Pogbagram

In March of this year, BYD announced the slogan "5 minutes to fully charge, 9 minutes to fully charge" at its flash charging technology launch event. The first batch of flash charging vehicles, thanks to the second-generation blade battery and new flash charging technology, can charge from 10% to 97% in just 9 minutes.

Surprisingly, this record, which was then known as the world's fastest mass-produced charging speed, only lasted for one month.

On the evening of April 7th, Lynk & Co released the new Lynk & Co 10 and 10+, and broke new records for charging performance.

The Lynk & Co 10, equipped with a 900V high-voltage architecture and a 95-degree shield gold brick battery, has reduced the charging time from 10% to 70% to 4 minutes and 22 seconds, and it only takes 8 minutes and 42 seconds to charge from 10% to 97%.

The speed at which the underlying hardware of pure electric vehicles iterates is indeed quite astonishing.

Let's take a look at the two new cars. The Lynk & Co 10 targets the 200,000 RMB market segment; it's essentially an annual update of the previous Z10. The newly launched Lynk & Co 10+, on the other hand, focuses on the 250,000 RMB market segment, and boasts impressive track performance and chassis hardware.

It outperformed the Porsche Taycan Turbo GT on the track.

1000 horsepower, fully charged in just 8 minutes

The most significant upgrade in this year's Lynk & Co 10 model year update is the three-electric system: the 95 kWh Shield Gold Brick Battery, in conjunction with the V4 Extreme Charging Station, can add an average of 2 kilometers of range per second, bringing the CLTC range to 816 kilometers.

As mentioned earlier, it takes more than 8 minutes to fully charge this battery, and only 4 minutes and 22 seconds to charge it from 10% to 70%, which is about half a minute faster than BYD's flash charging.

In addition to the top-of-the-line 95 kWh battery pack, the Lynk & Co 10 also offers a 77 kWh battery pack. It uses an 800V architecture, achieving a charging rate of 5.5C, and charging from 10% to 80% takes only 10.5 minutes.

The most challenging aspect of high-voltage fast charging is naturally temperature. To manage the high temperatures generated by high-power charging, Lynk & Co has implemented a dual-sided three-dimensional liquid cooling technology for the battery pack.

According to the official statement, compared with the common single bottom surface cooling, this solution doubles the heat dissipation area, halves the heat transfer path, and improves the overall heat exchange efficiency of the battery by 35%.

In terms of safety protection, the battery pack is encased in a protective structure known as "985." This structure includes 9 layers of side impact protection, 8 layers of bottom protection, and 5 layers of top support. Together with an AI BMS battery management system that can automatically learn driving habits, this forms the underlying framework for the vehicle's three-electric safety features.

With the battery as a safety net, the Lynk & Co 10+ boasts impressive power specifications.

The dual-motor all-wheel-drive system delivers a peak power of 680kW (approximately 925 horsepower) and a maximum torque of 913 Nm. The 0-100 km/h acceleration time is 3.2 seconds, and the mid-range acceleration from 80 to 120 km/h takes only 2.1 seconds.

At the launch event, Lynk & Co also announced the car's time on the Asian Ridge Circuit: 1 minute 40.14 seconds.

The lightweight design and fast response speed were a major factor in achieving this result.

Lynk & Co stated that they used aerospace-grade semi-solid magnesium alloy die-castings extensively in the electric drive housing. This single change reduced the housing weight by more than 6 kilograms, bringing the net weight of the entire electric drive system down to 75 kilograms.

The coaxial drive layout eliminates power transmission losses, and the four-wheel drive system has a switching response time of only 10 milliseconds. Combined with the G-TCS intelligent anti-slip control system with a 2-millisecond response, the wheels can execute the driver's intentions very crisply.

According to the official statement, even after more than 100 full-power accelerations and decelerations on the track, this electric drive system can maintain a maximum efficiency of 93.7% without overheating or degradation.

Since it's an annual model update, the new car basically retains the previous exterior design, with changes mainly in the details.

The front grille features a hidden, active opening design, and the outer shell is made of PMMA and ASA high-gloss injection molded material, which can prevent fading from prolonged exposure to sunlight. The door handles have a semi-hidden design, and in addition to the regular micro-switches, emergency mechanical pull cables are retained both inside and outside the vehicle.

The most obvious change is that the Lynk & Co 10+ has been fitted with a set of exclusive aerodynamic and chassis kits.

The large carbon fiber rear wing supports manual adjustment at two levels: 0 degrees and 12 degrees. When on the track, adjusting it to 12 degrees can provide 109.1 kg of downforce at high speeds.

Unsprung weight has also been reduced, with the four 21-inch forged wheels resulting in a total weight reduction of 7.88 kg. The tires are Michelin PS EV, and the Brembo perforated brake discs are paired with racing brake pads with a maximum temperature resistance of 650°C, extending the 100 km/h braking distance to 33.31 meters.

The chassis structure still uses all-aluminum material, with a front double wishbone and rear multi-link independent suspension. The CCD continuously variable damping electronically controlled shock absorption system and the lateral stabilizer bar are standard across the entire series.

Because of its large size, although the Lynk & Co 10 is a car that emphasizes sportiness, its interior space and configuration are very close to those of an executive sedan.

The cabin has an 81% usable floor area ratio, and all four seats in the front and rear rows are equipped with electric adjustment, ventilation, heating and massage functions as standard.

A refrigerator with an independent compressor, capable of both cooling and heating, is installed under the center console. It has a capacity of 5.7 liters and supports temperature regulation from -6°C to 50°C. Even when the person is away from the car, the refrigerator can continue to operate on battery power for up to 24 hours.

The Lynk & Co 10+ boasts an even more extreme interior atmosphere, with a large area of ​​the interior wrapped in Ultrasuede suede, complemented by an exclusive pulse yellow color scheme and racing-inspired perforated patterns, creating a striking visual impact.

The car's hardware computing power is handled by the Snapdragon 8295 chip, which runs the Flyme Auto 2 system.

In terms of driver assistance, Lynk & Co offers two different hardware solutions:

The Qianli Haohan H5 solution, based on the Orin Y chip, is used to handle high-frequency daily traffic scenarios; for more advanced all-scenario assisted driving, the Qianli Haohan H7 solution equipped with the Thor-U chip can be selected.

#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.

The upcoming OPPO Find X9 Ultra will replace all seven lenses – the main camera, wide-angle lens, two telephoto lenses, a Danxia color lens, the front camera, and a teleconverter…

None of them were heirlooms.

In its more than 20 years of operation, OPPO has very rarely completely overhauled its imaging hardware for a single product generation. Luo Jun said:

This kind of revolutionary upgrade seems to have never been done before.

Luo Jun is the director of OPPO's imaging algorithm, who has led the imaging algorithm direction of four generations of products from Find X6 Pro to Find X9 Ultra, as well as the "Master Mode" that has become a hallmark of OPPO's imaging.

On the eve of the Find X9 Ultra launch, iFanr interviewed Luo Jun. We talked about the newly designed "Master Mode," the unreleased phone, and the past, present, and future of OPPO's imaging technology.

We are trying to find an answer: What is the "realism" of computational photography?

Master Mode: Using Algorithms to Fight Algorithms

What is "Master Mode"?

Simply put, it's an image pipeline in OPPO cameras that's independent of the regular shooting mode.

Normal mode aims for brightness, pleasing colors, and instant good looks—the system automatically increases brightness, enhances color saturation, and performs strong sharpening and noise reduction. These operations make the photos eye-catching at first glance, but at the cost of making them look somewhat artificially "refined."

This is the most common controversy in the mobile imaging industry today: "algorithm-driven".

The "algorithmic flavor" is not a problem of any one company, but rather a structural byproduct of the development of computational photography to its current state.

As mobile phones use increasingly sophisticated algorithms to compensate for the physical limitations of their small bodies—multi-frame synthesis, AI noise reduction, HDR stacking, super-resolution reconstruction—each processing step adds computational traces to the photo: shadows are brightened, highlights are suppressed, noise is smoothed out to create an oil painting-like texture, and sharpening creates more "details."

The result is that everything looks good, but nothing seems to have any purity.

"Master Mode" takes a different approach.

It processes images with more restrained tone mapping, a more natural sharpening strategy, and a tonal logic closer to that of an optical camera. It preserves shadows where they should be dark and retains grain where there is noise, not pursuing "bright white beauty" in every picture, but instead pursuing the realistic texture of the photo.

There's an unwritten rule in the mobile phone industry: if a feature is controversial for two consecutive generations, it will most likely be scrapped in the third generation.

"Master Mode" perfectly embodied this image.

During the Find X7 Ultra era, user opinions on it were polarized. Those who liked it said it had a "camera-like" feel and a "high-end" look; those who disliked it said the image was "dark" and "not sharp." I even encountered a situation where, after sending a photo taken in Master Mode to a friend, their first reaction was: "Did you take this picture blurry?"

But Master Mode survived.

It has transformed from a niche tool for professional users into a photo mode that even college students actively choose.

What's even more intriguing is that users prefer the Master Mode for almost the same reason— it lacks an algorithmic feel.

In fact, the Master Mode runs the most advanced and computationally intensive algorithm pipeline in the entire imaging system.

This sense of contrast is a microcosm of OPPO's current imaging capabilities, and also reflects Luo Jun's complete understanding of computational photography over the past decade.

The masters of traditional filmmaking created the master mode.

Luo Jun majored in image algorithms and joined Sony through campus recruitment.

In the early 2000s, the Japanese imaging industry was the pinnacle of the global imaging industry. He worked on Handycam video recorders, Alpha SLR cameras, and witnessed the development of the NEX mirrorless series from scratch.

But what truly made him see the turning point in the industry was the Sony RX100.

At the time, it cost over 200,000 yen, while a typical point-and-shoot camera cost 50,000 to 60,000 yen. This one sold for over 10,000 yuan as soon as it was released, but it was indeed quite innovative.

Sony crammed a one-inch sensor and a Zeiss lens into a body the size of a shirt pocket. This marked the beginning of the miniaturization trend in imaging. Looking back today, the RX100 and later mobile phone imaging followed the same path: maximizing image quality within extremely limited physical space.

However, mobile phones have gone much further.

During his more than ten years at Sony, all of Luo Jun's image algorithms ran on dedicated ASIC chips. A single chip was developed every two years, covering multiple product lines, prioritizing stability and reliability.

But he gradually realized a fundamental misalignment:

Algorithms iterate rapidly, but ASICs are released every two years. The overall computing power and architecture are somewhat mismatched with the research approach of computational photography and AI—it's too slow.

Later, he discovered the NPU—a processing unit specifically designed for handling neural network computations. Algorithms could run at the software layer, dramatically increasing iteration speed.

However, the best platform for an NPU is not in a camera, but in a mobile phone.

In early 2017, Luo Jun saw OPPO demonstrate its periscope telephoto technology at MWC—10x hybrid zoom, something no one in the mobile phone industry had done at the time. He immediately recognized the company's potential and decided to join OPPO.

Interestingly, ten years later, the Find X9 Ultra, which he spearheaded, features an even better 10x optical telephoto lens, but that's another story.

This shift from traditional to mobile imaging determined the underlying logic behind his Master Mode. Many people feel that Master Mode "lacks algorithmic feedback, resembling straight-out-of-camera output," a comment that Luo Jun found interesting.

Professional cameras also have algorithms, and their ISP pipelines are quite complex, entirely implemented using chips. However, the effect is very similar to our Master Mode, so the user's frame of reference becomes—"I can achieve a camera-like effect using my phone's algorithms."

In his view, the idea that "cameras don't have algorithms" is a misconception. The camera's algorithms are simply embedded in the chip, invisible to the user.

The design of Master Mode stems from this understanding. The goal has never been to "remove the algorithm," but rather to make the algorithm like the ISP of a professional camera—doing a lot of work without you even realizing it.

OPPO internally calls it "using computation to compute" .

Luo Jun said that if your goal is to "make the algorithm invisible," then you can't pursue improvements in a single parameter. You need a systematic set of standards to define what "good" means.

He summarized this standard in four words: true to life.

Three years to reshape OPPO Imaging

At the end of 2021, Luo Jun was transferred back to China from Japan to take full control of the iteration direction of OPPO's imaging algorithms.

For all mobile phone manufacturers, shifting to self-developed imaging algorithms is a decision that involves high long-term investment but low short-term returns.

However, in order to make imaging—not just beautification—a core competitive advantage of OPPO's flagship phones, Luo Jun reorganized a team of hundreds of people working on imaging algorithms.

"Realistic reproduction" is a relatively abstract concept: what kind of images can be considered realistic, and what methods should be used to reproduce them?

Luo Jun breaks it down into three specific dimensions— light and shadow, detail, and color. He has a three-year plan in mind, hoping to reconstruct OPPO's imaging capabilities with three generations of products.

The Find X6 Pro was a turning point for imaging phones under Luo Jun's philosophy, as it primarily addressed the issue of light and shadow.

In an interview with iFanr, OPPO's Director of Imaging Cognition, Cheng Zhuo, stated that the goal of the Find X6 series is to establish "correct tonal relationships"—correcting distorted light and shadow curves.

This generation of Find features the industry's only large-sensor telephoto lens at the time—a 1/1.56-inch CMOS sensor with an equivalent 70mm lens, and is paired with a brand-new Super Light and Shadow Image Engine.

This system, for the first time, enables mobile phones to calibrate brightness information at the pixel level and calculate the light and shadow relationships between the subject, light, and environment. Luo Jun said:

Bright but not dazzling, dark but not black – these are our basic requirements for light and shadow.

Luo Jun also introduced a mode for professional users that could fully utilize the imaging capabilities of mobile phones into OPPO's imaging system, which they named "Hasselblad Professional Mode"—this was the prototype of "Master Mode".

Next, Luo Jun's team needed to address the details.

The Find X7 Ultra features the industry's first dual periscope quad main camera system, adding a telephoto lens that supports 6x optical zoom.

The increase in focal length is not just about "shooting farther." In Luo Jun's understanding, it has a more fundamental meaning:

With more focal lengths, there are more frames of reference. You can record the world from different perspectives, and the system can reconstruct more complete information.

Frame of reference—this is the core concept that Luo Jun uses to understand "realistic reproduction".

Reality is not an absolute objective standard; it depends on what you use as a reference. The viewfinder is one frame of reference, what the human eye sees is another, and the "good photo" imagined by the user is yet another.

The more focal lengths and details a system captures, the more complete the reference information it obtains, and the closer it gets to the "truth" in the user's mind.

The Find X7 Ultra further enhances the quality of light and shadow, especially in the mid-tones.

In everyday photos, the most significant amount of light and shadow information is concentrated in the midtone area—the transition zone between the brightest and darkest parts. If the midtones are coarse, the photo lacks realism.

It was also in this generation of imaging systems that OPPO officially launched "Master Mode". In Luo Jun's view, Master Mode is not exclusive to photographers, but rather returns the power to adjust the camera to the user – just like the levers and knobs on a camera.

However, the first-generation Master Mode had limited generalization capabilities and insufficient scenario compatibility, resulting in mixed user reviews. Some people loved it, while many others couldn't figure it out.

For Luo Jun, technical problems can always be solved, but how to uphold and communicate his ideas is a huge challenge.

The reason why Master Mode has been able to persist is perhaps because we have not compromised.

The lighting and details are there, but color is the last shortcoming.

Computational photography relies heavily on statistics. In complex lighting conditions, inaccurate white balance, skin tone shifts, and environmental color distortion are inherent limitations of statistical methods.

The Find X8 Ultra features a new lens—the Danxia Original Color Lens—dedicated to local color temperature sensing. It can identify the color temperature distribution in different areas of the image, distinguish between natural and artificial light sources, and independently reproduce skin tones and ambient colors.

Color mapping essentially involves two things: white balance and color mapping. White balance is a statistical method, and it's inherently inaccurate in some scenarios. With Danxia landforms, because they contain absolute information, there's a chance to correct deviations in scenes with interfering colors.

The role of Danxia is not to make colors look better, but to provide a physical anchor point for the color calculation pipeline—an objective reference benchmark that does not rely on statistical guessing.

See, it's another frame of reference.

With the Find X8 Ultra, another easily overlooked technological integration was completed: the processing algorithms for Master Mode and Photo Mode in the RAW domain were unified.

The RAW images produced by both modes are the same; the difference lies only in the backend—Photo mode uses a brighter and more pleasing tone mapping, while Master mode uses a more restrained approach to lighting and sharpening.

This means that "Master Mode" is no longer an independent functional branch; its underlying capabilities have become the core of the entire imaging system.

In Luo Jun's view, with the Find X8 Ultra generation, his original vision has finally been realized – light, shadow, detail and color, the three dimensions are combined into a complete form for the first time.

Thus, the new OPPO imaging brand "LUMO" was born.

Luo Jun's team's criteria for judging good images have gradually taken shape after three generations of product iterations— one of the benchmarks being the "continuity" of the photographs .

Photos taken with professional cameras also have noise, but the noise and graininess are continuous and look pleasing. I'd rather have some continuous noise than have patches of sharpness and blurriness in the image.

These standards didn't suddenly emerge during the development of a particular product generation; they originated from the traditional imaging genes ingrained in Luo Jun's bones—signal-to-noise ratio, continuity, and color mapping—only in a different medium, from cameras to mobile phones, from traditional optics to computational photography.

As new image processing algorithms gradually take shape, Luo Jun faces a new situation: the software side has done almost everything it can. The marginal benefits of algorithm iteration are diminishing.

What's next?

Find X9 Ultra: Echoes of a Decade

The answer is to do it again.

Luo Jun divides the development of mobile phone imaging into three stages:

The first phase began around 2015, with the core being device miniaturization—packing large sensors into mobile phones, stacking them from 1/3 inch all the way up to one inch;

The second phase began around 2021, when the algorithmic capabilities of AI and computational photography improved, allowing for the creation of decent-quality photos even without a large 1-inch sensor through algorithmic enhancement.

The third stage is now:

You can't rely on components or algorithms alone. It requires a combination of hardware and software, end-to-end innovation, to have any chance of pushing the effects forward.

The Find X9 Ultra is the product of this third stage—for which OPPO's imaging team went so far as to replace all seven lenses.

The main camera has been upgraded from a 50-megapixel 1-inch sensor to a 200-megapixel 1/1.2-inch sensor, the wide-angle lens has been upgraded from a 1/2.5-inch sensor to a 1/1.95-inch sensor, the first telephoto lens has been replaced with a larger sensor, and the second telephoto lens has been expanded from 6x optical zoom to 10x optical zoom. The color reproduction lens has been upgraded, the front camera has been upgraded from 32 megapixels to 50 megapixels, and even the teleconverter has been upgraded from 200 to 300.

The most challenging part to design was undoubtedly the 10x optical zoom telephoto lens.

Luo Jun showed iFanr the Find X9 Ultra's 10x telephoto lens—a 1/2.8-inch sensor paired with a 230mm lens group, but the entire module is only 29mm long, with the length of the prism being about half a little finger.

What's even more ingenious is that this prism isn't a single piece; it's made of three prisms joined together, with an air layer sealed in the middle to eliminate stray light. This process is unprecedented in the industry chain—no one has ever cut a prism into three pieces and glued them together, no one has ever sealed an air layer in the middle of a prism, and of course, no one has ever built such a production line.

Therefore, everything had to be started from scratch.

Luo Jun positioned this 10x telephoto lens as a "pocket teleconverter"—the teleconverter on the OPPO Find X9 Pro is more than ten centimeters long, while the "built-in teleconverter" on the X9 Ultra is only 29 millimeters long, but the image quality is the same.

That's why you can find all the mainstream focal lengths from 14mm to 230mm in the OPPO Find X9 Ultra, which is the classic "holy trinity" configuration of cameras.

In 2016, Luo Jun was impressed by OPPO's 10x periscope telephoto technology demonstration at MWC and decided to join the company. Ten years later, he and his team have embedded the best 10x optical telephoto lens to date into a mobile phone—for Luo Jun, this is an echo spanning a decade.

With the addition of 10x telephoto, the creative possibilities in Master Mode have expanded dramatically: videos can be shot at 10x or 20x zoom, and portrait mode has also added a 10x zoom range—something Luo Jun hadn't anticipated three years ago.

I probably never considered shooting these things with 10x zoom before, but suddenly I found that the material space has become much larger, which is quite interesting.

The new generation of Master Mode is also easier to use and share.

Luo Jun said that his favorite feature is the "recipe sharing" function. Users can adjust the shooting parameters and take photos in master mode, and the recipe will be embedded in the photo watermark.

When others see this photo, they can quickly import the same recipe and create a new one using ColorOS's one-click note function—it's so convenient for Xiaohongshu users who love to share their photos.

The premise for this feature to work is precisely that the previous three generations made the underlying pipeline of Master Mode sufficiently stable. If the pipeline is not mature, the recipe will fail in a different scenario after being shared.

Good computational photography is when you forget about computational photography.

Towards the end of the interview, we touched on a slightly abstract question: What is the "reality" of computational photography?

Luo Jun's answer consisted of only two sentences:

One is called "what you see is what you get," and the other is called "what you get is what you think."

What you see is what you get—that's what you get in the frame. But he believes the real key is the second half: users have expectations of what constitutes a good photo, and the job of the imaging system is to get as close to those expectations as possible.

When you take a photo, you're visualizing the effect of that photo. Whether it's what you see or what you imagine, that's the brain working on the post-processing.

Before you press the shutter, you already have an image in your mind. That image is your frame of reference.

This reminds me of my experience when I traveled to Sydney. I went there specifically to a famous photo spot, but it was a rainy day and there were a lot of people. After taking the photos, I wasn't very satisfied with them.

So, I thought of using Doubao to edit the photo—I added a sunset, removed the shadows, and after editing, I felt that this was what I wanted, but is this still considered photography?

Luo Jun told me:

It's definitely photography. But what percentage of what you envision in your mind is actually recorded, and what percentage is generated? This percentage varies depending on the tools and the context. The value of our imaging systems lies in maximizing the portion that is truly recorded. Otherwise, we could just rely on simple cameras.

From Master Mode to the reconstruction of OPPO Imaging, and then to the Find X9 Ultra—in Luo Jun's view, all of this has always pointed to the same goal:

Minimize the distance between the photo in your mind and the photo taken by your phone.

A true recreation is not only a recreation of reality itself, but also a recreation of the frame of reference in our minds.

Luo Jun said that in the future, image interaction must be simple for users—users can just pick it up and shoot without having to think about it, because the system already understands what you want.

I think that by then, the concept of realistic reproduction had already permeated the entire OPPO Crystal Imaging System.

Good computational photography is about making you forget about computational photography.

#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.

The competitive landscape that Starbucks China is facing now is completely different from that of previous years, as well as from its home country in the United States and other places.

During the sharing session at the 2026 Starbucks China Partner Forum, the fifth store manager of the first Starbucks store in China, the Beijing China World Trade Center store, said that when he was in charge, there were only two coffee shops in the China World Trade Center area; while the current 18th store manager said that there are already forty or fifty coffee shops in this area.

For a long time, the Chinese understanding of coffee was limited to Nestlé and Starbucks. But now, the meaning of coffee has been redefined. In Shenzhen, working-class people might order a 9.9 RMB Luckin Coffee iced espresso to perk them up in the morning; in Guangzhou, coffee lovers might order a JPG Latte and a roasted chicken and mushroom beef burger for lunch; and in Shanghai, sophisticated white-collar workers might meet clients at a boutique pour-over shop in Wutong District for a medium-light roast Geisha pour-over priced in the hundreds…

People have more choices besides Starbucks, and Starbucks also needs to make some changes to continue to be the primary choice.

What many people don't realize is that while the options for drinking coffee have expanded several times over in the past few years, Starbucks remains the best choice when it comes to sitting down for a cup of coffee. This is because even in the face of fiercer competition, Starbucks has not lowered the quality of its stores. Starbucks remains the top performer in terms of "good store locations, a large number of stores, large store areas, and plenty of seating."

The 2026 Starbucks China Partner Forum was a response to the competition in the Chinese coffee market. At this forum, Starbucks China launched its "1000 Stores, 1000 Faces" strategy. The official statement reads, "Starbucks China will fully leverage its strengths, enhance local innovation, and drive a leap forward in the Starbucks experience through the '1000 Stores, 1000 Faces' strategy. This will empower partners to create a unique community for each store, serving thousands of communities with differentiated products, experiences, and scenarios, thus paving a unique growth path for Starbucks and deepening its high-quality development in China."

Breaking it down, the strategy can be divided into 5 parts:

• Professional coffee first choice
• High-quality product innovation
• Scenario-based store expansion
• Partners create "one store, one community"
• AI helps connect people with humanity

In recent years, Starbucks' product capabilities have faced numerous challenges, such as the over-industrialization of dark roast beans and the slower pace of product innovation compared to domestic brands. Of course, Starbucks has also continuously reformed its products to meet these challenges, such as launching Reserve stores and Reserve beans, introducing medium and light roast beans as an option, and accelerating the pace of product innovation.

This time, "professional coffee selection" and "high-quality product innovation" are a very positive and clear response to prove that you can enjoy high-quality coffee in Starbucks stores, rather than just supplementing your caffeine intake.

Specifically, "professional coffee selection" has three meanings.

First, there's the comprehensive flavor selection. Alongside the 2026 Starbucks China Partner Forum and the "One Thousand Stores, One Thousand Faces" strategy, Starbucks launched "Springtime and Bright Scenery," a light roast coffee bean specially tailored for the Chinese market. This is currently the lightest roast on Starbucks' roasting curve. Combined with classic dark roasts and the "One Bean a Month" small-batch single-origin beans available at over 800 Starbucks Reserve stores in China, Starbucks now offers a very rich selection of coffee beans.

Then there's the wide variety of brewing methods. In addition to the existing 10 or more coffee brewing methods such as espresso, pour-over, French press, siphon, cold brew, and nitro cold brew, Starbucks will invest in new brewing equipment in more than 1,000 stores in office building business districts to launch "Daily Fresh Brew" coffee, which combines daily black coffee with milk coffee to meet the quality coffee needs of daily commuters.

Finally, regarding the refinement of barista promotion levels, in addition to the existing green apron (regular barista), black apron (barista), and brown apron (barista ambassador), Starbucks has officially launched a new coffee career path: "Regional Barista." In the future, each operations manager region will be assigned a dedicated Regional Barista to be responsible for the development of regional coffee culture. The "custom recipes" created by barista partners are also expected to be promoted regionally and even nationally.

"High-quality product innovation" is Starbucks' thinking at the product level. For example, leading the health trend, in response to the fact that many of Starbucks' past seasonal limited coffees were too sweet, and the current situation in China where coffee products are moving towards the direction of coffee milk tea, Starbucks still puts health in a high position. For example, the new "High Protein Latte PRO" uses milk with 6.0 grams of protein per 100 ml. One cup of "High Protein Latte PRO" series drinks contains 20 grams of native milk protein and 0 lactose.

In addition, Starbucks China stated that it will integrate global and Chinese culinary trends and inspirations. This summer, customers will see more refreshing and thirst-quenching fruity coffees and iced shaken teas, as well as a newly upgraded Frappuccino. In the future, it will continue to break down the combinations of coffee beans, brewing methods, flavors, sugar content, milk, and temperature to meet customers' highly customized needs.

To a large extent, Starbucks' strategy is somewhat against the trend.

The 9.9 yuan price barrier that Luckin Coffee built up has now collapsed. Products like 2.9 yuan Americanos and 3.9 yuan lattes have appeared on the market that are completely unprofitable. At the same time, the industry trend is to add sugar, flavorings and fats to make the flavor more appealing and easier to drink.

Starbucks' message is that it won't make coffee priced at 9.9 yuan or even 2.9 yuan, nor will it allow coffee to become like milk tea. Whether it's launching the "True Sugar-Free" series against the trend, the "Rose 20" series using high-quality ingredients, or the recent "High Protein Latte PRO" series, the idea is to maintain product quality and price by using better raw materials.

The "scenario-based store expansion" strategy is a traditional strength of Starbucks. Store expansion remains an important growth engine for Starbucks China, and Starbucks will adhere to a high-quality store expansion strategy, moving forward steadily in a more focused and stable manner to ensure the healthy growth of both new and existing stores. This also has three layers of meaning.

First, Starbucks is accelerating its expansion into untapped business districts and markets: Currently, Starbucks stores cover more than 1,000 county-level administrative regions, and this number will increase to 1,500 or more within the next three years. Simultaneously, it continues to expand into new business districts and meet existing demands in existing cities; even in Shanghai, where there are already more than 1,100 stores, there are still new business districts to explore.

Starbucks stores are distinctive enough that many could be considered tourist attractions, but Starbucks believes that's not enough. Their strategy includes creating localized experiences for each store, such as providing reliable and healthy meals for hospital stores, launching unique souvenirs for cultural and tourism stores, and opening more intangible cultural heritage experience stores in more cities that pay tribute to local culture.

In addition, Starbucks will expand into more store formats. Besides regular stores, Reserve stores, and themed stores (such as intangible cultural heritage themed stores), Starbucks will also expand into smaller stores such as the smallest Starbucks in scenic areas (10 square meters), coffee carts at concert venues, and modular office building convenience stores, etc., to meet the needs of different customer groups at different times.

In conjunction with "scenario-based store expansion," other complementary strategies include "one store, one community" and "AI-assisted humanistic connections." In Starbucks stores, partners and AI can become nodes for community connections and growth engines, making stores places for activities such as crafts, pets, cycling, and running, increasing the emotional connection between Starbucks, the community, and consumers.

Starbucks China in 2026 is unlike any other. Firstly, it has demonstrated remarkable resilience by not succumbing to the fierce competition in the coffee brand arena. In similar past competitions, such as in the automotive, smartphone, and television industries, most overseas brands have suffered setbacks at the hands of domestic brands. A more significant difference is that on April 2, 2026, Starbucks Coffee Company announced the formal completion of its strategic partnership with Boyu Capital.

The fund managed by Boyu Capital currently holds a 60% stake in Starbucks China's retail business, while Starbucks Global retains a 40% stake and will continue to act as the owner and licensor of the Starbucks brand and intellectual property, licensing it to the newly established joint venture. Approximately 8,000 Starbucks company-operated stores in mainland China will be converted to a franchise model and operated and managed by the joint venture.

Starbucks and Boyu Capital share a long-term vision of gradually expanding the number of stores in China to 20,000.

Before considering the future, we can also look at the past. Regardless of changes in the competitive landscape or equity structure, Starbucks' core competitiveness has remained unchanged at the store level.

The Starbucks Shanghai Roastery, which has been open for nine years, is now bustling with visitors early in the morning. While the average order value isn't low, as the only one of its kind in China and one of only two in Asia, it attracts coffee lovers not only from all over the country but also from around the world – that's the significance of the store. People don't always have to receive coffee from a delivery person; often, they want to sit in front of a barista and watch them use pour-over kettles, siphon pots, and other equipment to transform aroma and flavor, or watch green beans being transported by conveyor belt to the roaster, turning from green to brown, from burlap sacks to roasting pans.

Starbucks hopes that there will be more than just this one store; there will be every single one of them.

#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.

According to data from the National Bureau of Statistics, the monthly labor cost of a software engineer in China is roughly estimated to be around 20,000 to 30,000 yuan.

If we only consider the portion of his 8-hour workday when he actually performs his duties, it comes to approximately 110 to 170 yuan per hour.

Anthropic launched a new feature today called Claude Managed Agents, with one price listed at $0.08/hour, which is less than 0.6 yuan.

The number itself isn't the key point; the key point is that it means Anthropic has started billing by the hour. It's not just charging for the tokens used, but also calculating how long the agent has been running.

Managed Agents provides a complete set of off-the-shelf infrastructure, or what Anthropic calls an agent harness: including tool calls, memory systems, access control, long-term cloud operation, mutual monitoring between agents, and sandbox environments .

For example, suppose we want to hire someone to help you with your work, what problems might we encounter?

During the recruitment phase, it is necessary to prepare office space (server), install computers and configure operating systems (development environment), and write job descriptions (code logic).

During the work phase: The network connection drops halfway through, all progress is lost (session interruption), you want to check what he did but there is no record (unable to audit), and you are worried that he may be accessing company secrets (access control).

Claude Managed Agents' role in this process was to handle all these troublesome matters. Anthropic stated, "Stop building that shoddy makeshift setup yourself. Leave the infrastructure to me, and you can just focus on making money."

By setting up a console in Claude's official Agent or using the API, we can directly issue Agent requests. Claude Managed Agents will then provide them with a workspace, supervise their work, and ensure they don't misbehave .

Currently, Claude Managed Agent is in public beta, allowing anyone or any enterprise to quickly build a real, functional digital employee.

You can build an agent from scratch in just a few days.

Over the past two years, we've used countless agents, with developers launching their own agent products almost daily. Some are geared towards programming code, others towards design. Ultimately, these agents have all been unified into a large family: last year it was the Manus class, and this year it's the OpenClaw class.

However, if you want to deploy a more personalized agent, especially one that can be used by others, you need to handle the corresponding server yourself, set up complex mechanisms to prevent it from crashing, grant it secure access to the database, and manage the agent's context memory in a reasonable way.

Managed Agents handles all of this.

Its structure revolves around four concepts. The Agent defines who this employee is: what model they use, what system prompts they follow, and what tools they can invoke. The Environment is a pre-configured cloud container that comes pre-installed with runtime environments such as Python and Node.js.

A Session is a specific instance of task execution with a complete event history that can be viewed at any time. Events are the messages exchanged between us and the agent—task instructions, tool results, and status updates.

The complex "hand-crafted" agent model of the past has been compressed into a fully automated pipeline by Claude Managed Agents.

If you are a developer, you can directly call the API or use the CLI. A few lines of code are all it takes to create the agent, configure the runtime environment, start the session, and receive real-time event streams. The entire process is clearly documented, and it takes about half an hour to get it running from scratch.

If you don't write code, the Claude Console provides a complete visual interface. Selecting a model, writing system prompts, connecting to the MCP tool, and attaching external services are all completed with a few clicks. After configuration, you can test directly in the interface to see how the agent responds. If you're not satisfied, adjust it; if you are, let it run continuously.

The console's build page features an input box asking "What do you want to build?", next to which is a template library covering ready-made roles such as researcher, data analyst, customer service assistant, and incident response coordinator. Each role is pre-connected to tools like Slack, Notion, Asana, GitHub, and Jira. Simply select a template, modify the description, and you're ready to go.

However, simply becoming a Claude member is not enough. Currently, you also need to have an API plan, which means linking a credit card with a certain amount of tokens, in order to use the Managed Agent.

Managed Agents has a core engineering decision, related to the Harness project which has been discussed recently, that determines whether the system can truly be used in production.

Anthropic used a particularly poignant analogy in its official engineering blog to explain the structural design of Managed Agents.

They felt that early agent architectures were very much like "keeping a pet." Developers were used to stuffing the Claude (brain), the sandbox for executing code (hands and feet), and its memory (session logs) all into a huge server container.

This container has become incredibly delicate; we can't let it die. If the container freezes or crashes, the AI's brain and limbs will be destroyed, and the user's task data will be instantly wiped clean. The container is running both user credentials and Claude-generated code simultaneously; if there is a prompt word injection attack, the credentials will be directly exposed.

Anthropic's solution is to completely separate the "brain" and "hands," turning the container into "cattle and horses" that can be sacrificed at any time, that is, changing from raising pets to raising cattle and horses.

The scheduler (brain) no longer resides inside the container. It commands the container as if calling an external tool. What if a container crashes while executing dangerous code? The brain doesn't panic at all; it logs the error code and then unhesitatingly starts a new container to continue working.

The memories left by the agent are no longer crammed into the crowded brain of some AI or container. After operating separately, all memories are stored individually in an external session log. It's like an external hard drive.

The brain commands the hands through standardized invocation methods, regardless of whether the hands are containers, external services, or anything else. If one hand malfunctions, it switches to another, and the brain continues working; if the brain crashes, it recovers from the dialogue log and gets back to work.

This design resulted in a significant performance improvement. Before decoupling, each conversation had to wait for the container to fully initialize, and the system would spend a long time starting up a heavy container containing a large amount of scheduling logic.

Now, the first response time has been reduced by more than 90%, and the security boundary has become clearer as a result—the code generated by Claude runs in a sandbox, the credentials are in a safe outside the sandbox, and there are dedicated agents to isolate the two, so the agents can never get the original credentials.

More importantly, it enables the Agent to truly have the ability to work stably over the long term.

Anthropic mentioned that Notion has already built an enterprise agent internally using Managed Agents to help engineers write code and knowledge workers give presentations.

Rakuten deployed agents for sales, marketing, finance, and HR using Managed Agents, with each specialized agent going live within a week.

After Sentry's debugging agent discovers a bug, it automatically writes a patch and opens a pull request (PR). Developers receive a fix that they can review directly, and the entire process requires no human intervention.

In essence, large modeling companies used to provide model APIs, which processed each of our messages; Anthropic's change was to wrap the message-based API into an Agent API that could directly deliver jobs.

Back to that number $0.08/session-hour

This change is first reflected in the pricing structure of Claude Managed Agents. According to the official blog, the billing for Managed Agents includes a token fee (standard API price, $3/M input and $15/M output for Sonnet 4.6), plus $0.08/session-hour (billed based on actual runtime, excluding idle time), and a separate web search fee of $10 per 1000 times.

Anthropic provides an example: a one-hour coding session using Opus 4.6 with 50K inputs and 15K output tokens costs approximately $0.70.

Instead of hiring a dedicated employee, companies can now create their own internal agents through Managed Agents. The concept of digital employees has been taken a step further.

Furthermore, for Anthropic, this also means that revenue is now directly linked to the level of automation a company has; the more agents a company runs, the more Anthropic earns. This is the same logic behind AWS shifting from "selling servers" to "selling runtime," opening up a market much larger than simply selling subscriptions.

With the development of large model technology, the golden age of simply comparing parameters and benchmark scores seems to be fading. After all, truly powerful large models are restricted from being made publicly available.

The real battleground has returned to "how to make these brilliant minds work on the factory assembly line in the most stable and cheapest way." The launch of Claude Managed Agents is a milestone in the maturation of AI infrastructure.

Looking back at Claude's updates this year, whether in models or products, they have almost all addressed our pain points about what AI can do.

On the one hand, we are continuously improving the capabilities of our models to avoid being distracted by external factors such as video, browser, and image models; on the other hand, from Cowork to the subsequent frantic patching and replication of all the features of OpenClaw, and now to the launch of a platform specifically for developing and deploying Agents, each step has demonstrated an extremely keen product perspective.

Anthropic is pioneering a new release model, shifting from "We're releasing a faster, better tool" to "We've got you covered with the complete infrastructure for building your digital workforce."

Reference link:
Claude Managed Agents blog update:
https://claude.com/blog/claude-managed-agents
Claude Managed Agents Architecture Design Blog:
https://www.anthropic.com/engineering/managed-agents
Start setting up your own Agents in the Claude console:
https://platform.claude.com/workspaces/default/agent-quickstart

#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.

For a considerable period, the iPhone was a very streamlined product line, with a flagship product released every September. It was clear, restrained, and relied on high profits for profitability. On the other hand, domestic mobile phone brands like Xiaomi used a variety of product combinations to compete with the iPhone and also to compete in markets that Apple did not cover.

But starting next year, the number of new iPhones will be as many as that of Xiaomi's main brand, covering various price ranges from 3,000 yuan to 15,000 yuan.

Finally, Apple was no longer satisfied with fighting a battle only once a year. They wanted to turn the iPhone into a new business that operated all year round, stimulated consumption all year round, and could be repriced all year round.

7 iPhones a year, covering all price ranges.

According to Bloomberg, Apple is expected to launch six new iPhones between the second half of this year and the first half of next year, with announcements scheduled for the fall and spring of the following year.

• Fall 2026 Launch Event: iPhone 18 Pro, iPhone 18 Pro Max, Foldable iPhone Ultra
• 2027 Spring Launch Event: iPhone 18e, iPhone Air 2, iPhone 18.

Last year's iPhone 17 Pro series underwent a complete redesign inside and out, so this year's iPhone 18 Pro series will not have any changes in appearance. The main upgrades are as follows:

• The area of ​​Lingdong Island will shrink by 35%, and it may become a hole-punched island in the upper left corner.
• 2nm process A20 Pro chip
• The image features a variable aperture and a larger telephoto aperture.
• The camera control buttons have been simplified, completely removing capacitive touch functionality while retaining only basic pressure sensitivity.
• New "Wine Red" Color Scheme
• Battery capacity is expected to be increased to 5100mAh.

However, we all know that the real star of this year will be Apple's first foldable screen iPhone Fold, which may actually be named "iPhone Ultra".

This week, Nikkei reported that the foldable iPhone has entered trial production but has encountered setbacks, potentially delaying shipments. Bloomberg, however, offers a different perspective: while the complexity of the new display and the phone itself may lead to supply shortages in the weeks following its release, Apple still plans to launch it after the iPhone 18 Pro series.

Insiders revealed that Apple engineers have solved two major pain points: foldable screen quality and phone durability, and the foldable iPhone is expected to achieve a truly crease-free design. To keep the unfolded thickness within 5 millimeters, Apple abandoned the Face ID module and telephoto lens, adopting a Touch ID solution via the power button.

Unlike other foldable phones on the market, the foldable iPhone has a 5.5-inch external screen with a near-square aspect ratio, while when unfolded, it displays a 7.8-inch screen in landscape mode, more like a landscape iPad Mini. Apple plans to adjust iOS 27 to make foldable iPhone apps more similar to iPad apps on the new device.

It is certain that the foldable iPhone will not be cheap, with a starting price that may exceed $2,000, which is approximately 13,000 yuan.

The iPhone 18, originally slated for release this fall, will be delayed until next spring, sitting alongside the iPhone 18e.

The message conveyed by this change is quite clear: it aims to further widen the gap with the iPhone Pro in terms of positioning. According to leaks, the update to the iPhone 18 will be very minor, with virtually no changes to the exterior design. The updates will focus on internal configurations such as the A20 chip and 12GB of RAM.

Even more interestingly, the iPhone Air, which was thought to have been canceled, will also be relaunched next year. The Information reports that Apple is already redesigning the iPhone Air, with a key update being the addition of a second rear camera, likely a 48MP ultra-wide-angle camera. In addition to these upgrades, the phone will also be lighter, have a larger battery capacity, and may even feature the same heat sink as the iPhone 17 Pro.

Because the current iPhone Air already utilizes its internal space to its fullest potential, it's difficult to accommodate a second camera. Apple will almost have to redesign the internal structure of the iPhone Air 2 from scratch. Apple will also use a thinner Face ID module to further save space.

In short, starting with the iPhone 18 series, Apple will release six iPhones: iPhone Ultra, iPhone 18 Pro Max, iPhone 18 Pro, iPhone Air 2, iPhone 18, and iPhone 18e.

But that's not all. 2027 will mark the 20th anniversary of the iPhone's release, and Apple plans to launch a commemorative iPhone with an unprecedented curved glass design.

This device, codenamed V72, will be an "all-glass" iPhone—the screen and the glass on the back will be curved along the four edges, completely eliminating the black bezels around the screen. It is currently unclear whether the screen itself will be curved.

The phone's frame will be compressed into an extremely narrow metal strip, located in the center of the current iPhone, to house the phone's buttons.

If the foldable iPhone is also released annually, then Apple will release 7 iPhones next year.

According to Bloomberg, John Ternus, Apple's "top successor" and then head of hardware engineering, spearheaded this major transformation of the iPhone—a roadmap dubbed the "three-year plan," starting with last year's major upgrade to the iPhone 17 Pro and the release of the iPhone Air, then this year's launch of the foldable iPhone, and finally next year's 20th-anniversary iPhone.

By launching more SKUs, the iPhone is transforming from a single phone into a family of phones spanning the price range of 3,000 to 20,000 yuan, covering everything from entry-level to ultra-high-end.

From one iPhone a year a decade ago, to the dual versions of the iPhone 6 and iPhone 6 Plus, the entry-level iPhone SE, and finally four iPhones a year in 2020, Apple has finally adopted a "sea of ​​models" strategy, aiming to cover all price segments above 3000 yuan. From now on, there will be roughly one iPhone to choose from every 2000 yuan.

• Price range of 3000-4000 yuan: iPhone e
• Price range of 5000-6000 yuan: iPhone standard version
• Price range of 7000-8000 yuan: iPhone Air
• Price range of 8000-9000 yuan: iPhone Pro
• Price range of 9000-10000 yuan: iPhone Pro Max
• Price range of 10,000-20,000 yuan: 20th Anniversary iPhone, iPhone Ultra (predicted)

Capitalize on both low and high prices; A-chips dominate the market.

The act of "buying an iPhone" is about to be completely changed.

In the past, Apple wasn't against lowering prices, but it rarely released new products specifically for the entry-level market. This segment was mainly filled by older iPhones. Whether it was Apple's official website or third-party e-commerce channels, they would continue to sell basic iPhones from 1-2 years ago, generally at prices 1000-2000 yuan lower than the newer models.

While older iPhones can still perform well, not everyone is willing to buy a two-year-old model – after all, buying one means it will have two years less of updates than a new phone, and its performance will become more strained over time.

Now, Apple offers an even better option: the iPhone "e" series.

Using the same chip as the standard version, but with one less core, this "crippled" chip still provides near-flagship performance, while also utilizing the inventory that would otherwise be wasted. Most of the other components are also readily available, yet it offers consumers a "new" iPhone at a lower price.

In contrast to "pushing down," Apple is also trying to break through the current ceiling of the iPhone and attempt to "pull up."

The iPhone Air was originally an attempt to use its design and feel as selling points to rival large screens or imaging capabilities, but the product was unsuccessful and failed to impress consumers.

The foldable iPhone is the next attempt—domestic manufacturers like Huawei have proven that foldable flagship phones can be positioned and priced higher than top-of-the-line candybar phones. This market is far more mature than "thin and light" flagships, and relying on the iPadOS ecosystem, the wide and large screens of the foldable iPhone have even more room to shine.

Clearly, Apple is no longer satisfied with the current price range of the iPhone; they want to further expand the iPhone's reach, hoping to have products that cover both the low-end and high-end markets.

The confidence and reason for doing this are all because of chip A.

Apple's utilization of its A-chip inventory has reached an astonishing level: including the seven iPhones to be released next year, as well as the iPad mini, the digital iPad, and the new MacBook Neo, Apple has already created as many as 10 SKUs around the A-chip, and most of them will be updated annually.

For products like the iPhone e series, iPad mini, and MacBook Neo, which use A-series chips and lack core inventory, Apple hardly needs to invest in production, thus "turning waste into treasure".

This week, tech media outlet Culpium reported that due to better-than-expected sales of the MacBook Neo, Apple may have to restart production of these two-year-old chips.

Chips are a product with high investment and slow returns. The A chip is launched every year, with R&D expenses reaching the level of $1 billion. The price of the A16 chip, which was released almost 4 years ago, has reached $100 per chip, but the leading advantage only lasts for one or two years.

With the prices of components, especially memory, skyrocketing in the past two years, the price of new 2nm process chips will only rise further, increasing Apple's cost pressure and forcing it to dilute its market share.

Since chip manufacturing is so expensive, we certainly can't let go of any opportunity to generate revenue. We need to ship more iPhones with A-chips, and we also need to find ways to use up the damaged versions and old inventory. After all, the performance of A-chips is indeed strong enough. Even the damaged version of the A18 Pro from two years ago is still quite good on a computer.

For established rivals like domestic Android manufacturers, chips can only be purchased from suppliers. They have to bear the ever-increasing costs and have no extra inventory to hedge against the risks.

Xiaomi, OPPO, and vivo have not only raised prices across the board for their new products, but have also quietly increased the retail prices of their existing models, with some models seeing increases of up to 1,500 yuan, leading to a sharp increase in sales pressure. IDC predicts that China's mobile phone market shipments will decline by 2.2% in 2026.

In contrast, Apple and Huawei, brands that already had higher profit margins, were able to weather the cost increases more easily thanks to their stronger control over the supply chain. They even began to make a counterattack on the low-end market: Apple's iPhone 17e offered more storage at the same price, and Huawei relaunched its Enjoy brand, returning to the nearly extinct market of budget phones.

At the same time, manufacturers like Xiaomi are unlikely to produce budget smartphones this year. Even if they do, the configurations will likely be significantly downgraded, making it difficult to maintain the "cost-effectiveness" selling point.

Whether it's the previously comfortable low-price market or the high-end price segment where Apple wasn't particularly adept, Apple is now targeting both, squeezing the entire market's survival space.

For all mobile phone manufacturers, the good days are over when there are as many new iPhones as there are new Xiaomi phones.

The only exception? Perhaps it's Samsung, which is struggling to sell enough memory.

#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.