenge. As an electrically driven full-size humanoid robot, the Expedition A3's ability to perform such highly dynamic, explosive, and fluid movements demonstrates that its motion control algorithms and instantaneous power output have reached a new level.

More importantly, this martial arts actor-level performance also solves a long-standing pain point in the industry: the "stiffness" of robot movements.

The A3 combo demonstrates excellent overall coordination, which, according to the official description, is due to its fully flexible waist, which replicates the range of motion of the human waist in a 1:1 ratio.

In addition to its lightweight leg exoskeleton structure, this robot's TCP end effector speed can reach up to 2 m/s. In simpler terms, its punch speed might be faster than your reaction time.

Of course, it's a little disappointing that there wasn't a specific demonstration of dexterity in this video.

However, if the Expedition A3 were only good at looking cool, it would be nothing more than an expensive, oversized toy. This brings us to the biggest practical upgrade: battery life. A few months ago, their predecessor, the Expedition A2, was taking an ascetic approach—traveling 100 kilometers in the sweltering heat, but its actual battery life was only about 2 hours.

The Expedition A3 uses an embedded dual-battery system, which not only looks slimmer and more compact, but also extends the overall battery life to a maximum of 8 hours on a full charge.

What does 8 hours mean?

The standard human workday is only 8 hours, which means that for the first time, the Expedition A3 has the capability to fully cover a human shift. With the addition of fast battery swapping technology, it can theoretically operate 24 hours a day—a true "007" workhorse, and is planned for mass production by 2026.

In addition to the evolution of physical abilities, Expedition A3 has also maxed out its emotional intelligence.

The A3 features an end-to-end large model, with the biggest change being the "wake word removal". You no longer need to awkwardly shout commands; it can respond directly based on the context, and even supports waking you up by tapping your shoulder.

Based on these capabilities, Logic Studio positions the A3 for scenarios like guided tours and commercial performances, which is actually a precise and effective strategy. Imagine a robot in a shopping mall that can perform a Wing Chun routine for you and even help you carry a 3kg object – wouldn't that be far more eye-catching than a real person handing out flyers?

At the end of the video, the master grants Yuanzheng A3 permission to "descend the mountain." This is not only a movie Easter egg but also an industry metaphor. As more and more humanoid robots begin to pursue superhuman athleticism, long battery life, and more natural interaction, robots will hopefully truly integrate into human commercial society.

Of course, there is still a long way to go from "being able to go down the mountain" to "traveling all over the world". Cost control, scenario adaptation, security standards, and user acceptance are all challenges.

With the Year of the Horse approaching, various humanoid robots are descending from the mountains one after another. We will soon see who will become the true "master".

#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.

ifanr | Original Link · View Comments · Sina Weibo

The first part of the national standard for solid-state batteries for automobiles is expected to be released in July.

According to a report by Cailian Press, Wang Fang, chief scientist of the China Automotive Technology and Research Center (CATARC), recently stated at an industry conference that the draft for comments on GB/T "Solid-state batteries for electric vehicles – Part 1: Terminology and Classification" was completed in December 2025, and the comment period will last until February 28, 2026.

It is understood that the national standard will clarify the definitions of terms such as liquid battery, hybrid solid-liquid battery (semi-solid battery), and solid battery (all-solid battery).

The report indicates that the China Automotive Technology and Research Center (CATARC) will organize verification tests in January-February 2026 to further improve testing methods and confirm judgment indicators; it is expected to be reviewed and approved in April 2026 and officially released in July.

According to our research, the standard "Solid-state batteries for electric vehicles – Part 1: Terminology and classification" has been made available on the National Standards Information Public Service Platform.

The content shows that the main drafting units are China Automotive Technology and Research Center Co., Ltd., CATL (Contemporary Amperex Technology Co., Limited), and Hefei Guoxuan High-Tech Power Energy Co., Ltd.

my country successfully completed its first sea search and recovery mission for the return capsule of a manned spacecraft.

According to Xinhua News Agency, on February 11, during the low-altitude demonstration and verification of the Long March 10 carrier rocket system and the maximum dynamic pressure escape flight test of the Mengzhou manned spacecraft system, the Mengzhou manned spacecraft successfully carried out maximum dynamic pressure escape and safely splashed down at sea.

At 12:20 on the 11th, the maritime search and rescue team completed the search and recovery mission of the return capsule. This is China's first manned spacecraft search and recovery mission at sea , accumulating important experience for subsequent space station application and development missions and manned lunar landing missions.

The report states that this test is another research and development flight test organized and implemented after the Long March 10 carrier rocket tethered ignition, the Mengzhou manned spacecraft zero-altitude escape flight, and the comprehensive verification of the lunar lander's landing and takeoff. It marks an important stage breakthrough in the research and development of my country's manned lunar exploration project.

According to the China Manned Space Engineering Office, this test features many highlights, including a new type of rocket, a new type of spacecraft, a new launch pad, and a new mission for the sea recovery of the rocket and spacecraft. Both the rocket and spacecraft participating in the test were in their prototype state.

The rocket uses a core stage single-stage configuration and underwent two tethered ignition tests in the early stages; the spacecraft's return capsule underwent a zero-altitude escape flight test in the early stages.

To conduct this test, all relevant test products underwent adaptive modifications in accordance with reusable requirements and procedures. The Wenchang Space Launch Site overcame various difficulties by adopting a strategy of building and using simultaneously to ensure the test was carried out on schedule. The landing site system conducted targeted training and drills focusing on the technical challenges of the spacecraft's first sea splashdown and recovery of the return capsule.

It is worth mentioning that this test was the first ignition flight of the Long March 10 carrier rocket in its prototype state, the first maximum dynamic pressure escape test of a spacecraft in China, the first sea splashdown of the return capsule and the first stage of the rocket in China, and the first ignition flight test mission carried out at the newly built launch pad of the Wenchang Space Launch Site.

The success of this test verified the flight performance of the rocket's first-stage ascent and recovery phases, as well as the spacecraft's maximum dynamic pressure escape and recovery functions. It also verified the compatibility of the interfaces of various engineering systems, accumulating valuable flight data and engineering experience for subsequent manned lunar exploration missions.

Apple's AI Siri delayed again: capabilities significantly reduced, may arrive in September.

According to Bloomberg, Apple has internally adjusted its schedule, and some features of the new Siri will be delayed until iOS 26.5, or even iOS 27 which will be released in September . It will take several versions to complete the functionality.

Regarding the release schedule, the new Siri was initially promised to be available in early 2025, but was subsequently delayed until March 2026 with the release of iOS 26.4. However, the engineering team has recently been asked to switch to iOS 26.5 as the primary testing environment.

From an overall planning perspective, Apple is splitting the Siri update, which was originally concentrated in a single version, into multiple phases. Some features are expected to be released in iOS 26.5 in May, while the rest will be moved to iOS 27.

• During the current testing phase, the new Siri has revealed several shortcomings, including inconsistent understanding of voice requests, excessively long response times, and a tendency to fail in complex command scenarios. Testers also reported that Siri would interrupt commands midway through if the user spoke too quickly, resulting in a poor testing experience.
• The "deep access to personal data" capability, originally touted as a killer selling point, is being significantly weakened. This feature allows Siri to directly retrieve information from personal content such as text messages and invoke corresponding app functions. However, in the iOS 26.5 build, it only exists as a "preview" that needs to be manually enabled, indicating that the experience is not yet mature.
• Another capability that has seen slow progress is advanced voice control based on App Intents, which aims to complete multiple in-app operations with a single sentence, such as finding, editing, and sending an image. This mechanism has been implemented in the beta system, but its stability is insufficient.
• It's worth noting that internal builds of iOS 26.4 and 26.5 also include two unreleased new capabilities: web search question answering and image generation. The former is similar in form to Perplexity, capable of integrating web information and providing summaries and links; the latter is based on TuLeyuan, but current test results are unsatisfactory.
• For iOS 27, iPadOS 27 and macOS 27, Apple is also pushing forward a project codenamed "Campo", with the goal of making Siri a system-level AI entry point that is more like a chatbot, deeply integrated with core applications such as Mail, Calendar and Safari;

It is understood that a major reason for the slow progress of Siri upgrades is Apple's strong stance on privacy. Apple insists on performing inference as much as possible on the device, or using only servers with privacy protection mechanisms, and avoids directly using real user data for training. This conflicts with the working principle of LLM models.

On the computing power side, Cook recently told internally that Apple Silicon is being used to build data center solutions specifically designed for Apple devices. This cloud AI chip project, codenamed Baltra, will support more demanding Siri and generative AI tasks in the future.

Clearly, Apple doesn't want the new Siri project to drag on past the crucial spring of 2026, but the reality is that AI Siri's troubles continue, and currently, Apple doesn't have a better solution.

Less than 48 hours later, two co-founders of xAI announced their departure.

According to Jiemian News and public information, xAI's two co-founders, Tony Wu and Jimmy Ba, announced their departures on the 10th and 11th, respectively.

On February 10, Wu Yuhuai posted on social media announcing his departure from xAI. He expressed his nostalgia for the company and gratitude to all his colleagues, as well as his thanks to Musk for his "trust in the mission".

Although Wu Yuhuai did not reveal his future plans, he mentioned, "It's time to start my next chapter. This is an era full of possibilities: a small team equipped with AI can move mountains and redefine possibilities."

Information shows that Wu Yuhuai is a key figure at xAI in the fields of mathematical reasoning and symbolic AI. During the live stream of the Grok-3 model launch last year, he sat next to Elon Musk as a core researcher; Musk hailed the model as "the smartest AI on Earth."

It is understood that before Wu Yuhuai left the company, xAI underwent a restructuring, and many of his job responsibilities were transferred to Guodong Zhang.

Public information shows that Zhang Guodong is a core member of xAI, mainly responsible for machine learning optimization and training. Previously, his team was dedicated to building Artificial General Intelligence (AGI) and open-sourcing the Grok series of large models. Before joining xAI, his professional experience covered top institutions such as Microsoft Research Asia, Google Brain, and DeepMind.

On February 11, Jimmy Baht announced that he had left Musk's startup on Tuesday.

In his article, Jimmy Bar stated, "It's time to recalibrate my perception of the big picture. 2026 will be extraordinary, and likely the busiest (and most decisive) year in terms of humanity's future."

He also expressed his gratitude to Musk, mentioning that he was fortunate to participate in the co-founding of xAI in its early stages, cherished this extraordinary journey together, and promised to maintain close contact with the company as a close friend of the team in the future.

It is understood that Jimmy Bach previously held a senior position, reporting directly to Elon Musk and overseeing most of the company's operations. However, at the end of last year, many of his core responsibilities were reassigned to the other two co-founders, Wu Yuhuai and Zhang Guodong.

According to xAI's organizational chart from early last year, he was also in charge of managing a team of over a thousand AI mentors, a position that was taken over by Diego Pasini in September of last year.

In addition, Tesla Vice President Raj Jegannathan also recently announced his departure.

In his article, Raj Jegannathan mentioned that his 13-year career is difficult to explain in a single article, but his journey at Tesla has been a process of continuous evolution.

It's an honor to serve in the design, construction, and operation of the world's largest AI cluster, from the technical details to making impactful contributions in IT, security, sales, and service.

Tencent x Tesla: Launching WeChat Connectivity and Destination Services

Recently, Tencent and Tesla jointly upgraded the driver's cabin, launching two major functions: "WeChat Connect" and "Destination Services".

It is understood that these two functions leverage Tencent's ecosystem and technological advantages to jointly create a seamless, intelligent, and convenient new cockpit experience. This initial rollout will be via OTA (Over-The-Air) update, covering over one million Model 3 and Model Y vehicles in the Chinese market, and will subsequently be standard equipment on all new vehicles.

The WeChat Connect feature, based on Tencent's mobile connectivity capabilities, enables seamless location sharing between mobile phones and in-vehicle systems. Users can share location cards from WeChat conversations, or POI information from various sources such as Tencent Maps, Baidu Maps, and Gaode Maps, to the Tesla in-vehicle system with a single click and initiate navigation directly.

Destination services provide an intelligent experience around travel scenarios. When a user navigates to a specific destination such as a coffee shop, restaurant, or scenic spot, the system will intelligently trigger nearby services. It supports launching merchant mini-programs via mobile phone push or QR code scanning by the passenger, enabling convenient operations such as instant stop and pick-up, and pre-booking tickets.

In the future, more features will be launched to help users enjoy a more convenient and intelligent travel experience.

Doubao and Ant Afu join the Spring Festival red envelope battle

Following the "Yuanbao Pai Red Envelope" and "Qianwen Invites You to Drink Milk Tea" campaigns, Ant Fortune and Doubao have also recently launched their own Chinese New Year red envelope campaigns.

• Ant Financial's Afu: New users who register on Afu between February 11th and February 12th will receive a 16.8 yuan Alipay red envelope.
• Doubao: On the eve of Chinese New Year, February 16, it will give away more than 100,000 technological gifts and New Year cash red envelopes of up to 8,888 yuan to audiences across the country through CCTV's Spring Festival Gala.

Previously, Tencent's Yuanbao announced the distribution of 1 billion yuan in New Year red envelopes; Alibaba's Qianwen launched "Spend 3 billion yuan to win prizes" to offer users free milk tea, coffee, and meals.

Reports indicate that ByteDance is developing an AI chip.

According to Reuters, ByteDance is pushing forward with the development of its own AI chips and is in talks with Samsung about contract manufacturing and supply, with the core objective of ensuring a stable supply of computing power for its AI.

The report, citing sources familiar with the matter, states that ByteDance is developing an AI chip codenamed "SeedChip," primarily for inference tasks, and is in talks with Samsung regarding manufacturing cooperation. The goal is to obtain samples by the end of March, with plans to mass-produce at least 100,000 chips this year, potentially increasing to 350,000 chips later.

In response to the news, a ByteDance spokesperson told Reuters that the information regarding the self-developed chip project was "inaccurate," but did not elaborate further; Samsung declined to comment.

The report also mentioned that ByteDance's AI-related procurement budget this year may exceed 160 billion yuan, more than half of which will still be used to purchase Nvidia chips (including H200), while continuing to promote its self-developed chips.

Recently, with the Seedance 2.0 model gaining popularity, although the platform claims that the attention it receives has exceeded expectations and that it has suspended the use of real-person materials, the fact is that the generation of this type of high-quality video with "synchronized audio and video" will increase the demand for inference computing power.

ByteDance's increased procurement and self-developed chip development is more like securing key supplies, leaving room for peak computing power, and reducing its dependence on a single supply chain.

Musk: We want to build a factory on the moon to produce AI satellites.

According to The New York Times, Musk stated at the xAI all-hands meeting on February 10th that the company needs to build a factory on the moon to manufacture AI satellites and equip them with a giant catapult to launch them into space.

The report noted that Musk declared at the meeting, "We must go to the moon." He stated that this move would help xAI stay ahead of other companies and enable it to build its AI infrastructure more powerfully.

Musk added, "It's hard to imagine what intelligence of that scale would think, but it would be incredibly exciting to see all of this happen."

Recently, xAI and SpaceX announced their merger to advance their plan to build an AI data center in outer space. Elon Musk subsequently posted a message stating:

SpaceX has shifted its focus to building a self-growing city on the moon, which we may be able to achieve in less than 10 years, while it will take more than 20 years on Mars.

Musk pointed out that the planets will only align every 26 months (six months travel time), at which point it will be possible to go to Mars. "But we can launch to the moon every 10 days (two days travel time). This means we can iterate and build lunar cities much faster than Martian cities."

He also stated that SpaceX will also strive to build a Martian city, and will begin implementation in about 5 to 7 years, but the priority is to ensure the future of civilization, while the moon will be built much faster.

Rokid Glasses now supports OpenClaw and features custom intelligent agents.

On February 11, Rokid announced that its Rokid Glasses Lingzhu platform has officially launched the "Custom Intelligent Agent" feature.

According to the official introduction, users can connect Rokid Glasses to any desired backend through the standard SSE (Server-Sent Events) interface, and can privately deploy OpenClaw, DeepSeek R1, Qwen3, Kimi K2.5, etc.

According to Rokid, Rokid Glasses, in conjunction with OpenClaw, can achieve the following:

• Privacy and data sovereignty: By accessing locally deployed agents (such as OpenClaw running on a Mac Mini or home server), users' camera footage, private knowledge base, and personal memories are processed directly in a closed loop locally;
• More powerful personal assistant capabilities: Through ClawHub's skill ecosystem, OpenClaw can access the file system, operate the browser, send IM messages, and even execute Python scripts;
• A completely free large model base: utilize Kimi K2.5's Thinking Mode to tackle complex reasoning, or deploy a local quantized version of Qwen 3 to enjoy fully localized edge interaction.

Rokid states that Rokid Glasses are responsible for "seeing" and "hearing," while OpenClaw is responsible for "thinking" and "execution," and this is the ultimate form of AR.

Yu Donglai announced his "formal retirement after the Chinese New Year".

According to the Henan Daily WeChat official account, on February 11, Yu Donglai, the founder of Pang Donglai, announced on social media that he would officially retire after the Spring Festival and become an advisor. The work of Pang Donglai will be overseen by the Pang Donglai Decision-Making Committee.

In his article, Yu Donglai mentioned that "he is sixty years old. In order to promote a scientific and advanced development model and encourage young people to take on their responsibilities, he will officially retire after the Lunar New Year and become an advisor . The work of Pang Donglai will be overseen by the Pang Donglai Decision-Making Committee."

Yu Donglai also pointed out, "We are determined to contribute to the well-being of our country and society, and to inspire more people to live a relaxed and happy life… I believe it will definitely become even better!"

It is reported that Pang Donglai's total sales exceeded 23.531 billion yuan in 2025. Among them, the supermarket format occupied the core position, with sales reaching 12.643 billion yuan, the jewelry format had sales of 2.451 billion yuan, and the tea sales reached 1.063 billion yuan.

Jony Ive: Touchscreens are not suitable as the primary interaction method for cars.

Ferrari recently unveiled the interior design of its first all-electric model, the "Luce." The interior features multiple physical buttons that can be used to adjust functions such as air conditioning and seat heating.

It is understood that Luce's design was undertaken by the LoveFrom team, led by former Apple Chief Design Officer Jony Ive.

Jony Ive recently revealed his thoughts on vehicle infotainment: touchscreens are not suitable as the primary interface for cars.

Jony Ive mentioned that using a touchscreen while driving causes the driver's eyes to leave the road, which is why touchscreens are not suitable as the primary interface technology.

Jony Ive also pointed out that the feel of each physical button is different, so blind operation is possible without looking at it.

When asked by a reporter, "Then why does the iPhone use touch?", Jony Ive directly stated, "Because the core idea behind developing the iPhone was to create a universal interface that could display any scene (such as a computer, keyboard, or camera)."

I don't think designing car interiors is about trying to solve problems like we design the iPhone (referring to designing a universal interface).

Zhipu launches the new model GLM-5

Just now, Zhipu officially launched and open-sourced its latest model, GLM-5.

According to reports, GLM-5 is a product of the move towards Agentic Engineering: in terms of coding and agent capabilities, it achieves state-of-the-art performance in open source, and its user experience in real programming scenarios is close to that of Claude Opus 4.5. It excels in complex systems engineering and long-term agent tasks.

GLM-5 adopts a brand-new platform: the parameter size has been expanded from 355B (32B activations) to 744B (40B activations), and the pre-training data has been increased from 23T to 28.5T ; a brand-new "Slime" framework has been built to support larger model sizes and more complex reinforcement learning tasks.

In addition, GLM-5 integrates DeepSeek Sparse Attention for the first time , which significantly reduces the model deployment cost while maintaining the effect of long texts.

In terms of specific manifestations:

• In the globally authoritative Artificial Analysis rankings, GLM-5 ranks fourth globally and first among open source technologies.
• GLM-5 achieves alignment with Claude Opus 4.5 in terms of programming capabilities, and achieves state-of-the-art (SOTA) status among open-source models in industry-recognized mainstream benchmarks.
• GLM-5 achieved the highest scores among open-source models in SWE-bench-Verified and Terminal Bench 2.0, with scores of 77.8 and 56.2 respectively, outperforming Gemini 3 Pro.
• GLM-5 achieved top performance in BrowseComp (networked retrieval and information understanding), MCP-Atlas (large-scale end-to-end tool invocation), and τ²-Bench (automatic proxy tool planning and execution in complex scenarios).

It is worth mentioning that GLM-5 has completed deep inference adaptation with domestic computing platforms such as Huawei Ascend, Moore's Threads, Cambricon, Kunlun Chip, Muxi, Suiyuan, and Hygon. Through underlying operator optimization and hardware acceleration, GLM-5 has achieved stable operation with high throughput and low latency on domestic chip clusters.

Effective immediately, GLM-5 is open-sourced on both Hugging Face and ModelScope platforms, with model weights licensed under the MIT License. GLM-5 is also included in the GLM Coding Plan Max package.

Online experience

Z.ai: https://chat.z.ai

Zhipu Qingyan APP/Web Version: https://chatglm.cn

Open source link

GitHub: https://github.com/zai-org/GLM-5

Hugging Face: https://huggingface.co/zai-org/GLM-5

The model-making Spring Festival battle has quietly begun.

We have observed that DeepSeek has recently undergone a quiet update.

When you disable DeepSeek's Deep Thinking and online search features, and ask DeepSeek for the context length and knowledge base expiration date, if it replies that it is 1M (previously 128k) and the knowledge base expiration date is May 2025, it means that it has entered the gray-scale testing phase of the new version.

However, as of now, multimodal capabilities are still not available in this version.

Earlier, the South China Morning Post, citing sources familiar with the matter, reported that DeepSeek was more likely to release a minor update for its V3 series models during the Chinese New Year. However, the source also revealed that the real highlight was still on its way.

DeepSeek's next-generation flagship model is expected to be a trillion-parameter-level base model. Due to the significant increase in scale, the training speed has slowed down considerably, leading to a delay in the release process.

Furthermore, we have observed that the MiniMax Agent now features the latest MiniMax-M2.5 text model, which is available for selection.

No further information is available at this time.

Samsung Galaxy S26 series expected to be released at the end of the month

Yesterday, Samsung officially announced on its website that it will hold a global Galaxy product launch event at 2:00 AM on February 26, 2026, where it is expected to release new products such as the Samsung Galaxy S26 series smartphones and the Galaxy Buds 4 series headphones.

Based on the released teaser video, the Galaxy S26 series is expected to include three new models. Combined with previous information, the Galaxy S26 series will offer the S26, S26+, and S26 Ultra.

Based on the information currently available, all three new phones will adopt a rear camera module design similar to the Z Fold 7, and all will feature a straight front and rear panel with a straight-edged mid-frame.

It's worth mentioning that the new phone offers a brand-new hardware-level screen privacy feature. Reportedly, this feature will achieve pixel-level privacy through nanometer-level optical control and supports full-screen privacy mode.

In terms of performance, the S26 series may adopt a dual-platform strategy, with some regions using the Exynos 2600 processor (the world's first 2nm process chip); the S26 Ultra is also expected to be upgraded to 60W wired fast charging.

The all-electric Highlander has been launched, featuring the same family design.

Yesterday, Toyota unveiled the all-new all-electric Highlander, positioned as a mid-to-large SUV.

In terms of appearance, the all-electric Highlander adopts Toyota's latest C-shaped daytime running light design, and the overall lines of the vehicle are more straight and rugged; the floating A-pillar has been replaced with an all-black roof design; and the taillights have been changed to a through-type light strip design.

The overall vehicle size has increased: the body length has reached 5m, the wheelbase has increased to 3m, the width has also evolved to 1.98m, while the body height has decreased to 1.7m.

The new vehicle will be available in both front-wheel drive and all-wheel drive versions, and will have two battery options: 77 kWh and 95.8 kWh, offering a maximum range of 514 kilometers (320 miles).

Inside, the new vehicle features a 6-seat layout, with the second row offering power folding; it's equipped with a 14-inch central control screen (offering 5G network connectivity) and a 12.3-inch instrument panel, and the front seats feature double-layered laminated glass. Notably, the all-electric Highlander comes standard with a dashcam.

Ant Financial has open-sourced its full-modal large model, Ming-Flash-Omni 2.0.

On February 11, Ant Group released the open-source full-modal large model Ming-Flash-Omni 2.0.

In multiple public benchmark tests, Ming-Flash-Omni 2.0 has demonstrated outstanding performance in key capabilities such as visual language understanding, controllable speech generation, and image generation and editing, with some metrics surpassing Gemini 2.5 Pro, becoming a new benchmark for open-source full-modal large model performance.

It is reported that Ming-Flash-Omni 2.0 is also the industry's first unified audio generation model for all scenarios , capable of generating speech, ambient sound effects, and music simultaneously on the same audio track. Users can precisely control timbre, speech rate, tone, volume, emotion, and dialect simply by giving commands in natural language.

The model achieves an extremely low inference frame rate of 3.1Hz during the inference phase, enabling real-time high-fidelity generation of minute-long audio clips, maintaining industry-leading performance in inference efficiency and cost control.

It is worth mentioning that Ming-Flash-Omni 2.0 is trained on the Ling-2.0 architecture (MoE, 100B-A6B) and has been comprehensively optimized around the three major goals of "seeing more accurately, hearing more clearly, and generating more stably".

Currently, the model weights and inference code of Ming-Flash-Omni 2.0 have been released in open-source communities such as Hugging Face. Users can also experience and use it online through the official Ant Financial platform, Ling Studio.

UU Delivery has discontinued its "New Year's Greetings on Behalf" service.

Recently, UU Express launched a service called "New Year's Greetings on Your Behalf," which includes "paying respects to elders on your behalf and kowtowing on their behalf." The highest package is priced at 999 yuan, which has sparked heated discussions.

Yesterday, UU Run responded to the aforementioned service.

The platform stated that its initial intention in launching this service was to meet the specific needs of those who are unable to return home for the Lunar New Year due to being in different locations, having mobility issues, or being overseas. They hope that this service will help them convey their thoughts and feelings, compensate for the regret of not being able to greet people in person, and provide greater emotional support.

UU Delivery stated that its "New Year's Greetings Delivery" service is not a desecration or misinterpretation of traditional etiquette. The platform also stated:

To maintain a positive, healthy, and harmonious online environment and avoid unnecessary misunderstandings and conflicts arising from commercial innovation, we have, after careful evaluation, proactively removed some of the controversial services from our platform. For orders that cannot be fulfilled due to the removal of these services, we will proactively offer users compensation of three times the original amount, refunded to the original payment method. Thank you for your understanding.

Meanwhile, UU Run also stated that based on interesting requests from users (such as "on-call duty", "on-the-spot blind date", "on-the-spot pet care", "on-the-spot care for elderly people in hospitals"), they will evaluate and study the possibility of listing these services as soon as possible to help more users in need.

In addition, UU Express announced that it will officially launch the "Enjoy Your New Year's Eve Dinner with Peace of Mind" Spring Festival public welfare campaign to give back to society with practical actions.

On Lunar New Year's Eve, the platform will fully waive delivery fees for all frontline workers who are unable to return home to reunite with their families due to their commitment to their posts—including police officers, medical staff, firefighters, sanitation workers, etc.—for their families' errand orders related to delivering New Year's Eve dinners to them (the delivery driver's income will not be affected).

Weibo releases "2025 Hot Search Trends Report"

Yesterday, Weibo released the full version of the "2025 Weibo Hot Search Trend Report".

According to the report, "National Games", "the movie Nezha's popularity", "AI wave" and "college entrance examination" all made it into the "Top Ten Events"; and in the ranking of trending "keywords", "response" appeared the most times, while "Lei Jun", "Nezha", "concert", "Xiaomi" and "mobile phone" were also included.

It is worth mentioning that artificial intelligence and robotics almost dominate the keywords for emerging trends, with DeepSeek appearing on the list the most.

miHoYo announces the "dismissal" of its long-term legal counsel

On February 11, miHoYo's legal department issued a "Statement Regarding the Termination of Cooperation with Shanghai Huiye Law Firm," announcing:

We terminate all cooperative relationships with our former permanent legal counsel, Huiye Law Firm, and cease entrusting them with any legal matters related to miHoYo. At the same time, we permanently blacklist Shanghai Huiye Law Firm as a partner of the company, and will never cooperate with them again.

According to information from miHoYo:

Previously, miHoYo appointed Shanghai Huiye Law Firm as its permanent legal counsel to provide litigation legal services. During the period when Huiye Law Firm provided legal services to miHoYo, Zhou Kaichang, a member of the management committee and senior partner of Huiye Law Firm, knowingly violated the relevant provisions of the Lawyers Law Law by privately assigning intern lawyer Wu to continue to help Yuanjing Company collect and preserve litigation evidence and provide it to Yuanjing Company, despite knowing that the litigation case involving Yuanjing Energy Co., Ltd. v. miHoYo involved a conflict of interest.

Mihoyo stated that it has filed a complaint with the Xuhui District Justice Bureau of Shanghai regarding Zhou Kaichang's actions. Following an investigation by the Xuhui District Justice Bureau, the bureau has imposed administrative penalties on Zhou Kaichang in accordance with the law.

In addition, Envision Energy Co., Ltd. has applied to the Jiangyin People's Court to withdraw its lawsuit against miHoYo for online infringement liability dispute, and the Jiangyin People's Court has ruled to allow the withdrawal.

It is reported that Envision Energy sued miHoYo in June last year because the villain in "Bloody Zone" was named "Envision Industry", and the latter countersued for commercial defamation.

New way to play with Yuanbao: Enjoy watching the NBA All-Star Game in "Yuanbao Party"

On February 11, Tencent announced that Yuanbaopai officially launched the "Watch the NBA All-Star Game Together" special event, which includes four key battles: the Celebrity Game, the Rising Stars Game, the Individual Games, and the All-Star Game itself.

From February 14th to 16th, users can enter any Yuanbao Pai account, click the "+" button below, and watch the 2026 NBA All-Star Game with other Pai users. They can also discuss the game in real time by tagging Yuanbao within the Pai account and enjoy a 1080P high-definition live stream.

It is understood that this event has deeply integrated Tencent Sports' high-quality content resources. In addition to the benefits of high-definition live streaming, Yuanbao Pai has also innovatively incorporated AI interactive gameplay. "Yuanbao" will become an all-around "football expert". Users can @Yuanbao in the group at any time to discuss player data, check the rules of the game, or conduct in-depth tactical analysis in real time.

Currently, Yuanbaopai's "Listen and Watch Together" feature connects with QQ Music and Tencent Video, two high-quality content libraries.

A trailer has been released for the short film "Sandiwara," starring Michelle Yeoh.

According to Sina Movies, a trailer has been released for the short film "Sandiwara," starring Michelle Yeoh and directed by Sean Baker.

Michelle Yeoh plays five roles: video blogger, commentator, vendor, waiter, and singer. Set in a bustling night market in Penang, Malaysia, the story focuses on local culture, people, and life.

It is reported that the film will premiere at the Berlin Film Festival, and Michelle Yeoh will also receive a Lifetime Achievement Award at this year's Berlin Film Festival.

The animated film "The Eight Immortals!" is scheduled for release this summer.

Yesterday, the animated comedy film "The Eight Immortals!" was scheduled for release in the summer of 2026 and released a teaser poster titled "Crossing the Sea Under Cover of Darkness".

It is said that the story revolves around eight ordinary people from humble backgrounds who, facing a powerful villainous deity, cleverly outwit and outmaneuver him with their wit and folk wisdom, and are ultimately praised by the world as the "Eight Immortals".

The film is produced by Oriental DreamWorks and CMC Pictures, and directed by Mou Zhengyang. It interprets this classic legend with a comedic style and youthful innovation.

#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.

ifanr | Original Link · View Comments · Sina Weibo