Pogbagram

BYD launched a new product in a very low-key manner yesterday.

There was no dazzling light show, no impassioned speeches from executives, and not even a formal press conference. Instead, a Weibo account named "Linghui Auto" quietly registered and went online, followed by four familiar yet unfamiliar models listed in the Ministry of Industry and Information Technology's new car catalog.

This is probably BYD's quietest brand launch in recent years, because this newcomer is not responsible for creating surprises, but only for solving problems.

The emergence of Linklink signifies that BYD has officially separated its B2B (business-to-business) business from its main brand, establishing an independent operational unit. According to public information, this trademark was applied for registration as early as 2010. It's like a seed buried for sixteen years, finally sprouting only today, when BYD's annual sales have exceeded 4.6 million vehicles and its main brand urgently needs to move upmarket.

Old car gets a new look

A close examination of the product information disclosed by the Ministry of Industry and Information Technology reveals that Linkex's initial lineup consists of mature products already available on BYD's technology shelf, essentially a 35-year-old programmer finding new employment.

The first batch of four models announced—the Link e5, e7, e9, and M9—cover almost the entire range from compact sedans to mid-to-large MPVs. And their prototypes are all familiar faces we see on the streets.

The Link e5 will undoubtedly be a major force in the future travel market.

Its prototype is the Qin PLUS EV, which we are extremely familiar with. In the past few years, this car has almost become synonymous with making a living thanks to its extremely high cost performance and reliable electric powertrain system. The Linghui e5 inherits this legacy, and apart from minor adjustments to the grille details and a new brand logo, it basically retains the original car's outline.

For commercial vehicles, this unchanging nature is actually a virtue. It means they have a very high degree of parts commonality, keeping maintenance costs to a minimum. Equipped with a 100kW motor, it is more than enough for urban commuting, while also achieving a balance between range and energy consumption.

The slightly higher-end Link e9 and e7 demonstrate BYD's other ambitions in the B-end market—covering high-end travel scenarios.

The Linghui e9 is built on the body of the BYD Han DM-i, with a body length of nearly 5 meters and a wheelbase of 2920mm. This space advantage, originally intended to appeal to family users, has now been transferred to the ride-hailing and official vehicle market.

Similarly, the Link e7, derived from the upcoming Seal 06 EV, retains the spacious headroom and off-road capability of an SUV, while the headlights have been redesigned to differentiate it.

The most unique product is the Link M9.

This is the only plug-in hybrid model among the four vehicles, and also the only overseas model introduced to the Chinese market at this time. Its prototype is the export version of the BYD Xia, which was previously mainly targeted at overseas markets such as Mexico.

As an MPV, it is equipped with BYD's latest fifth-generation DM technology, consisting of a 1.5L engine and an electric motor forming a plug-in hybrid system. Demand for MPVs has always been strong in the business reception, hotel shuttle, and rental markets, but this market has long been dominated by older models from joint venture brands. The entry of the Link M9 is clearly an attempt to leverage the low operating costs of new energy vehicles to shake up this stagnant market.

From the perspective of product definition logic, Link Auto demonstrates an extreme pragmatism.

It doesn't have any fancy fancy features. Although the official statement claims that it will be equipped with the "Eye of the Gods" assisted driving system or the "Cloud Chariot" body control system in the future, the configuration of the currently declared models focuses more on the vehicle's economy, durability, and ease of maintenance.

For taxi companies, ride-hailing platforms, or government procurement departments, they don't need a 0-100 km/h acceleration time of 3 seconds or a complex sensor matrix. What they need is a robust, durable, long-range, and quick-repair tool that can generate revenue.

This strategy also represents an efficient use of resources.

After several years of market experience, the production mold costs for models like the Qin PLUS, Han, and Hiace have long been diluted, and their supply chains are highly mature. Assigning these models, which are in the later stages of their lifecycle or have ample production capacity, to Linke will not only allow them to continue to be utilized but also further reduce costs through economies of scale.

For passengers, the space and air conditioning in the back of the Link e5 are exactly the same as in the Qin PLUS EV; for drivers, the familiar driving feel and universal maintenance system mean they don't need to pay any extra learning costs.

Linklink was created to make the "utility vehicle" attribute even purer.

Link REIT is the support, BYD is the marksman.

Rebranding and relaunching several mature older models might seem like a natural progression. However, for BYD, establishing an independent brand at this juncture is far more complex than it appears on the surface.

Over the past few years, BYD has rapidly captured a huge market share with its blockbuster models such as the Qin, Han, and Dolphin. However, there are always two sides to every coin; a significant portion of these massive sales has flowed into the ride-hailing market. When you spend over 200,000 yuan to buy a Han DM-i and park it on the side of the road, someone might very well open your back door and utter those four mysterious numbers that will send your blood pressure soaring.

The confusion between the images of private cars and commercial vehicles is a persistent problem that plagues all car companies aspiring to move upmarket.

For BYD, this is a knot that must be untied. The large and stable order volume in the B-end market is the cornerstone for maintaining factory operations and spreading R&D costs, but if the B-end image is allowed to erode the C-end perception, the brand premium will never be able to rise.

This is the significance of Linkex's existence. It separates those models that are mainly used for "making a living" and allows BYD's main brands—Dynasty and Ocean—to finally free up resources to talk about luxury, individuality, and family warmth.

To understand the brilliance of BYD's move, let's look at the previous attempts of its competitor, GAC Aion. Aion was also eager to shed its "ride-hailing" label, but it chose a more difficult path—launching a new, higher-positioned, and more luxurious brand, namely Haobo.

Since Aion is considered a ride-hailing brand, I'll create a completely new and more expensive brand to prove that I can also be high-end.

But the market is ruthless, and building a completely new high-end brand is as difficult as building a skyscraper from scratch. Consumers are unlikely to change their perceptions in a short period of time and accept a million-dollar supercar that has sprung up from a ride-hailing brand. Haobo's arduous exploration in the market also proves how steep this upward climbing path is.

BYD is different. Instead of trying to cover up its past with a new brand, it lets the new brand take away the so-called "dirty work" and "hard work." This kind of opportunistic approach faces less resistance and yields results faster.

If the support does all the hard work, the marksman can focus on farming and carrying the whole game.

On the other hand, Link REIT is also revitalizing BYD's existing assets.

The molds, production lines, and supply chains for these initial models are all readily available, eliminating the need to redistribute huge R&D costs. While other brands are still struggling with the gross profit margin of new cars, Link REIT's approach of reusing existing platforms may allow them to profit from the very first car sold.

In the market environment of 2026, low cost will be the biggest advantage. If a price war really breaks out, Link REIT has the confidence to push prices even lower and snatch up those B2B orders that value cost-effectiveness the most.

This, in turn, protects the main brand. By transferring older technologies to Link REIT to generate revenue in the B2B market, new technologies—such as advanced intelligent driving systems and better motors and batteries—can be used more effectively in new Dynasty and Ocean models, truly widening the gap between "commercial" and "family" driving experiences.

#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 year's AI industry is somewhat like the smartphone market in 2008. Everyone knows that touchscreens are the future, but manufacturers are all making "Nokias with touchscreens".

The era of agents has arrived; that's a consensus. But how do you create a good agent model? Following conventional thinking, it might still be the same: more parameters, deeper and broader network structures, and larger datasets.

300B isn't enough, so we go with 1T; 1T isn't enough, so we go with 10T. It's as if as long as the model is made big enough, agent capabilities will naturally emerge, just like as long as Nokia's screen is big enough, the iPhone will appear on its own.

While large parameter models do offer a broader knowledge base and a more robust foundation, simply "optimizing" the previous generation of models will not make our agents any better. The Agent era needs agentic models .

Is there a more efficient, truly disruptive approach that doesn't rely on stacking parameters but on architecture optimization; doesn't require dedicated cloud servers and can run locally; and is both an all-rounder and has targeted optimizations?

On February 2, Step Star released and open-sourced its latest base model, Step 3.5 Flash, which is a more efficient underlying support model for agents. It adopts a sparse MoE architecture with a total of 196 billion parameters, but each token only activates about 11 billion parameters.

This is a rather counterintuitive statistic in the large-scale model industry, seemingly "lagging behind" compared to competitors vying for trillions of parameters. However, this seemingly "lagging" choice may hold the biggest secret of the Agent era.

The L3 model can no longer climb the ladder of L1.

If this had happened six months ago, Leap Star might have been doing something else entirely.

In his latest blog post, Zhu Yibo, co-founder and CTO of Step2, mentioned that during the Step2 model era, they were also staunch believers in the Scaling Law. Like all model manufacturers at the time, they diligently climbed the ladder called parameters, designing a larger number of parameters than DeepSeek V3, and even training it several months earlier than their competitors.

The result was that, despite its impressive benchmark scores, the traditional component stacking logic was ultimately outmatched by DeepSeek R1's inference paradigm.

The reason is simple: DeepSeek R1 represents a leap forward, from L1 Chatbots to L2 Reasoners. Continuing to use Chatbot thinking to build reasoning models may not necessarily lead to failure, but it is destined to hit a wall.

This is not only a retrospective of Jieyue Xingchen's experience, but also a microcosm of the entire industry. After much painful reflection, they discovered a neglected truth: Chatbots in the L1 era and Agents in the L3 era require two different models .

According to OpenAI's five-level architecture, we are experiencing a leap from L2 Reasoner to L3 Agent.

In the L1 Chatbot era , the core requirement was fluency in conversation; the model only needed to respond quickly and express itself naturally. What we might need was a "liberal arts student" who could recite encyclopedias, knowledgeable enough to effortlessly mimic Lu Xun's style. At this point, an output speed of 20-30 tokens per second perfectly matched human reading habits.

The L2 Reasoner era : With the emergence of long thought chains, we need models that can represent complete, deep thinking. Looking at long thought processes, we think they are intelligent, and we are even willing to wait tens of seconds for a more accurate result.

The L3 Agent era : the characteristics have completely changed. The context of the work scenario resides in the 32K-128K range. We no longer read the output word by word, but only focus on "when can the results be delivered".

In this scenario, continuing to use the heavy-parameter model from the L2 era is like taking the equipment from the Aotai Line to climb the Sheshan "Hutai Line". Although the reserves are sufficient, the efficiency is low and the computing power cost is extremely expensive.

To some extent, the agent can no longer be seen by the user, but rather used for the task. It needs to run efficiently in long contexts, with a codebase capable of easily handling hundreds of thousands of tokens; and it needs to improve speed, which directly determines the user experience; but the core is still planning and tool invocation.

If we continue to use heavy-duty models from the L1 era to run L3 tasks, it's like driving a Ferrari to deliver food—not only is it expensive, but it also simply can't run in congested, long-context road conditions.

This also explains why Step Tech dared to go against the grain and unveil the "new species" of Step 3.5 Flash, focusing on "speed" and "strong logic." This trade-off may seem out of step with the industry at first glance, but it opens up more possibilities.

In the world of martial arts, speed is the only way to break through defenses: the "violent aesthetics" of the Agent era.

In the Agent era, "speed" is no longer an added bonus, but a matter of life and death for the model.

Zhu Yibo pointed out a detail that is easily overlooked: in the chatbot era, it was enough for the model output to be faster than human reading speed (20-30 tokens/s), otherwise we wouldn't be able to read it all. But in the agent era, this standard is completely invalid .

Why? Because users don't want to see the process. When AI helps us write code, search for information, or book flights, we don't stare at the screen watching it transcribe word by word; we just want the result.

At this stage, speed is no longer a matter of experience, but rather a matter of productivity itself, directly determining the efficiency of task delivery.

To achieve this ultimate speed, LeapStar took a huge gamble on its technological path.

While its peers were blindly following the trend of Linear Attention, Step 3.5 Flash insisted on choosing the SWA (Sliding Window Attention) architecture. This hybrid attention layout can process token calculations faster on the one hand, and solve the secondary bottleneck of long context processing on the other.

Simply put, it doesn't memorize the entire 256K text by rote; instead, it allocates attention in a focused and rhythmic manner, much like a human. This allows it to not only avoid becoming less intelligent when processing massive amounts of data but also significantly reduce computational costs.

What appears to be a "reverse approach" is actually the sophisticated calculation of the Agent era, where "skillful methods yield great results." This is because, under current hardware conditions, SWA is most favorable for speculative sampling. This technical trade-off directly boosts the inference speed of single-request code-based tasks to a maximum of 350 tokens/s .

The lightning-fast "instant kill" is the decisive moment that transforms AI from a "toy" into a productivity tool. On the first day of its release, Step 3.5 Flash made it onto the OpenRouter Fastest Models list .

Reject "memorization-based learners"; high intelligence is the primary productive force.

Running fast cannot come at the cost of "lowering intelligence." " High intelligence " is also essential to measuring whether a model is suitable as an agent.

The general consensus, whether among users or most model manufacturers, is that the larger the parameters, the stronger the capabilities. However, Step 3.5 Flash's top-ranking performance in the mathematical field, achieved with appropriate dimensions and extreme post-training , yielded results comparable to models with large parameters.

• It scored 97.3 points in AIME 2025 (American Invitational Mathematics Competition);
• He scored 85.4 points in the IMO Answer Bench (International Mathematical Olympiad Benchmark);
• In HMMT 2025 (Harvard-MIT Mathematics Competition), he even scored 96.2 points .

What does this mean? These scores are all the highest among top-tier open-source models in China.

If the Parallel Collaborative Reasoning (PaCoRe) mode is enabled, its score even approaches perfect. Behind this phenomenon of "intelligence overflow" lies an extremely subtle but accurate industry truth: past models were like "memorizers," relying on rote memorization of massive amounts of data to get by; while Step 3.5 Flash is a true "problem solver."

In an agent's workflow, this capability is fatal. Because real-world tasks are full of unknowns, what we need is not a parrot that can only repeat knowledge points, but a "super brain" that can understand complex instructions, break down task logic, and self-correct.

Reasoning ability proves intelligence, but the agent still needs to be reliable in its work. Step 3.5 Flash achieved first place in domestic open source in several key scenarios.

Coding ability: Top tier globally

• SWE-bench Verified: 74.4 points (Bug fixes from a real open-source project)
• Terminal-Bench 2.0: 51 points ( China's No.1 open source , automating terminal tasks)
• LiveCodeBench-V6: 86.4/88.9 points ( Number one open source code in China , for real-time coding and debugging)

Agent core capabilities: Multiple domestic open source firsts

• τ²-Bench: 88.2 points ( No. 1 open source in China , multi-step task planning)
• xbench-DeepSearch: 54 points ( Number one open source in China , for deep search and information integration)
• BrowseComp: 69 points (Top tier, web browsing and context management)

No matter how impressive the data is, it must withstand the test of real-world scenarios . In the following typical scenarios, Step 3.5 Flash also proved that "skillful yet powerful, fast as lightning" is not just a slogan.

While it's commonly believed that writing analysis reports with Deep Research requires models that are well-written and eloquent, it actually relies on strong logical reasoning and the ability to utilize tools effectively.

Give it a vague topic, such as "scientific education for infants and toddlers aged 0-3 years," and it won't just make things up. Instead, like a real human researcher, it will break down the task, plan the path, search online, reflect and revise, and then give us a detailed report of tens of thousands of words that even novice parents can understand.

In Scale AI's Research Rubrics benchmark, its score even surpassed similar systems from OpenAI and Gemini. This further demonstrates that it has already achieved a "logical closed loop" capable of operating independently.

Step 3.5 Flash can also be integrated into the Claude Code environment. When the model acts as a professional data analyst and faces complex data analysis tasks, it can not only write code to clean the data, assist in daily data processes, and align data formats, but also directly generate workflow reports.

Whether it's Deep Research or Vibe Coding projects, these are either done on Step 3.5's official website or by calling APIs, but Step 3.5 Flash's ambitions go far beyond simply pulling AI capabilities from cloud servers.

Zhu Yibo revealed that he even bought a device out of his own pocket to run the model. Currently, Step 3.5 Flash is the most powerful model available, bar none, capable of running smoothly with 256K ultra-long context on a 128GB RAM MacBook using 4-bit quantization.

This phrase, "without exception," truly highlights the stubbornness of technologists. Perhaps it also hints at the ultimate ambition of Leapfrog Space's "AI + Terminal" strategy: the most powerful brain shouldn't just live in expensive H100 clusters; it should live in your computer, and even in future mobile phones .

While other manufacturers are still vying for funding and valuations, LeapStar has quietly lowered the cost barrier for high-performance agents once again. This confirms the strategic prediction: the "show" of large-scale model competition is over, and the industry has officially entered a "survival of the fittest" phase.

Another dark horse has emerged in the Spring Festival AI battle, shaking up the landscape of big data models.

In the recent noisy battle of AI releases, Step 3.5 Flash, this dark horse, has definitely been somewhat underestimated. It's not just a "high-performance, cost-effective" model; rather, it's somewhat like DeepSeek's sudden emergence a year ago, paving a new path for the AI ​​industry.

In the real world where computing power is not unlimited, the real winner is who can solve more complex problems with a more sophisticated architecture and fewer resources.

The "rags-to-riches" era, where simply piling up specs could secure funding and manipulating rankings could earn applause, is long gone. The battles to come belong to those who are not only "smart" but also "quick-witted."

This insistence on "miniaturization and high efficiency" essentially stems from StepStar's dedication to the mission of AGI. Zhu Yibo once remarked that the significance of persisting in training the basic model, besides the commercial advantages, lies in fulfilling that "long-held AGI dream."

For LeapStar, the path to AGI is not about taking a big gamble, but about a forward-thinking methodology and an accurate assessment of the needs of the times. As Yin Qi, the newly appointed chairman of LeapStar, said in an interview, "Developing solid basic models and exploring the upper limits of intelligence is LeapStar's mission."

From Step 1 to Step 3.5, from multimodal to voice, from the cloud to the terminal, the integration of AI with physical space and the deep integration with terminal hardware are all necessary steps to leapfrog the stars and reach the ultimate AGI.

When AI capabilities truly "enter ordinary households," the end point of the technology competition will no longer be the arms race of computing power. AI will begin to better serve us, and everyone—SMEs, individual developers, and students—will be able to use top-tier agent capabilities at low cost. AGI will then no longer be just a game for giants.

With ingenuity and speed , Step 3.5 Flash offers its answer, and represents another step closer to AGI's mission.

Those who persist in their dreams will eventually forge their own path. And that path is becoming increasingly clear.

#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

In January, Geely Automobile surpassed BYD to become the top-selling Chinese independent brand, with total sales reaching 270,167 units, a 14% increase compared to the previous month.

The Geely Galaxy brand played a crucial role in this success, with sales reaching 82,990 units in January, accounting for 67% of total new energy vehicle sales and 31% of Geely Automobile's total sales. The sales were mainly contributed by the Geely Star Wish and Galaxy Starship 7 EM-i, with 41,676 and 11,816 units respectively (both including exports).

Meanwhile, the Geely Galaxy M9, positioned at a higher level, has consistently sold over 10,000 units since its launch, and the newly launched Starship 7 EM-i has also performed well. However, in the core 100,000-150,000 RMB family SUV market, Geely Galaxy still needs a product that can compete with its old rivals.

This is the newly unveiled Geely Galaxy M7.

As a mid-size plug-in hybrid SUV, the new car is planned to be launched in the first quarter of 2026, with an expected price range of 100,000 to 130,000 yuan, directly competing with plug-in hybrid SUVs such as BYD Song PLUS DM-i, Song Pro DM-i, and Haval Menglong.

Geely summarizes the selling points of the Galaxy M7 as three "golden" features: "golden brick battery", "golden range system" and "golden driving control system".

Among them, the "Gold Brick Battery" and the "Gold Calculation Endurance System" both point to the core competitiveness of Geely Galaxy M7 – its endurance performance.

The Geely Galaxy M7 is equipped with the EM-i plug-in hybrid system, consisting of a 1.5L hybrid-specific engine and a front-mounted drive motor, and offers two lithium iron phosphate battery pack options: 18.4 kWh and 29.8 kWh.

Official data shows that the new car has a maximum pure electric range of 225 km and a combined range of 1730 km with a full tank of gas and a full battery. In a low-battery state, the fuel consumption is 4.69 L/100 km, and in a fully depleted state, it can be as low as 3.35 L/100 km.

In the 150,000 RMB plug-in hybrid SUV segment, a pure electric range of 225 km is quite competitive. For reference, BYD's previously launched 2026 Song Pro DM-i long-range version has a CLTC pure electric range of 220 km.

In terms of dimensions, the Galaxy M7 is positioned as a five-seater mid-size SUV, with a length, width and height of 4770 × 1905 × 1685 mm and a wheelbase of 2785 mm. Compared with the current Galaxy L7, the body length has increased by 60 mm.

Compared to its competitor, the BYD Song PLUS DM-i, the Galaxy M7 is 5 mm shorter in length but has a 20 mm longer wheelbase, making the overall size of the two vehicles similar.

In terms of design, the Galaxy M7 adopts the latest family design language of Geely's Galaxy M series.

The new car features a closed grille with a continuous LED daytime running light strip at the front, which forms a cohesive visual connection with the upward-sweeping headlights on both sides. Chrome trim is embedded in the air intakes on both sides, and the air intakes on both sides below are closed and decorated with vertical silver chrome trim, making the overall look more refined.

The side profile continues the styling of a fastback coupe SUV, with the roofline naturally sloping downwards and connecting to the roof via a blackened D-pillar, creating a floating effect; the waistline rises in the opposite direction at the rear door and extends to the rear of the vehicle, enhancing the dynamic expression of the side profile.

The rear of the new car retains the family's common through-type taillights, and the lower part of the rear bumper features a two-section air deflector to enhance the sporty visual appeal.

Stepping inside, the Geely Galaxy M7 features a large floating touchscreen in the center console, equipped with the Flyme Auto intelligent cockpit system.

The center armrest area inside the car retains a large number of physical buttons and crystal-textured knobs, which enhances both practicality and the luxurious atmosphere.

In terms of intelligent features, some high-end models will be equipped with the "Qianli Haohan H3 Intelligent Assisted Driving System", which supports functions such as highway navigation, automatic lane changing and intelligent parking.

Regarding the chassis, the official statement emphasizes that it will continue the tuning experience of the Lotus team, attempting to strike a balance between handling quality and the comfort orientation of a family SUV, corresponding to what it calls the "Golden Driving Control System".

In terms of product timing, the launch of the Geely Galaxy M7 is largely intended to succeed the Galaxy L7.

As a former mainstay model, the Geely Galaxy L7 has seen sluggish sales over the past year, with only 189 units sold in December last year.

In the current plug-in hybrid SUV market priced around 150,000 yuan, the BYD Song Pro DM-i has consistently achieved monthly sales of over 10,000 units for the past year, with peak monthly sales exceeding 20,000 units. The Changan Qiyuan Q07 has also broken through BYD's dominance with its cost-effectiveness, maintaining stable monthly sales of around 10,000 units.

In this environment, if Geely Galaxy M7 wants to stand out from the competition, in addition to leading in a single parameter, it may also need to respond to market expectations with a surprisingly attractive price.

Geely Automobile's sales target for 2026 is 3.45 million vehicles, representing a year-on-year increase of approximately 14%. Of these, the Geely brand aims for 2.75 million vehicles; the JK brand for 300,000 vehicles; and the Lynk & Co brand for 400,000 vehicles.

With plans to launch one to two new products every quarter, 2026 will still be a period of intensive product launches for Geely.

If the Geely Galaxy M7 can successfully break through this goal, it will become one of Geely Galaxy's most important volume-selling models in 2026.

#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

SpaceX announces acquisition of xAI, creating a $1.25 trillion space and AI giant.

Early this morning, SpaceX issued a statement announcing the completion of its acquisition of xAI, integrating the two companies' businesses into a vertically integrated system covering rocket launches, space internet, direct-connect mobile communications, and artificial intelligence computing.

The statement indicated that the merged company will cover businesses such as rocket launches, space internet, direct-connect mobile device communications, and real-time information platforms.

Musk stated that within the next two to three years, "AI computation in space will become the lowest-cost method," which will enable companies to train models and process data more efficiently, thereby accelerating physics research and the invention of new technologies.

Bloomberg, citing sources familiar with the matter, reported that the merged company is valued at $1.25 trillion (approximately 8.7 trillion yuan at the current exchange rate) and is expected to issue shares at $526.59 per share. SpaceX still plans to proceed with its initial public offering this year, which could reach $50 billion.

Previously, xAI completed a financing round in January of this year, reaching a valuation of $230 billion; SpaceX, on the other hand, was valued at approximately $800 billion in its share sale plan last December.

The report also mentioned that SpaceX is seeking regulatory approval to launch up to 1 million satellites into Earth orbit to support its space data center concept.

The merger will further strengthen synergies in capital, computing power, and talent among Musk's businesses, including the social platform X, which has already merged with xAI, and its chatbot Grok. xAI currently spends approximately $1 billion per month, and the merger is seen as an important step in alleviating its high cost pressures.

As one of Musk's most stable businesses, SpaceX is currently the only company in the United States capable of routinely carrying out manned missions to the International Space Station, and is also an important launch service provider for NASA and the U.S. Department of Defense.

Its Starlink network has deployed more than 9,000 satellites, resulting in significant revenue growth and is considered an important source of cash flow to support the xAI business in the future.

Jensen Huang: "Never promised" to invest $100 billion in OpenAI

According to Bloomberg, Nvidia CEO Jensen Huang recently stated in an interview that the widely circulated plan to invest $100 billion in OpenAI "was never a promise," and emphasized that Nvidia will evaluate each round of financing "step by step."

He pointed out that OpenAI had invited Nvidia to invest up to $100 billion, which the company was "honored" to do, but the actual investment would be made gradually depending on the situation.

According to reports, in a letter of intent signed last September, Nvidia planned to support OpenAI in building data centers with a capacity of at least 10GW and provide advanced chips for training and deploying AI models.

However, The Wall Street Journal previously reported that there were doubts within Nvidia about the plan, with some people worried about the lack of discipline in OpenAI's business model and feeling uneasy about the competitive landscape.

Jensen Huang responded that the claims were "nonsense" and emphasized that he still believes in the importance of OpenAI, calling it "one of the most influential companies of our time."

Jensen Huang added that Nvidia's investment in the current funding round will not be close to $100 billion, but the overall investment will still be "very large" and may even become the largest investment in the company's history.

He also stated that the deployment progress of the first 1GW data center in the collaboration "depends on OpenAI," as the relevant infrastructure is led by OpenAI.

With the upcoming releases of the Zhipu GLM-5 and MiniMax M2.2, the Spring Festival is expected to be a peak season for large-scale model releases.

According to the South China Morning Post, in the final sprint before the Spring Festival, many cutting-edge artificial intelligence laboratories in China are launching new-generation large models in an attempt to seize exposure and user mindshare during the holiday.

Following Alibaba's launch of the Qwen3-Max-Thinking and Kimi 2.5 last week, Zhipu AI and MiniMax are also rumored to be updating their flagship models within the next two weeks.

According to sources, Zhipu AI plans to launch GLM-5 before the Spring Festival. This is the fifth generation iteration of the GLM series, and it is expected to bring "comprehensive and significant" upgrades in creative writing, programming, reasoning, and intelligent agent capabilities.

MiniMax will release M2.2, a minor update to M2.1, with a focus on enhancing programming capabilities.

In contrast, DeepSeek will not be releasing any of the "big surprises" that people are expecting during this year's Spring Festival.

Multiple sources indicate that DeepSeek is more likely to release only a minor update to the V3 series. Its next-generation flagship model is expected to be a trillion-parameter base model, but its release has been delayed due to training slowdowns caused by its massive scale.

In addition, ByteDance will launch a "three-piece suite" during the Spring Festival: the large language model Doubao 2.0, the image generation model Seedream 5.0, and the video generation model SeedDance 2.0. Alibaba is expected to release its flagship model Qwen 3.5 during the Spring Festival, focusing on strengthening complex reasoning, mathematical, and coding capabilities.

Meanwhile, the competition among tech giants for users during the Spring Festival has entered a fierce stage.

Giants like Alibaba, Tencent, and Baidu are investing huge resources to drive the growth of AI applications: Tencent's "Yuanbao" will distribute 1 billion yuan in digital red envelopes, Baidu will distribute 500 million yuan in red envelopes through its Wenxin App, and Alibaba also announced yesterday that it will invest 3 billion yuan to promote its Qianwen App.

OpenAI releases macOS version of its Codex application.

Early this morning, OpenAI officially launched its new Codex desktop application, further strengthening its competitiveness in the field of AI coding assistants.

Currently, the Codex application is temporarily available to all ChatGPT users on Mac, including free users and Go subscribers. To attract existing developers, OpenAI has also temporarily increased the rate limits for paid plans.

OpenAI states that the Codex application is designed to be a "command center" for developers, allowing them to manage multiple AI agents in a single interface, support parallel task execution, automate repetitive tasks, and provide customizable interaction styles.

The application integrates the GPT-5.2-Codex model, enabling complex software development processes to be completed within hours. The company claims that its internal team once relied on Codex to complete the development of Sora for Android in 28 days.

Fortune reports that Codex's launch comes at a time of intensifying competition in the AI ​​coding assistant market. Anthropic released a similar app, Claude Code Cowork, in January, and Cursor is also experiencing rapid growth.

OpenAI says that over 1 million developers have used Codex in the past month, including large companies like Cisco and startups like Harvey and Sierra.

Tesla has officially announced that its third-generation humanoid robot will be unveiled this year, with an estimated annual production of one million units.

Yesterday, Tesla announced that its third-generation humanoid robot is about to be unveiled and confirmed that the product will be officially released this year. Tesla stated that this generation of robots has been redesigned from first principles and can learn new skills by observing human behavior, with a target annual production of one million units.

According to reports, the new generation of robots can complete tasks through demonstrations, verbal descriptions, and even video examples, and Musk says its performance will "surprise everyone."

The official statement emphasizes that this will be Tesla's first truly mass-produced humanoid robot. Previously, a pilot production line for Optimus was already running at the Fremont factory, and a larger, third-generation production line is expected to be completed and put into operation this year.

Previously, Tesla's official account, TeslaAI, posted on Weibo that the cost of a single third-generation humanoid robot after mass production is expected to be controlled within $20,000 (approximately RMB 139,000).

Audi admits renaming the A4 to A5 was a mistake and will revert to the traditional naming system.

According to Australian media outlet Drive, Audi has confirmed that its strategy of renaming its classic A4 model to A5 was a "mistake" and plans to restore the A4 name in the next facelift.

Audi's current CEO, Gernot Döllner, recently stated in a media interview at the Munich Motor Show: "Yes, it was a mistake, and we have corrected it."

We will revert to our original naming system: A represents sedan, Q represents SUV, and the numbers correspond to vehicle size and market segment.

The report points out that Audi had attempted to differentiate its models by using odd numbers for gasoline-powered vehicles and even numbers for electric vehicles in its showrooms. As a result, the A4 was merged into the A5 series, the A6 was planned to be renamed A7, and the high-performance S/RS series were adjusted accordingly.

However, this system quickly caused confusion in the global market: dealer inquiries increased, consumers struggled to understand the relationship between the A5 and the original A5 coupe, and the A6L in the Chinese market, due to its special status, did not change its name, further exacerbating the inconsistency in the system.

In interviews with Autocar and Drive, Dorner emphasized that the decision to rename the car was made "too late," which was one of the key mistakes, as the production run was rushed with only a few months to go.

Regarding whether the A4 name will be reinstated in the future, he clearly stated that "it is possible," and used A6 as the anchor point for the future naming system.

At the same time, Audi's electrification strategy is also being adjusted. As the brand slows down its plan to "sell only electric vehicles in most markets from 2033 onwards," the positioning of the A4 name, which was originally intended to be used only for pure electric sedans, has become less clear.

Audi's global sales declined by 11.8% in 2024, with electric vehicle sales falling by 8% year-on-year, which also became one of the background factors for the reorganization of the naming system.

Yu Chengdong confirms Wenjie M6 will debut this spring.

Yesterday, Yu Chengdong, Huawei's Executive Director and Chairman of the Consumer Business Group, announced on Weibo that the Huawei M6 will be unveiled this spring. Prior to the official announcement, winter testing images of the M6 ​​had already been released.

Based on publicly available images, the new car continues the brand's design language, but the front fascia does not feature a continuous daytime running light strip, presenting a more distinctive visual style. The body lines are rounded, and the roof-mounted lidar indicates that it will be equipped with Huawei's advanced intelligent driving assistance system, ADS.

Last December, Yu Chengdong revealed that the product matrix of the five "Boundaries" under HarmonyOS Intelligent Mobility would be completed this year, with the Wenjie M6 being a key component. In addition, the Wenjie brand will also promote the iterative upgrade of existing models, while the "Zunjie" brand also plans to launch an ultra-high-end luxury sedan.

Alipay's 2026 "Collect Blessings" campaign launches today: a new "Health Blessing" has been added, allowing users to receive blessings simply by "touching" each other offline.

Today, Alipay's "Collect Blessings" campaign officially launched. This year's campaign adds a "Health Blessing" to the traditional "Five Blessings," which is distributed by the health AI application "Ant Fortune." The campaign will continue until Lunar New Year's Eve.

After users enter the "Collect Blessings" event from the Alipay homepage and claim the "Health Blessing," they will simultaneously receive a health red envelope that arrives in real time and can be used for offline Alipay payments within 7 days.

In addition to the new lucky cards, the lucky card collection activity has been further expanded this year.

• Online, users can scan the character "福" (Fu, meaning good fortune) or physical objects or photos containing the character "马" (Ma, meaning horse) or horse elements through the "灵光扫福" (Lingguang Scan Fu) function in the Alipay App to obtain Fu cards.
• Offline, you can also trigger the acquisition of lucky cards by "tap" in various scenarios such as payment, picking up express delivery, riding a bike, charging, entering a park, and visiting an exhibition.

This year's event features 19 sets of themed lucky cards, covering the classic Five Blessings and multiple IPs such as games, trendy toys, and movies. Users can choose any 6 sets to collect. Collecting a complete set will lock in a 1.68 yuan red envelope. All red envelopes will be drawn at 22:18 on New Year's Eve.

During the event, experiencing the "Ant Fortune" feature will give you a chance to win rare cards and universal fortune.

Since its launch in 2016, Alipay's "Collect Blessings" campaign has been held for 11 consecutive years, with over 1 billion users participating. The project leader stated that the campaign interprets tradition through technology and has become a "digital New Year custom" for hundreds of millions of users.

Tencent responds to allegations of using Yuanbao red envelopes to boost viral sharing: No-threshold red envelopes are different from incentivized sharing.

According to The Paper, Tencent's Yuanbao "Spring Festival Red Envelope" campaign has recently gone viral on multiple social media platforms, sparking discussions about whether the sharing mechanism violates WeChat's external link management regulations.

The report points out that after the event was launched, users could get extra chances to win prizes by sharing, up to a maximum of 30 chances. This mechanism promoted the rapid spread of the red envelopes on social networks.

In response to external criticism, Tencent issued an internal letter yesterday stating that the basic logic of the Yuanbao Spring Festival red envelope is "no threshold for receiving". Users can receive the basic red envelope without completing tasks such as helping or collecting cards, which is fundamentally different from the "inducement to share" model that the platform has been cracking down on for a long time.

Tencent stated that the team is in communication with legal, public relations, and relevant business departments to comprehensively assess the compliance boundaries of the activity and will continue to monitor user feedback and experience.

According to Beike Finance, the popularity of the event has also spawned gray market services.

On second-hand trading platforms, a large number of "Yuanbao Red Envelope Assistance Services" have appeared, with a single assistance costing about 1 yuan, and some merchants offering packages ranging from 10 to 13 yuan.

Reporters' tests revealed that the assistance service is usually completed within 10 minutes and can increase the number of chances to win a prize by 10. However, since the amount of a single cash red envelope is generally less than 1 yuan, the overall benefits are difficult to cover the expenses.

Some scalpers who run concert ticket-grabbing businesses have also joined in the assistance services, and even formed "assistance groups" to promote the services.

Apple reportedly patched the AirDrop vulnerability in Final Cut Pro.

According to X user @nikewrayy, Apple may be further tightening the licensing and verification mechanism for professional software, and Final Cut Pro sent via AirDrop is no longer available.

The leaked screenshots show that when a user attempts to run a copy of Final Cut Pro that is not authorized by the current Apple ID, the software is blocked during the "Validating Purchase…" stage on the launch page, and a warning pop-up appears.

No valid App Store purchase record found. Please delete this version of Final Cut Pro and download it from the App Store.

Previously, some users discovered that they could use AirDrop to transfer the Final Cut Pro.app package they had purchased from one Mac to another from the Apple Store and run it directly without logging into the purchase account.

It's worth noting that Apple just launched the Apple Creator Studio subscription service last month, which integrates apps including Final Cut Pro, Logic Pro, and Pixelmator Pro, as well as new AI features and premium content from the iWork suite, into a single subscription.

It is currently unclear whether this verification mechanism will be simultaneously deployed to other Apple professional creative software such as Logic Pro and Motion.

Zhuji Power raises $200 million in Series B funding, with additional investment from JD.com and NIO Capital.

Yesterday, LimX Dynamics announced the completion of a $200 million Series B funding round.

This round of financing was led by several well-known domestic and international institutions, including Stone Venture (UAE), Eastern Fortune Capital, Cornerstone Capital, and GF Securities, with participation from strategic industry investors such as JD.com and Zhongding Group. Existing shareholders, such as SAIC Motor's Shangqi Capital and NIO Capital, continued to increase their investment.

Zhuji Power stated that this round of financing will be used to promote breakthroughs in the core technology of embodied intelligence fusion of the cerebellum and cerebrum, as well as original innovation in general-purpose humanoid robots and modular base robots.

The company will continue to increase its R&D and marketing investment this year, and further expand into the Chinese and global markets through product engineering and supply chain ecosystem construction.

Founded in 2022 and headquartered in Shenzhen, Zhuji Power is an AI-driven humanoid robot company that has developed a variety of products, including full-size humanoid robots and multi-form robots such as TRON.

Oracle plans to raise up to $50 billion this year to bolster its cloud infrastructure development.

According to Bloomberg, Oracle plans to raise $45 billion to $50 billion this year through debt and equity financing to expand its cloud infrastructure to meet the contracted computing power demands from large customers, including AMD, Meta, Nvidia, OpenAI, TikTok, and xAI.

The company stated that the size of this round of financing reflects the financial pressure of building cloud computing infrastructure driven by AI.

The report indicates that Oracle will raise funds through two methods:

• Half of the proceeds came from mandatory convertible preferred stock, common stock, and an on-exchange offering of up to $20 billion.
• The other half comes from a one-off unsecured bond issuance in early 2026, led by Goldman Sachs with Citigroup handling the equity-related issuance.

Last year, Oracle issued $18 billion in bonds, one of the largest corporate bond issuances of the year.

It is worth noting that Oracle's stock price has fallen by about 50% since it hit an all-time high last September, wiping out about $460 billion in market value.

Analysts point out that the ongoing AI investment boom has raised concerns about the payback period. Oracle's free cash flow has turned negative due to data center construction and is expected to remain so until 2030. The company will still need to invest tens of billions of dollars in chip procurement and data center leasing over the next few years.

The report also mentioned that OpenAI has committed to investing approximately $300 billion in rented servers on Oracle Cloud, but it has not yet turned a profit, making outsiders more cautious about Oracle's capital expenditure return cycle.

Market concerns about Oracle's credit risk also pushed its credit default swap prices to their highest level since the 2008 financial crisis in December last year.

Analysts believe that Oracle's timing in releasing this news on Sunday is quite unusual, and may be intended to send a stabilizing signal to investors before the market opens this week in order to curb the continued decline in its stock price.

The father of the iPod: Re-examining privacy risks after becoming a father

According to Business Insider, Tony Fadell, the "father of the iPod," recently stated in a podcast interview that becoming a parent has led to a significant shift in his and many other Silicon Valley tech founders' views on privacy issues.

He pointed out that before having children, many tech professionals were more radical about privacy and willing to sacrifice personal data in the face of technological innovation; however, this attitude is changing in the face of risks such as deepfakes and social engineering.

Fadell noted that Meta CEO Mark Zuckerberg, Google co-founder Larry Page, and Sergey Brin have "completely different" ways of understanding the world after becoming parents.

My thoughts were very different before and after having children.

He said that many founders are now rethinking how much data they are willing to give up and how to protect the privacy of their families and children.

Fadell particularly emphasized the privacy challenges of the AI ​​era.

He believes that truly revolutionary AI devices of the future will often require a large amount of personal data and real-time input, which will force social and business leaders to make a more difficult trade-off between innovation and privacy.

He revealed that some tech founders even reflected on the past, wishing they could do it all over again, but the decisions made in the past could not be reversed.

At the same time, global regulators are increasing their scrutiny of AI and privacy issues.

xAI has faced investigations in multiple locations for its models generating unauthorized sexualized images of real people, including minors; Meta has also been questioned about how its chatbots interact with minors. The tension between privacy protection and the development of AI technology is rapidly becoming apparent.

The OPPO Find N6 has passed certifications in multiple countries and is expected to be launched soon.

According to My Mobile, the OPPO Find N6 foldable phone has recently passed multiple certifications, including those from TDRA in the UAE, IMDA in Singapore, and TUV, all with the model number CPH2765.

Meanwhile, TUV certification revealed the device's charging specifications, showing that it supports 80W wired fast charging, consistent with previous 3C information.

Regarding the specific configuration, based on recent leaks, the following is a summary:

• Equipped with a Snapdragon 8 Elite Gen5 processor;
• It features an 8.12-inch 2K LTPO UTG main screen and a 6.62-inch secondary screen;
• The rear camera features a triple-lens setup with 50MP + 50MP + 200MP sensors, and also includes a 2MP multispectral lens.
• It features a built-in "6K-level" dual-cell battery solution with a total typical capacity of approximately 6000mAh;
• It weighs approximately 225g.

It is worth noting that this device will be the first to support an AI stylus and will offer two versions: a standard version and a Beidou satellite communication version.

Multiple sources indicate that the Find N6 will most likely be released in the Chinese market after the Spring Festival, followed by a simultaneous launch in overseas markets. Following the pattern of last year's Find N5, OPPO may continue its strategy of releasing the phone in China on the same day as globally.

Step Star releases a faster and more powerful open-source agent base model, Step 3.5 Flash.

Yesterday, Step Star announced the launch of its new open-source foundation model, "Step 3.5 Flash," which boasts faster inference speed, stronger agent capabilities, and more stable performance in long-chain tasks.

• It adopts a sparse MoE architecture with a total of 196 billion parameters, and a single token only activates about 11 billion parameters;
• By introducing MTP-3 technology, the model can predict 3 tokens at a time, thereby significantly improving generation efficiency.
• For long text processing, a hybrid architecture of 3:1 sliding window and global attention (SWA + Full Attention) is adopted, which can efficiently process 256K context and maintain stability in complex tasks.

Official data shows that Step 3.5 Flash inference speed can reach up to 350 TPS in single-request code tasks. In agent scenarios and mathematical tasks, its performance is close to that of closed-source models.

Official benchmark charts show that enabling Parallel Thinking resulted in further performance improvements in tests such as XBench-DeepSearch.

Step 3.5 Flash states that its goal is not only to be "smarter," but also to be "trustworthy," "responsive," and "cost-controlled."

The official documentation also showcased several application examples, including high-speed mathematical reasoning, an automated programming platform for generating 3D Earth visualizations, and the ability to break down price comparison tasks in edge-cloud collaborative scenarios, integrate data from Taobao, JD.com, and Pinduoduo, and provide purchase recommendations, demonstrating its agent capabilities in complex task planning and execution.

StepStar also announced that training for its next-generation model, Step 4, has begun.

GitHub: https://github.com/stepfun-ai/Step-3.5-Flash/tree/main

Hugging Face: https://huggingface.co/stepfun-ai/Step-3.5-Flash

Lingguang App has a new feature: upload images to generate apps, and the API capabilities have been fully enhanced.

Yesterday, Lingguang, a general AI assistant under Ant Group, announced a comprehensive upgrade to its core function "Flash Apps," officially launching two new capabilities: "Upload Image to Generate Apps" and "Desktop Widgets," and integrating nearly 20 API tools.

According to reports, this update aims to accelerate the evolution of AI applications from "information providers" to "interactive tool deliverers" by lowering the development threshold and expanding the interaction dimensions.

In the newly added "Upload Image to Create Application" function, the system uses multimodal technology to intelligently analyze image content, enabling UI interface replication, structured extraction of key information, and scene style recognition.

Users can simply upload static images such as hand-drawn sketches, student grade sheets, or product sales rankings, and combine them with personalized instructions to transform them into online tools with data visualization and collaboration functions with a single click.

This upgrade also significantly expands the underlying capabilities, adding nearly 20 APIs including sound effect synthesis, Large Language Model (LLM) calls, calendar services, text-to-speech, persistent storage, gyroscope, and vibration feedback. The opening of these interfaces allows users to build applications with more complex logic and interactive experiences.

This update also opens up hardware sensor permissions, bringing the possibility of motion-sensing interaction to lightweight applications.

By utilizing the phone's gyroscope and vibration feedback module, developers can create mini-games or simulation applications with immersive experiences (such as "Hand Relaxation Simulator"), triggering customized sound effects and physical feedback at scoring or completion points.

Data shows that as of December 2025, users have created more than 12 million flash applications on the Lingguang platform.

The 2026 Spring Festival travel rush has begun: inter-regional travel is expected to reach 9.5 billion trips.

According to Xinhua News Agency, the 2026 Spring Festival travel rush officially started yesterday and will last for 40 days until March 13.

The report points out that thanks to the 9-day "extra-long" holiday, the demand for returning home and traveling will overlap, and the total number of cross-regional personnel flows in the whole society is expected to reach 9.5 billion, setting a new historical record.

Xinhua News Agency predicts that driving will remain the main mode of transportation, accounting for about 80%; railway and civil aviation passenger volumes are expected to reach 540 million and 95 million respectively, with the overall scale and daily peak expected to exceed the historical level for the same period.

Qianwen App invests 3 billion yuan to launch a free order campaign during the Spring Festival.

Yesterday, Qianwen App announced a 3 billion yuan investment to launch the "Spring Festival Invitation Plan," offering free meals and entertainment to people across the country during the Spring Festival, allowing them to experience a new lifestyle in the AI ​​era.

Alibaba's ecosystem businesses, including Taobao Flash Sale, Fliggy, Damai, Hema, Tmall Supermarket, and Alipay, will also join the Spring Festival offensive.

According to the official announcement, the plan will be officially launched on February 6. In addition to large cash red envelopes, the core selling point is "eating, drinking and playing, with non-stop free meals".

It is understood that this investment of 3 billion yuan is not only the largest in Alibaba's history of Spring Festival activities, but also the highest amount in the AI ​​competition among major companies during this year's Spring Festival.

Qianwen App stated that through this substantial investment, it hopes to transform AI assistants from simple chat tools into "AI life agents," and by cultivating the habit of "asking AI for help," it aims to solve the pain point of users repeatedly switching between multiple apps, and achieve a one-stop consumption experience from decision-making to fulfillment.

Related reading: Just after Yuanbao became a wealth management product, Alibaba's Qianwen invested another 3 billion yuan.

Premium nuts are approaching 200 yuan per kilogram, prompting netizens to exclaim, "We can't afford them!"

According to the Luzhong Morning Post, as the Lunar New Year of the Horse approaches, the prices of roasted nuts and seeds have risen overall, with some high-end nuts approaching or even exceeding 200 yuan per kilogram, triggering a wave of complaints from consumers on social media who feel that "it's becoming increasingly unaffordable."

When reporters visited the chain brand Xueji Fried Nuts, they found that the price of basic sunflower seeds and peanuts had reached 23.8 yuan per kilogram, while nuts such as pistachios, freshly roasted cashews, and open-shelled pine nuts were generally close to 100 yuan per kilogram, and hand-peeled pine nuts were as high as 218 yuan per kilogram.

In comparison, the current average price of pork is about 10 yuan per jin (500g), a significant price difference. Due to its high prices, Xueji Fried Nuts has been jokingly referred to by netizens as "Xueji Jewelry Store" or "The Hermes of Fried Nuts."

Besides Xueji, Qiwang peanuts, another brand on the shopping mall's central island, also generally prices its nuts at over 100 yuan per jin (500g). Despite the price increase, many young consumers are still willing to queue up to buy them, showing that high-end roasted nuts remain very attractive during the Spring Festival consumption season.

Industry insiders analyze that the price increase of high-end roasted nuts and seeds is mainly driven by three factors:

• The upgrading of consumption has led to the concentration of brands in shopping malls, and the rent and decoration costs have driven up the selling price.
• Upgrades in raw materials and processes, such as the brand's claim of strictly controlling the number of sunflower seeds;
• Demand surged during the Spring Festival, driving up prices in the short term.

Meanwhile, nut gift boxes continue to hold their position as the "king of New Year's goods" during the Spring Festival. According to data from the International Nut and Dried Fruit Association, Chinese consumers consume approximately 8 million tons of nuts and kernels annually, accounting for 15% of global production. Hema's data shows that since the end of last year, its nut gift box sales have increased by 168% year-on-year.

According to data from iiMedia Research, over 80% of consumers spend between 20 and 100 yuan per month on nuts and roasted seeds, with the highest percentage of consumers purchasing them once or more per month. The Chinese nut and roasted seed market exceeded 300 billion yuan in 2024 and is projected to grow to 428.3 billion yuan by 2029.

A new poster for "Michael Jackson: Road to Stardom" has been released.

A new poster for the Michael Jackson biopic "Michael Jackson: Road to Stardom" has been released. The film will be released in North America on April 24 this year.

The poster's main visual features the costume and appearance of Michael Jackson's nephew, Jaafar Jackson.

He wore a replica red zip-up leather jacket with silver mesh design on the shoulders, matching his look in the "Beat It" music video. Paired with aviator sunglasses and a slightly shy smile, he was praised by many fans as "remarkably similar" to Michael Jackson.

The tagline "His Story Begins" on the poster has sparked further speculation about the film's structure. Besides promising to depict Michael's journey from the Jackson 5 to global superstardom, some fans believe this might indicate a multi-part format to cover a more complete phase of Michael's life.

The film was originally scheduled for release last year but was delayed until this year due to production delays. With the release of the poster, the trailer is expected to be released during the Grammy Awards or next week's Super Bowl.

The re-release of "Return to the Wolf Pack" is scheduled for February 19th.

The film "Return to the Wolf Pack" has announced that it will be re-released in theaters nationwide on February 19 (the third day of the Lunar New Year).

Directed by Yi Feng and produced by Li Weiyi, the film tells the true story of how she rescued the dying wolf cub Green in 2010 and ultimately helped him return to the grasslands and rejoin his wolf pack. Since its release in 2017, the film has moved a large audience with its emotional bond that transcends species and its authentic footage.

Their story also led to the establishment of China's first wolf ecological protection and monitoring station. In 2021, wolves were listed as a Class II protected animal in China.

The film "Return to the Wolf Pack" has a Douban rating of 8.3.

The film "The Swordsman" has officially announced its global simultaneous release during the Spring Festival.

The martial arts action film "The Swordsman" will be released simultaneously at home and abroad on February 17, making it one of the few Chinese martial arts blockbusters to be released globally at the same time this Spring Festival.

Directed by Yuen Woo-ping and produced by Wu Jing and Yu Baimei, the film is adapted from the comic of the same name by Xu Xianzhe. The story revolves around Dao Ma, the "number two fugitive in the world," who takes on a special escort mission. He and his companions escort the target from the Western Regions to Chang'an, but unexpectedly discover that the person they are escorting is actually Zhi Shilang, the "number one fugitive in the world."

Multiple forces subsequently entered the fray, and the conflict surrounding the escort mission escalated continuously.

#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

Musk just announced at Tesla's earnings call that the Model S and Model X will cease production in the second quarter of 2026.

These two flagship models, which once led the global automotive industry's electrification wave, are about to retire more than a decade after their release.

The Model S was first unveiled in 2012, followed by the Model X in 2015.

In an era when gasoline-powered cars still held absolute dominance, these two vehicles shocked the entire automotive industry with their futuristic design and outstanding performance, making people truly believe for the first time that electric vehicles could be the future of luxury cars.

However, times have changed, and the pioneers of yesteryear have now reached their end.

Despite Tesla updating the interior, powertrain, and battery technology of the Model S/X multiple times over the past decade, key aspects such as the vehicle's core styling, dynamic performance, and suspension chassis have remained unchanged.

As the most expensive models in Tesla's current lineup, the Model S and Model X, despite undergoing multiple updates and minor facelifts, still lack industry-standard features such as the 800V architecture and ultra-fast charging. Their once-proud suspension and chassis are also gradually becoming outdated.

As a result, sales of both models declined year by year in major global automotive markets.

In 2025, Tesla sold 50,850 "Other Models," a category that includes the Model S, Model X, and Cybertruck, representing a 40.2% year-over-year decline in overall sales.

Musk stated that the reason for canceling the vehicle project was to make room at Tesla's Fremont factory for the production of its Optimus humanoid robot.

It's time to retire the Model S and X projects in a glorious way, as we truly move towards a future based on autonomy, so if you're interested in buying a Model S or X, now is a great time to place your order.

Tesla's official Chinese website still offers vehicles for immediate sale, with the Model S starting at 842,900 yuan and the Model X starting at 882,900 yuan.

Judging from the financial data, Tesla is in a difficult turning point.

Tesla's total revenue for 2025 is $94.827 billion, a 3% decrease year-over-year, while net profit is approximately $3.794 billion, a 46% decrease year-over-year . In the fourth quarter of last year, revenue was $24.9 billion, a 3.1% decrease year-over-year; operating profit was $1.41 billion, an 11% decrease year-over-year.

This financial performance is not surprising. Tesla's outdated vehicle design and increasingly fierce competition from traditional automakers in the US, Europe, and China have severely hampered demand for Tesla.

Although Tesla launched derivative models such as the "Refreshed Model Y", "Model YL/YP" and "Reduced Standard Edition" this year in an attempt to reverse the decline, sales still decreased by 8.5% year-on-year compared to 2024, with a cumulative sales volume of approximately 1.6 million vehicles.

At the same time, Tesla also lost its title as the world's best-selling electric vehicle company, which was taken over by BYD.

In fact, Musk's interests have long since shifted away from electric vehicle manufacturing itself.

In his view, Tesla's future lies not in hardware, but in FSD (Full Self-Driving), Robotaxi (driverless taxis), and Optimus (humanoid robots).

Musk attributed the company's current poor performance to subsidy cuts and macroeconomic factors, acknowledging that "the company will face a few tough quarters," but he firmly believes that the company will eventually rebound as Tesla's artificial intelligence plans are realized.

Musk optimistically predicts that by the end of 2025, 50% of the U.S. population will be able to use Tesla's robotaxis.

However, as we discussed yesterday, compared to long-term Robotaxi operators like Waymo, Baidu's Robotaxi, and Pony.ai, Tesla's current fleet size is relatively small and operates only in a limited number of US cities. Furthermore, its technological solutions are far from being ready for large-scale operation.

Realizing this vision will likely require a considerable amount of time and numerous technological breakthroughs.

However, it's not all bad news. Tesla's Full Self-Driving (FSD) software saw a 38% year-over-year increase in purchases in 2025, and currently has nearly 1.1 million paid subscribers worldwide.

To further promote the adoption of Full Self-Driving (FSD), Tesla announced that it would stop selling the FSD buy-out version starting February 14 and switch entirely to a monthly subscription model, which will undoubtedly help expand the FSD user base.

In addition to the robots, as is customary, Musk once again made some promises.

The Semi electric truck is scheduled to begin mass production in the first half of 2026, and the next-generation Roadster is planned for release in April 2026. In addition, Tesla plans to launch the Cybertruck, a fully autonomous off-road vehicle capable of playing a role in urban and intercity freight transport.

All of the above indicates that Tesla is gradually transforming from an electric vehicle manufacturer into an artificial intelligence company, which is a key support for its trillion-dollar market value.

At the end of the earnings call, Musk once again quoted former Intel CEO Andy Grove's famous saying: "Only the paranoid survive."

He emphasized that the series of changes Tesla is currently making are "all out of necessity, not because we want to do them, but because we have no other choice."

These words were both a response to external doubts and a defense of his strategic choices.

The discontinuation of Model S and Model X is not only the natural end of the product life cycle, but also symbolizes Tesla's complete departure from its identity as a "pure car manufacturer".

Musk has embarked on another high-stakes gamble, though the outcome may not be known for several years.

But one thing is certain: regardless of success or failure, Tesla is no longer the Tesla that amazed the world with the Model S. Tesla Model S/X production discontinued, capacity transferred to robots; Musk: Only the paranoid survive.

#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