Apple

Apple releases new beta builds of all its flashy new Liquid Glass-ified OS updates

Apple, beta, iOS, iPadOS, macOS, macos developer beta, Tech / Mike M. / June 24, 2025

Should you install these betas?

Selecting from among several beta OS versions in the Settings app on iOS 18. Credit: Andrew Cunningham

We are not highlighting this second round of developer betas because we think you should go out and install them on the Macs, iPhones, iPads, and Apple Watches that you use daily. These are still early versions, and they’re likely to have significant performance, battery, and stability problems relative to the current publicly available versions of the software.

But generally speaking, these second developer builds are the first ones I install on my secondary test devices—a collection of mostly older devices that have been replaced but are still considered current enough to run the new update. The initial builds are usually little more than a tech demo and can have major show-stopping bugs (an M1 iPad Air with the first developer beta on it simply stopped responding to any input, including a hard restart, and I needed to set it aside so its battery could drain all the way before I could do anything else with it), but the second betas tend to be somewhat more amenable to normal everyday use.

The new iOS and iPadOS betas will run on just about any hardware that can currently install and run iOS and iPadOS 18, with a couple of older exceptions. The macOS beta will run on any Apple Silicon Mac and on a handful of Intel Macs released in 2019 and 2020. The other betas will generally run on anything that supports the current versions, with some caveats (Liquid Glass effects only show up on newer Apple TV 4K boxes, for example, while the first-gen Apple TV 4K and the old 1080p Apple TV will run the update but without Liquid Glass).

If you don’t have spare devices you can dedicate to testing, we’d recommend waiting until the public beta in July before you even think about running any of these betas, and only after backing up all the important data on those devices. Rolling back to an older software version is doable, but a bit of a pain. Alternatively, those with Apple Silicon Macs who want to test the latest versions could try setting up a virtual machine using an app like VirtualBuddy or one of the others that leverages Apple’s built-in Virtualization framework.

Apple releases new beta builds of all its flashy new Liquid Glass-ified OS updates Read More »

The MacBook Air is the obvious loser as the sun sets on the Intel Mac era

Apple, Apple silicon, Features, Intel Macs, macos 26, macos 26 tahoe, macos tahoe, Tech / Kris Guyer / June 17, 2025

In the end, Intel Macs have mostly gotten a better deal than PowerPC Macs did.

For the last three years, we’ve engaged in some in-depth data analysis and tea-leaf reading to answer two questions about Apple’s support for older Macs that still use Intel chips.

First, was Apple providing fewer updates and fewer years of software support to Macs based on Intel chips as it worked to transition the entire lineup to its internally developed Apple Silicon? And second, how long could Intel Mac owners reasonably expect to keep getting updates?

The answer to the first question has always been “it depends, but generally yes.” And this year, we have a definitive answer to the second question: For the bare handful of Intel Macs it supports, macOS 26 Tahoe will be the final new version of the operating system to support any of Intel’s chips.

To its credit, Apple has also clearly spelled this out ahead of time rather than pulling the plug on Intel Macs with no notice. The company has also said that it plans to provide security updates for those Macs for two years after Tahoe is replaced by macOS 27 next year. These Macs aren’t getting special treatment—this has been Apple’s unspoken, unwritten policy for macOS security updates for decades now—but to look past its usual “we don’t comment on our future plans” stance to give people a couple years of predictability is something we’ve been pushing Apple to do for a long time.

With none of the tea leaf reading left to do, we can now present a fairly definitive look at how Apple has handled the entire Intel transition, compare it to how the PowerPC-to-Intel switch went two decades ago, and predict what it might mean about support for Apple Silicon Macs.

The data

We’ve assembled an epoch-spanning spreadsheet of every PowerPC or Intel Mac Apple has released since the original iMac kicked off the modern era of Apple back in 1998. On that list, we’ve recorded the introduction date for each Mac, the discontinuation date (when it was either replaced or taken off the market), the version of macOS it shipped with, and the final version of macOS it officially supported.

For those macOS versions, we’ve recorded the dates they received their last major point update—these are the feature-adding updates these releases get when they’re Apple’s latest and greatest version of macOS, as macOS 15 Sequoia is right now. After replacing them, Apple releases security-only patches and Safari browser updates for old macOS versions for another two years after replacing them, so we’ve also recorded the dates that those Macs would have received their final security update. For Intel Macs that are still receiving updates (versions 13, 14, and 15) and macOS 26 Tahoe, we’ve extrapolated end-of-support dates based on Apple’s past practices.

A 27-inch iMac model. It’s still the only Intel Mac without a true Apple Silicon replacement. Credit: Andrew Cunningham

We’re primarily focusing on two time spans: from the date of each Mac’s introduction to the date it stopped receiving major macOS updates, and from the date of each Mac’s introduction to the date it stopped receiving any updates at all. We consider any Macs inside either of these spans to be actively supported; Macs that are no longer receiving regular updates from Apple will gradually become less secure and less compatible with modern apps as time passes. We measure by years of support rather than number of releases, which controls for Apple’s transition to a once-yearly release schedule for macOS back in the early 2010s.

We’ve also tracked the time between each Mac model’s discontinuation and when it stopped receiving updates. This is how Apple determines which products go on its “vintage” and “obsolete” hardware lists, which determine the level of hardware support and the kinds of repairs that the company will provide.

We have lots of detailed charts, but here are some highlights:

For all Mac models tracked, the average Mac receives about 6.6 years of macOS updates that add new features, plus another two years of security-only updates.
If you only count the Intel era, the average is around seven years of macOS updates, plus two years of security-only patches.
Most (though not all) Macs released since 2016 come in lower than either of these averages, indicating that Apple has been less generous to most Intel Macs since the Apple Silicon transition began.
The three longest-lived Macs are still the mid-2007 15- and 17-inch MacBook Pros, the mid-2010 Mac Pro, and the mid-2007 iMac, which received new macOS updates for around nine years after their introduction (and security updates for around 11 years).
The shortest-lived Mac is still the late-2008 version of the white MacBook, which received only 2.7 years of new macOS updates and another 3.3 years of security updates from the time it was introduced. (Late PowerPC-era and early Intel-era Macs are all pretty bad by modern standards.)

The charts

If you bought a Mac any time between 2016 and 2020, you’re generally settling for fewer years of software updates than you would have gotten in the recent past. If you bought a Mac released in 2020, the tail end of the Intel era when Apple Silicon Macs were around the corner, your reward is the shortest software support window since 2006.

There are outliers in either direction. The sole iMac Pro, introduced in 2017 as Apple tried to regain some of its lost credibility with professional users, will end up with 7.75 years of updates plus another two years of security updates when all is said and done. Buyers of 2018–2020 MacBook Airs and the two-port version of the 2020 13-inch MacBook Pro, however, are treated pretty poorly, getting not quite 5.5 years of updates (plus two years of security patches) on average from the date they were introduced.

That said, most Macs usually end up getting a little over six years of macOS updates and two more years of security updates. If that’s a year or two lower than the recent past, it’s also not ridiculously far from the historical average.

If there’s something to praise here, it’s interesting that Apple doesn’t seem to treat any of its Macs differently based on how much they cost. Now that we have a complete overview of the Intel era, breaking out the support timelines by model rather than by model year shows that a Mac mini doesn’t get dramatically more or less support than an iMac or a Mac Pro, despite costing a fraction of the price. A MacBook Air doesn’t receive significantly more or less support than a MacBook Pro.

These are just averages, and some models are lucky while others are not. The no-adjective MacBook that Apple has sold on and off since 2006 is also an outlier, with fewer years of support on average than the other Macs.

If there’s one overarching takeaway, it’s that you should buy new Macs as close to the date of their introduction as possible if you want to maximize your software support window. Especially for Macs that were sold continuously for years and years—the 2013 and 2019 Mac Pro, the 2018 Mac mini, the non-Retina 2015 MacBook Air that Apple sold some version of for over four years—buying them toward the end of their retail lifecycle means settling for years of fewer updates than you would have gotten if you had waited for the introduction of a new model. And that’s true even though Apple’s hardware support timelines are all calculated from the date of last availability rather than the date of introduction.

It just puts Mac buyers in a bad spot when Apple isn’t prompt with hardware updates, forcing people to either buy something that doesn’t fully suit their needs or settle for something older that will last for fewer years.

What should you do with an older Intel Mac?

The big question: If your Intel Mac is still functional but Apple is no longer supporting it, is there anything you can do to keep it both secure and functional?

All late-model Intel Macs officially support Windows 10, but that OS has its own end-of-support date looming in October 2025. Windows 11 can be installed, but only if you bypass its system requirements, which can work well, but it does require additional fiddling when it comes time to install major updates. Consumer-focused Linux distributions like Ubuntu, Mint, or Pop!_OS may work, depending on your hardware, but they come with a steep learning curve for non-technical users. Google’s ChromeOS Flex may also work, but ChromeOS is more functionally limited than most other operating systems.

The OpenCore Legacy Patcher provides one possible stay of execution for Mac owners who want to stay on macOS for as long as they can. But it faces two steep uphill climbs in macOS Tahoe. First, as Apple has removed more Intel Macs from the official support list, it has removed more of the underlying code from macOS that is needed to support those Macs and other Macs with similar hardware. This leaves more for the OpenCore Configurator team to have to patch in from older OSes, and this kind of forward-porting can leave hardware and software partly functional or non-functional.

Second, there’s the Apple T2 to consider. The Macs with a T2 treat it as a load-bearing co-processor, responsible for crucial operating system functions such as enabling Touch ID, serving as an SSD controller, encoding and decoding videos, communicating with the webcam and built-in microphone, and other operations. But Apple has never opened the T2 up to anyone, and it remains a bit of a black box for both the OpenCore/Hackintosh community and folks who would run Linux-based operating systems like Ubuntu or ChromeOS on that hardware.

The result is that the 2018 and 2019 MacBook Airs that didn’t support macOS 15 Sequoia last year never had support for them added to the OpenCore Legacy Patcher because the T2 chip simply won’t communicate with OpenCore firmware booted. Some T2 Macs don’t have this problem. But if yours does, it’s unlikely that anyone will be able to do anything about it, and your software support will end when Apple says it does.

Does any of this mean anything for Apple Silicon Mac support?

Late-model Intel MacBook Airs have fared worse than other Macs in terms of update longevity. Credit: Valentina Palladino

It will likely be at least two or three years before we know for sure how Apple plans to treat Apple Silicon Macs. Will the company primarily look at specs and technical capabilities, as it did from the late-’90s through to the mid-2010s? Or will Apple mainly stop supporting hardware based on its age, as it has done for more recent Macs and most current iPhones and iPads?

The three models to examine for this purpose are the first ones to shift to Apple Silicon: the M1 versions of the MacBook Air, Mac mini, and 13-inch MacBook Pro, all launched in late 2020. If these Macs are dropped in, say, 2027 or 2028’s big macOS release, but other, later M1 Macs like the iMac stay supported, it means Apple is likely sticking to a somewhat arbitrary age-based model, with certain Macs cut off from software updates that they are perfectly capable of running.

But it’s our hope that all Apple Silicon Macs have a long life ahead of them. The M2, M3, and M4 have all improved on the M1’s performance and other capabilities, but the M1 Macs are much more capable than the Intel ones they supplanted, the M1 was used so widely in various Mac models for so long, and Mac owners can pay so much more for their devices than iPhone and iPad owners. We’d love to see macOS return to the longer-tail software support it provided in the late-’00s and mid-2010s, when models could expect to see seven or eight all-new macOS versions and another two years of security updates afterward.

All signs point to Apple using the launch date of any given piece of hardware as the determining factor for continued software support. But that isn’t how it has always been, nor is it how it always has to be.

Andrew is a Senior Technology Reporter at Ars Technica, with a focus on consumer tech including computer hardware and in-depth reviews of operating systems like Windows and macOS. Andrew lives in Philadelphia and co-hosts a weekly book podcast called Overdue.

The MacBook Air is the obvious loser as the sun sets on the Intel Mac era Read More »

Coming to Apple OSes: A seamless, secure way to import and export passkeys

Apple, Biz & IT, iOS, iPadOS, macOS, Security, wwdc 2025 / Rejus Almole / June 13, 2025

As the video explains:

This new process is fundamentally different and more secure than traditional credential export methods, which often involve exporting an unencrypted CSV or JSON file, then manually importing it into another app. The transfer process is user initiated, occurs directly between participating credential manager apps and is secured by local authentication like Face ID.

This transfer uses a data schema that was built in collaboration with the members of the FIDO Alliance. It standardizes the data format for passkeys, passwords, verification codes, and more data types.

The system provides a secure mechanism to move the data between apps. No insecure files are created on disk, eliminating the risk of credential leaks from exported files. It’s a modern, secure way to move credentials.

The push to passkeys is fueled by the tremendous costs associated with passwords. Creating and managing a sufficiently long, randomly generated password for each account is a burden on many users, a difficulty that often leads to weak choices and reused passwords. Leaked passwords have also been a chronic problem.

Passkeys, in theory, provide a means of authentication that’s immune to credential phishing, password leaks, and password spraying. Under the latest “FIDO2” specification, it creates a unique public/private encryption keypair during each website or app enrollment. The keys are generated and stored on a user’s phone, computer, YubiKey, or similar device. The public portion of the key is sent to the account service. The private key remains bound to the user device, where it can’t be extracted. During sign-in, the website or app server sends the device that created the key pair a challenge in the form of pseudo-random data. Authentication occurs only when the device signs the challenge using the corresponding private key and sends it back.

This design ensures that there is no shared secret that ever leaves the user’s device. That means there’s no data to be sniffed in transit, phished, or compromised through other common methods.

As I noted in December, the biggest thing holding back passkeys at the moment is their lack of usability. Apps, OSes, and websites are, in many cases, islands that don’t interoperate with their peers. Besides potentially locking users out of their accounts, the lack of interoperability also makes passkeys too difficult for many people.

Apple’s demo this week provides the strongest indication yet that passkey developers are making meaningful progress in improving usability.

Coming to Apple OSes: A seamless, secure way to import and export passkeys Read More »

Apple’s Craig Federighi on the long road to the iPad’s Mac-like multitasking

Apple, iOS, ipad, iPadOS, macOS, Tech, wwdc / Kris Guyer / June 12, 2025

Federighi talks to Ars about why the iPad’s Mac-style multitasking took so long.

iPads! Running iOS 26! Credit: Apple

CUPERTINO, Calif.—When Apple Senior Vice President of Software Engineering Craig Federighi introduced the new multitasking UI in iPadOS 26 at the company’s Worldwide Developers Conference this week, he did it the same way he introduced the Calculator app for the iPad last year or timers in the iPad’s Clock app the year before—with a hint of sarcasm.

“Wow,” Federighi enthuses in a lightly exaggerated tone about an hour and 19 minutes into a 90-minute presentation. “More windows, a pointier pointer, and a menu bar? Who would’ve thought? We’ve truly pulled off a mind-blowing release!”

This elicits a sensible chuckle from the gathered audience of developers, media, and Apple employees watching the keynote on the Apple Park campus, where I have grabbed myself a good-but-not-great seat to watch the largely pre-recorded keynote on a gigantic outdoor screen.

Federighi is acknowledging—and lightly poking fun at—the audience of developers, pro users, and media personalities who have been asking for years that Apple’s iPad behave more like a traditional computer. And after many incremental steps, including a big swing and partial miss with the buggy, limited Stage Manager interface a couple of years ago, Apple has finally responded to requests for Mac-like multitasking with a distinctly Mac-like interface, an improved file manager, and better support for running tasks in the background.

But if this move was so forehead-slappingly obvious, why did it take so long to get here? This is one of the questions we dug into when we sat down with Federighi and Senior Vice President of Worldwide Marketing Greg Joswiak for a post-keynote chat earlier this week.

It used to be about hardware restrictions

People have been trying to use iPads (and make a philosophical case for them) as quote-unquote real computers practically from the moment they were introduced 15 years ago.

But those early iPads lacked so much of what we expect from modern PCs and Macs, most notably robust multi-window multitasking and the ability for third-party apps to exchange data. The first iPads were almost literally just iPhone internals connected to big screens, with just a fraction of the RAM and storage available in the Macs of the day; that necessitated the use of a blown-up version of the iPhone’s operating system and the iPhone’s one-full-screen-app-at-a-time interface.

“If you want to rewind all the way to the time we introduced Split View and Slide Over [in iOS 9], you have to start with the grounding that the iPad is a direct manipulation touch-first device,” Federighi told Ars. “It is a foundational requirement that if you touch the screen and start to move something, that it responds. Otherwise, the entire interaction model is broken—it’s a psychic break with your contract with the device.”

Mac users, Federighi said, were more tolerant of small latency on their devices because they were already manipulating apps on the screen indirectly, but the iPads of a decade or so ago “didn’t have the capacity to run an unlimited number of windowed apps with perfect responsiveness.”

It’s also worth noting the technical limitations of iPhone and iPad apps at the time, which up until then had mostly been designed and coded to match the specific screen sizes and resolutions of the (then-manageable) number of iDevices that existed. It simply wasn’t possible for the apps of the day to be dynamically resized as desktop windows are, because no one was coding their apps that way.

Apple’s iPad Pros—and, later, the iPad Airs—have gradually adopted hardware and software features that make them more Mac-like. Credit: Andrew Cunningham

Of course, those hardware limitations no longer exist. Apple’s iPad Pros started boosting the tablets’ processing power, RAM, and storage in earnest in the late 2010s, and Apple introduced a Microsoft Surface-like keyboard and stylus accessories that moved the iPad away from its role as a content consumption device. For years now, Apple’s faster tablets have been based on the same hardware as its slower Macs—we know the hardware can do more because Apple is already doing more with it elsewhere.

“Over time the iPad’s gotten more powerful, the screens have gotten larger, the user base has shifted into a mode where there is a little bit more trackpad and keyboard use in how many people use the device,” Federighi told Ars. “And so the stars kind of aligned to where many of the things that you traditionally do with a Mac were possible to do on an iPad for the first time and still meet iPad’s basic contract.”

On correcting some of Stage Manager’s problems

More multitasking in iPadOS 26. Credit: Apple

Apple has already tried a windowed multitasking system on modern iPads once this decade, of course, with iPadOS 16’s Stage Manager interface.

Any first crack at windowed multitasking on the iPad was going to have a steep climb. This was the first time Apple or its developers had needed to contend with truly dynamically resizable app windows in iOS or iPadOS, the first time Apple had implemented a virtual memory system on the iPad, and the first time Apple had tried true multi-monitor support. Stage Manager was in such rough shape that Apple delayed that year’s iPadOS release to keep working on it.

But the biggest problem with Stage Manager was actually that it just didn’t work on a whole bunch of iPads. You could only use it on new expensive models—if you had a new cheap model or even an older expensive model, your iPad was stuck with the older Slide Over and Split View modes that had been designed around the hardware limitations of mid-2010s iPads.

“We wanted to offer a new baseline of a totally consistent experience of what it meant to have Stage Manager,” Federighi told Ars. “And for us, that meant four simultaneous apps on the internal display and an external display with four simultaneous apps. So, eight apps running at once. And we said that’s the baseline, and that’s what it means to be Stage Manager; we didn’t want to say ‘you get Stage Manager, but you get Stage Manager-lite here or something like that. And so immediately that established a floor for how low we could go.”

Fixing that was one of the primary goals of the new windowing system.

“We decided this time: make everything we can make available,” said Federighi, “even if it has some nuances on older hardware, because we saw so much demand [for Stage Manager].”

That slight change in approach, combined with other behind-the-scenes optimizations, makes the new multitasking model more widely compatible than Stage Manager is. There are still limits on those devices—not to the number of windows you can open, but to how many of those windows can be active and up-to-date at once. And true multi-monitor support would remain the purview of the faster, more-expensive models.

“We have discovered many, many optimizations,” Federighi said. “We re-architected our windowing system and we re-architected the way that we manage background tasks, background processing, that enabled us to squeeze more out of other devices than we were able to do at the time we introduced Stage Manager.”

Stage Manager still exists in iPadOS 26, but as an optional extra multitasking mode that you have to choose to enable instead of the new windowed multitasking system. You can also choose to turn both multitasking systems off entirely, preserving the iPad’s traditional big-iPhone-for-watching-Netflix interface for the people who prefer it.

“iPad’s gonna be iPad”

The $349 base-model iPad is one that stands to gain the most from iPadOS 26. Credit: Andrew Cunningham

However, while the new iPadOS 26 UI takes big steps toward the Mac’s interface, the company still tries to treat them as different products with different priorities. To date, that has meant no touch screens on the Mac (despite years of rumors), and it will continue to mean that there are some Mac things that the iPad will remain unable to do.

“But we’ve looked and said, as [the iPad and Mac] come together, where on the iPad the Mac idiom for doing something, like where we put the window close controls and maximize controls, what color are they—we’ve said why not, where it makes sense, use a converged design for those things so it’s familiar and comfortable,” Federighi told Ars. “But where it doesn’t make sense, iPad’s gonna be iPad.”

There will still be limitations and frustrations when trying to fit an iPad into a Mac-shaped hole in your computing setup. While tasks can run in the background, for example, Apple only allows apps to run workloads with a definitive endpoint, things like a video export or a file transfer. System agents or other apps that perform some routine on-and-off tasks continuously in the background aren’t supported. All the demos we’ve seen so far are also on new, high-end iPad hardware, and it remains to be seen how well the new features behave on low-end tablets like the 11th-generation A16 iPad, or old 2019-era hardware like the iPad Air 3.

But it does feel like Apple has finally settled on a design that might stick and that adds capability to the iPad without wrecking its simplicity for the people who still just want a big screen for reading and streaming.

Apple’s Craig Federighi on the long road to the iPad’s Mac-like multitasking Read More »

New Apple study challenges whether AI models truly “reason” through problems

AI, AI benchmarks, AI research, Apple, Apple Research, machine learning, simulated reasoning / Kris Guyer / June 12, 2025

Puzzle-based experiments reveal limitations of simulated reasoning, but others dispute findings.

An illustration of Tower of Hanoi from Popular Science in 1885. Credit: Public Domain

In early June, Apple researchers released a study suggesting that simulated reasoning (SR) models, such as OpenAI’s o1 and o3, DeepSeek-R1, and Claude 3.7 Sonnet Thinking, produce outputs consistent with pattern-matching from training data when faced with novel problems requiring systematic thinking. The researchers found similar results to a recent study by the United States of America Mathematical Olympiad (USAMO) in April, showing that these same models achieved low scores on novel mathematical proofs.

The new study, titled “The Illusion of Thinking: Understanding the Strengths and Limitations of Reasoning Models via the Lens of Problem Complexity,” comes from a team at Apple led by Parshin Shojaee and Iman Mirzadeh, and it includes contributions from Keivan Alizadeh, Maxwell Horton, Samy Bengio, and Mehrdad Farajtabar.

The researchers examined what they call “large reasoning models” (LRMs), which attempt to simulate a logical reasoning process by producing a deliberative text output sometimes called “chain-of-thought reasoning” that ostensibly assists with solving problems in a step-by-step fashion.

To do that, they pitted the AI models against four classic puzzles—Tower of Hanoi (moving disks between pegs), checkers jumping (eliminating pieces), river crossing (transporting items with constraints), and blocks world (stacking blocks)—scaling them from trivially easy (like one-disk Hanoi) to extremely complex (20-disk Hanoi requiring over a million moves).

Figure 1 from Apple's — Figure 1 from Apple’s “The Illusion of Thinking” research paper. Credit: Apple

“Current evaluations primarily focus on established mathematical and coding benchmarks, emphasizing final answer accuracy,” the researchers write. In other words, today’s tests only care if the model gets the right answer to math or coding problems that may already be in its training data—they don’t examine whether the model actually reasoned its way to that answer or simply pattern-matched from examples it had seen before.

Ultimately, the researchers found results consistent with the aforementioned USAMO research, showing that these same models achieved mostly under 5 percent on novel mathematical proofs, with only one model reaching 25 percent, and not a single perfect proof among nearly 200 attempts. Both research teams documented severe performance degradation on problems requiring extended systematic reasoning.

Known skeptics and new evidence

AI researcher Gary Marcus, who has long argued that neural networks struggle with out-of-distribution generalization, called the Apple results “pretty devastating to LLMs.” While Marcus has been making similar arguments for years and is known for his AI skepticism, the new research provides fresh empirical support for his particular brand of criticism.

“It is truly embarrassing that LLMs cannot reliably solve Hanoi,” Marcus wrote, noting that AI researcher Herb Simon solved the puzzle in 1957 and many algorithmic solutions are available on the web. Marcus pointed out that even when researchers provided explicit algorithms for solving Tower of Hanoi, model performance did not improve—a finding that study co-lead Iman Mirzadeh argued shows “their process is not logical and intelligent.”

Figure 4 from Apple's — Figure 4 from Apple’s “The Illusion of Thinking” research paper. Credit: Apple

The Apple team found that simulated reasoning models behave differently from “standard” models (like GPT-4o) depending on puzzle difficulty. On easy tasks, such as Tower of Hanoi with just a few disks, standard models actually won because reasoning models would “overthink” and generate long chains of thought that led to incorrect answers. On moderately difficult tasks, SR models’ methodical approach gave them an edge. But on truly difficult tasks, including Tower of Hanoi with 10 or more disks, both types failed entirely, unable to complete the puzzles, no matter how much time they were given.

The researchers also identified what they call a “counterintuitive scaling limit.” As problem complexity increases, simulated reasoning models initially generate more thinking tokens but then reduce their reasoning effort beyond a threshold, despite having adequate computational resources.

The study also revealed puzzling inconsistencies in how models fail. Claude 3.7 Sonnet could perform up to 100 correct moves in Tower of Hanoi but failed after just five moves in a river crossing puzzle—despite the latter requiring fewer total moves. This suggests the failures may be task-specific rather than purely computational.

Competing interpretations emerge

However, not all researchers agree with the interpretation that these results demonstrate fundamental reasoning limitations. University of Toronto economist Kevin A. Bryan argued on X that the observed limitations may reflect deliberate training constraints rather than inherent inabilities.

“If you tell me to solve a problem that would take me an hour of pen and paper, but give me five minutes, I’ll probably give you an approximate solution or a heuristic. This is exactly what foundation models with thinking are RL’d to do,” Bryan wrote, suggesting that models are specifically trained through reinforcement learning (RL) to avoid excessive computation.

Bryan suggests that unspecified industry benchmarks show “performance strictly increases as we increase in tokens used for inference, on ~every problem domain tried,” but notes that deployed models intentionally limit this to prevent “overthinking” simple queries. This perspective suggests the Apple paper may be measuring engineered constraints rather than fundamental reasoning limits.

Figure 6 from Apple's — Figure 6 from Apple’s “The Illusion of Thinking” research paper. Credit: Apple

Software engineer Sean Goedecke offered a similar critique of the Apple paper on his blog, noting that when faced with Tower of Hanoi requiring over 1,000 moves, DeepSeek-R1 “immediately decides ‘generating all those moves manually is impossible,’ because it would require tracking over a thousand moves. So it spins around trying to find a shortcut and fails.” Goedecke argues this represents the model choosing not to attempt the task rather than being unable to complete it.

Other researchers also question whether these puzzle-based evaluations are even appropriate for LLMs. Independent AI researcher Simon Willison told Ars Technica in an interview that the Tower of Hanoi approach was “not exactly a sensible way to apply LLMs, with or without reasoning,” and suggested the failures might simply reflect running out of tokens in the context window (the maximum amount of text an AI model can process) rather than reasoning deficits. He characterized the paper as potentially overblown research that gained attention primarily due to its “irresistible headline” about Apple claiming LLMs don’t reason.

The Apple researchers themselves caution against over-extrapolating the results of their study, acknowledging in their limitations section that “puzzle environments represent a narrow slice of reasoning tasks and may not capture the diversity of real-world or knowledge-intensive reasoning problems.” The paper also acknowledges that reasoning models show improvements in the “medium complexity” range and continue to demonstrate utility in some real-world applications.

Implications remain contested

Have the credibility of claims about AI reasoning models been completely destroyed by these two studies? Not necessarily.

What these studies may suggest instead is that the kinds of extended context reasoning hacks used by SR models may not be a pathway to general intelligence, like some have hoped. In that case, the path to more robust reasoning capabilities may require fundamentally different approaches rather than refinements to current methods.

As Willison noted above, the results of the Apple study have so far been explosive in the AI community. Generative AI is a controversial topic, with many people gravitating toward extreme positions in an ongoing ideological battle over the models’ general utility. Many proponents of generative AI have contested the Apple results, while critics have latched onto the study as a definitive knockout blow for LLM credibility.

Apple’s results, combined with the USAMO findings, seem to strengthen the case made by critics like Marcus that these systems rely on elaborate pattern-matching rather than the kind of systematic reasoning their marketing might suggest. To be fair, much of the generative AI space is so new that even its inventors do not yet fully understand how or why these techniques work. In the meantime, AI companies might build trust by tempering some claims about reasoning and intelligence breakthroughs.

However, that doesn’t mean these AI models are useless. Even elaborate pattern-matching machines can be useful in performing labor-saving tasks for the people that use them, given an understanding of their drawbacks and confabulations. As Marcus concedes, “At least for the next decade, LLMs (with and without inference time “reasoning”) will continue have their uses, especially for coding and brainstorming and writing.”

Benj Edwards is Ars Technica’s Senior AI Reporter and founder of the site’s dedicated AI beat in 2022. He’s also a tech historian with almost two decades of experience. In his free time, he writes and records music, collects vintage computers, and enjoys nature. He lives in Raleigh, NC.

New Apple study challenges whether AI models truly “reason” through problems Read More »

Apple details the end of Intel Mac support and a phaseout for Rosetta 2

Apple, Apple silicon, Macs, Rosetta, Rosetta 2, Tech / Kris Guyer / June 10, 2025

The support list for macOS Tahoe still includes Intel Macs, but it has been whittled down to just four models, all released in 2019 or 2020. We speculated that this meant that the end was near for Intel Macs, and now we can confirm just how near it is: macOS Tahoe will be the last new macOS release to support any Intel Macs. All new releases starting with macOS 27 will require an Apple Silicon Mac.

Apple will provide additional security updates for Tahoe until fall 2028, two years after it is replaced with macOS 27. That’s a typical schedule for older macOS versions, which all get one year of major point updates that include security fixes and new features, followed by two years of security-only updates to keep them patched but without adding significant new features.

Apple is also planning changes to Rosetta 2, the Intel-to-Arm app translation technology created to ease the transition between the Intel and Apple Silicon eras. Rosetta will continue to work as a general-purpose app translation tool in both macOS 26 and macOS 27.

But after that, Rosetta will be pared back and will only be available to a limited subset of apps—specifically, older games that rely on Intel-specific libraries but are no longer being actively maintained by their developers. Devs who want their apps to continue running on macOS after that will need to transition to either Apple Silicon-native apps or universal apps that run on either architecture.

Apple details the end of Intel Mac support and a phaseout for Rosetta 2 Read More »

Apple tiptoes with modest AI updates while rivals race ahead

AI, Apple, Apple developers, apple intelligence, machine learning, Siri, wwdc, wwdc 2025, Xcode / Kris Guyer / June 10, 2025

Developers, developers, developers?

Being the Worldwide Developers Conference, it seems appropriate that Apple also announced it would open access to its on-device AI language model to third-party developers. It also announced it would integrate OpenAI’s code completion tools into its XCode development software.

Craig Federighi stands in front of a screen with the words — Apple Intelligence was first unveiled at WWDC 2024. Credit: Apple

“We’re opening up access for any app to tap directly into the on-device, large language model at the core of Apple,” said Craig Federighi, Apple’s software chief, during the presentation. The company also demonstrated early partner integration by adding OpenAI’s ChatGPT image generation to its Image Playground app, though it said user data would not be shared without permission.

For developers, Apple’s inclusion of ChatGPT’s code-generation capabilities in XCode may represent Apple’s attempt to match what rivals like GitHub Copilot and Cursor offer software developers in terms of AI coding augmentation, even as the company maintains a more cautious approach to consumer-facing AI features.

Meanwhile, competitors like Meta, Anthropic, OpenAI, and Microsoft continue to push more aggressively into the AI space, offering AI assistants (that admittedly still make things up and suffer from other issues, such as sycophancy).

Only time will tell if Apple’s wariness to embrace the bleeding edge of AI will be a curse (eventually labeled as a blunder) or a blessing (lauded as a wise strategy). Perhaps, in time, Apple will step in with a solid and reliable AI assistant solution that makes Siri useful again. But for now, Apple Intelligence remains more of a clever brand name than a concrete set of notable products.

Apple tiptoes with modest AI updates while rivals race ahead Read More »

Apple drops support for just three iPhone and iPad models from iOS and iPadOS 26

Apple, iOS, iPadOS, Tech / Rejus Almole / June 10, 2025

Every year, Apple releases new versions of iOS and iPadOS, and most years those updates also end support for a handful of devices that are too old or too slow or otherwise incapable of running the new software.

Though this year’s macOS 26 Tahoe release was unkind to Intel Macs, the iOS 26 and iPadOS 26 releases are more generous, dropping support for just two iPhone models and a single iPad. The iOS 26 update won’t run on 2018’s iPhone XR or XS, and iPadOS 26 won’t run on 2019’s 7th-generation iPad. Any other device that can currently run iOS or iPadOS 18 will be able to upgrade to the new versions and pick up the new Liquid Glass look, among other features.

Apple never provides explicit reasoning for why it drops the devices it drops, though they can usually be explained by some combination of age and technical capability. The 7th-gen iPad, for example, was still using a 2017-vintage Apple A10X chip despite being introduced a number of years later.

The iPhone XR and XS, on the other hand, use an Apple A12 chip, the same one used by several still-supported iPads.

Apple usually provides security-only patches for dispatched iDevices for a year or two after they stop running the newest OS, though the company never publishes timelines for these updates, and the iPhone and iPad haven’t been treated as reliably as Macs have. Still, if you do find yourself relying on one of those older devices, you can still probably wring a bit of usefulness out of it before you start missing out on critical security patches.

Apple drops support for just three iPhone and iPad models from iOS and iPadOS 26 Read More »

Bill Atkinson, architect of the Mac’s graphical soul, dies at 74

Andy Hertzfeld, Apple, Bill Atkinson, Folklore.org, Mac history, Macintosh, Obituaries, obituary, retrotech, Steve Jobs, Tech, Vintage computing / Rejus Almole / June 9, 2025

Using HyperCard, Teachers created interactive lessons, artists built multimedia experiences, and businesses developed custom database applications—all without writing traditional code. The hypermedia environment also had a huge impact on gaming: 1993 first-person adventure hit Myst originally used HyperCard as its game engine.

An example of graphical dithering, which allows 1-bit color (black and white only) to imitate grayscale. Credit: Benj Edwards / Apple

For the two-color Macintosh (which could only display black or white pixels, with no gradient in between), Atkinson developed an innovative high-contrast dithering algorithm that created the illusion of grayscale images with a characteristic stippled appearance that became synonymous with early Mac graphics. The dithered aesthetic remains popular today among some digital artists and indie game makers, with modern tools like this web converter that allows anyone to transform photos into the classic Atkinson dither style.

Life after Apple

After leaving Apple in 1990, Atkinson co-founded General Magic with Marc Porat and Andy Hertzfeld, attempting to create personal communicators before smartphones existed. Wikipedia notes that in 2007, he joined Numenta, an AI startup, declaring their work on machine intelligence “more fundamentally important to society than the personal computer and the rise of the Internet.”

In his later years, Atkinson pursued nature photography with the same artistry he’d brought to programming. His 2004 book “Within the Stone” featured close-up images of polished rocks that revealed hidden worlds of color and pattern.

Atkinson announced his pancreatic cancer diagnosis in November 2024, writing on Facebook that he had “already led an amazing and wonderful life.” The same disease claimed his friend and collaborator Steve Jobs in 2011.

Given Atkinson’s deep contributions to Apple history, it’s not surprising that Jobs’ successor, Apple CEO Tim Cook, paid tribute to the Mac’s original graphics guru on X on Saturday. “We are deeply saddened by the passing of Bill Atkinson,” Cook wrote. “He was a true visionary whose creativity, heart, and groundbreaking work on the Mac will forever inspire us.”

Bill Atkinson, architect of the Mac’s graphical soul, dies at 74 Read More »

Apple’s AI-driven Stem Splitter audio separation tech has hugely improved in a year

AI, Apple, izotope, logic, RX11, stem splitter / Kris Guyer / June 9, 2025

Consider an example from a song I’ve been working on. Here’s a snippet of the full piece:

After running Logic’s original Stem Splitter on the snippet, I was given four tracks: Vocals, Drums, Bass, and “Other.” They all isolated their parts reasonably well, but check out the static and artifacting when you isolate the bass track:

The vocal track came out better, but it was still far from ideal:

Now, just over a year later, Apple has released a point update for Logic that delivers “enhanced audio fidelity” for Stem Splitter—along with support for new stems for guitar and piano.

screenshot of logic's new stem splitter feature — Logic now splits audio into more stems.

The difference in quality is significant, as you can hear in the new bass track:

And the new vocal track, though still lacking the pristine fidelity of the original recording, is nevertheless greatly improved:

The ability to separate out guitars and pianos is also welcome, and it works well. Here’s the piano part:

Pretty impressive leap in fidelity for a point release!

There are plenty of other stem-splitting tools, of course, and many have had a head start on Apple. With its new release, however, Apple has certainly closed the gap.

Izotope’s RX 11, for instance, is a highly regarded (and expensive!) piece of software that can do wonders when it comes to repairing audio and reducing clicks, background noise, and sibilance.

RX11 screenshot — RX11, ready to split some stems.

It includes a stem-splitting feature that can produce four outputs (vocal, bass, drums, and other), and it produces usable audio—but I’m not sure I’d rank its output more highly than Logic’s. Compare for yourself on the vocal and bass stems:

In any event, the AI/machine learning revolution has certainly arrived in the music world, and the rapid quality increase in stem-splitting tools in just a few years shows just what these AI systems are capable of when trained on enough data. I remain especially impressed by how the best stem splitters can extract not just a clean vocal but also the reverb/delay tail. Having access to the original recordings will always be better—but stem-splitting tech is improving quickly.

Apple’s AI-driven Stem Splitter audio separation tech has hugely improved in a year Read More »

What to expect from Apple’s Worldwide Developers Conference next week

Apple, ios 26, ipados 26, macos 26, Tech, visionOS, watchos 26, wwdc 2025 / Mike M. / June 8, 2025

i wwdc what you did there

We expect to see new designs, new branding, and more at Apple’s WWDC 2025.

Apple’s Worldwide Developers Conference kicks off on Monday with the company’s standard keynote presentation—a combination of PR about how great Apple and its existing products are and a first look at the next-generation versions of iOS, iPadOS, macOS, and the company’s other operating systems.

Reporting before the keynote rarely captures everything that Apple has planned at its presentations, but the reliable information we’ve seen so far is that Apple will keep the focus on its software this year rather than using the keynote to demo splashy new hardware like the Vision Pro and Apple Silicon Mac Pro, which the company introduced at WWDC a couple years back.

If you haven’t been keeping track, here are a few of the things that are most likely to happen when the pre-recorded announcement videos start rolling next week.

Redesign time

Reliable reports from Bloomberg’s Mark Gurman have been saying for months that Apple’s operating systems are getting a design overhaul at WWDC.

The company apparently plans to use the design of the Vision Pro’s visionOS software as a jumping-off point for the new designs, introducing more transparency and UI elements that appear to be floating on the surface of your screen. Apple’s overarching goal, according to Gurman, is to “simplify the way users navigate and control their devices” by “updating the style of icons, menus, apps, windows and system buttons.”

Apple’s airy, floaty visionOS will apparently serve as the inspiration for its next-generation software design. Credit: Apple

Any good software redesign needs to walk a tightrope between freshening up an old look and solving old problems without changing peoples’ devices so much that they become unrecognizable and unfamiliar. The number of people who have complained to me about the iOS 18-era redesign of the Photos app suggests to me that Apple doesn’t always strike the right balance. But a new look can also generate excitement and encourage upgrades more readily than some of the low-profile or under-the-hood improvements that these updates normally focus on.

The redesigned UI should be released simultaneously for iOS, iPadOS, and macOS. The Mac last received a significant facelift back in 2020 with macOS 11 Big Sur, though this was overshadowed at the time by the much more significant shift from Intel’s chips to Apple Silicon. The current iOS and iPadOS design has its roots in 2013’s iOS 7, though with over a decade’s worth of gradual evolution on top.

An OS by any other name

With the new design will apparently come a new naming scheme, shifting from the current version numbers to new numbers based on the year. So we allegedly won’t be seeing iOS 19, macOS 16, watchOS 12, or visionOS 3—instead, we’ll get iOS 26, macOS 26, watchOS 26, and visionOS 26.

The new numbers might be a little confusing at first, especially for the period of overlap where Apple is actively supporting (say) macOS 14, macOS 15, and macOS 26. But in the long run, the consistency should make it easier to tell roughly how old your software is and will also make it easier to tell whether your device is running current software without having to remember the number for each of your individual devices.

It also unifies the approach to any new operating system variants Apple might announce—tvOS starts at version 9 and iPadOS starts at version 13, for example, because they were linked to the then-current iOS release. But visionOS and watchOS both started over from 1.0, and the macOS version is based on the year that Apple arbitrarily decided to end the 20-year-old “macOS X” branding and jump up to 11.

Note that those numbers will use the upcoming year rather than the current year—iOS 26 will be Apple’s latest and greatest OS for about three months in 2025, assuming the normal September-ish launch, but it will be the main OS for nine months in 2026. Apple usually also waits until later in the fall or winter to start forcing people onto the new OS, issuing at least a handful of security-only updates for the outgoing OS for people who don’t want to be guinea pigs for a possibly buggy new release.

Seriously, don’t get your hopes up about hardware

Apple showed off Vision Pro at WWDC in 2023, but we’re not expecting to see much hardware this year. Credit: Samuel Axon

Gurman has reported that Apple had “no major new devices ready to ship” this year.

Apple generally concentrates its hardware launches to the spring and fall, with quieter and lower-profile launches in the spring and bigger launches in the fall, anchored by the tentpole that is the iPhone. But WWDC has occasionally been a launching point for new Macs (because Macs are the only systems that run Xcode, Apple’s development environment) and occasionally brand-new platforms (because getting developers on board with new platforms is one way to increase their chances of success). But the best available information suggests that neither of those things is happening this time around.

There are possibilities, though. Apple has apparently been at work behind the scenes on expanding its smart home footprint, and the eternally neglected Mac Pro is still using an M2 Ultra when an M3 Ultra already exists. But especially with a new redesign to play up, we’d expect Apple to keep the spotlight on its software this time around.

The fate of Intel Macs

It’s been five years since Apple started moving from Intel’s chips to its own custom silicon in Macs and two years since Apple sold its last Intel Macs. And since the very start of the transition, Apple has resisted providing a firm answer to the question of when Intel Macs will stop getting new macOS updates.

Our analysis of years of support data suggests two likely possibilities: that Apple releases one more new version of macOS for Intel Macs before shifting to a couple years of security-only updates or that Apple pulls the plug and shifts to security-only updates this year.

Rumors suggest that current betas still run on the last couple rounds of Intel Macs, dropping support for some older or slower models introduced between 2018 and 2020. If that’s true, there’s a pretty good chance it’s the last new macOS version to officially support Intel CPUs. Regardless, we’ll know more when the first betas drop after the keynote.

Even if the new version of macOS supports some Intel Macs, expect the list of features that require Apple Silicon to keep getting longer.

iPad multitasking? Again?

The perennial complaint about high-end iPads is that the hardware is a lot more capable than the software allows it to be. And every couple of years, Apple takes another crack at making the iPad a viable laptop replacement by improving the state of multitasking on the platform. This will allegedly be another one of those years.

We don’t know much about what form these multitasking improvements will take—whether they’re a further refinement of existing features like Stage Manager or something entirely new. The changes have been described as “more like macOS,” but that could mean pretty much anything.

Playing games

People play plenty of games on Apple’s devices, but they still aren’t really a “destination” for gaming in the same way that a dedicated console or Windows PC is. The company is apparently hoping to change that with a new unified app for games. Like Valve’s Steam, the app will reportedly serve as a storefront, launcher, and achievement tracker, and will also facilitate communication between friends playing the same game.

Apple took a similar stab at this idea in the early days of the iPhone with Game Center, which still exists as a service in the background on modern Apple devices but was discontinued as a standalone app quite a few years ago.

Apple has been trying for a few years now to make its operating systems more hospitable to gaming, especially in macOS. The company has added a low-latency Game Mode to macOS and comprehensive support for modern wireless gamepads from Microsoft, Sony, and Nintendo. The company’s Game Porting Toolkit stops short of being a consumer-friendly way to run Windows games on macOS, but it does give developers of Windows games an easier on-ramp for testing and porting their games to Apple’s platforms. We’ll see whether a unified app can help any of these other gaming features gel into something that feels cohesive.

Going home

A smart speaker about the size of a mason jar. — Might we see a more prominent, marketable name for what Apple currently calls the “HomePod Software”? Credit: Jeff Dunn

One of Apple’s long-simmering behind-the-scenes hardware projects is apparently a new kind of smart home device that weds the HomePod’s current capabilities with a vaguely Apple TV-like touchscreen interface. In theory, this device would compete with the likes of Amazon’s Echo Show devices.

Part of those plans involve a “new” operating system to replace what is known to the public as “HomePod Software” (and internally as audioOS). This so-called “homeOS” has been rumored for a bit, and some circumstantial evidence points to some possible pre-WWDC trademark activity around that name. Like the current HomePod software—and just about every other OS Apple maintains—homeOS would likely be a specialized offshoot of iOS. But even if it doesn’t come with new hardware right away, new branding could suggest that Apple is getting ready to expand its smart home ambitions.

What about AI?

Finally, it wouldn’t be a mid-2020s tech keynote without some kind of pronouncements about AI. Last year’s WWDC was the big public unveiling of Apple Intelligence, and (nearly) every one of Apple’s product announcements since then has made a point of highlighting the hardware’s AI capabilities.

We’d definitely expect Apple to devote some time to Apple Intelligence, but the company may be more hesitant to announce big new features in advance, following a news cycle where even normally sympathetic Apple boosters like Daring Fireball’s John Gruber excoriated the company for promising AI features that it was nowhere near ready to launch—or even to demo to the public. The executives handling Apple’s AI efforts were reshuffled following that news cycle; whether it was due to Gruber’s piece or the underlying problems outlined in the article is anyone’s guess.

Apple will probably try to find a middle road, torn between not wanting to overpromise and underdeliver and not wanting to seem “behind” on the tech industry’s biggest craze. There’s a decent chance that the new “more personalized” version of Siri will finally make a public appearance. But I’d guess that Apple will focus more on iterations of existing Apple Intelligence features like summaries or Writing Tools rather than big swings.

What to expect from Apple’s Worldwide Developers Conference next week Read More »

Endangered classic Mac plastic color returns as 3D-printer filament

3d printers, 3D printing, Apple, Apple IIgs, classic mac, computer collecting, Joe Strosnider, Macintosh, Macintosh Platinum, Macintosh SE, Macs, retrocomputing, retrotech, Tech, Vintage computing / Mike M. / June 5, 2025

On Tuesday, classic computer collector Joe Strosnider announced the availability of a new 3D-printer filament that replicates the iconic “Platinum” color scheme used in classic Macintosh computers from the late 1980s through the 1990s. The PLA filament (PLA is short for polylactic acid) allows hobbyists to 3D-print nostalgic novelties, replacement parts, and accessories that match the original color of vintage Apple computers.

Hobbyists commonly feed this type of filament into commercial desktop 3D printers, which heat the plastic and extrude it in a computer-controlled way to fabricate new plastic parts.

The Platinum color, which Apple used in its desktop and portable computer lines starting with the Apple IIgs in 1986, has become synonymous with a distinctive era of classic Macintosh aesthetic. Over time, original Macintosh plastics have become brittle and discolored with age, so matching the “original” color can be a somewhat challenging and subjective experience.

A close-up of “Retro Platinum” PLA filament by Polar Filament. Credit: Polar Filament

Strosnider, who runs a website about his extensive vintage computer collection in Ohio, worked for years to color-match the distinctive beige-gray hue of the Macintosh Platinum scheme, resulting in a spool of hobby-ready plastic by Polar Filament and priced at $21.99 per kilogram.

According to a forum post, Strosnider paid approximately $900 to develop the color and purchase an initial 25-kilogram supply of the filament. Rather than keeping the formulation proprietary, he arranged for Polar Filament to make the color publicly available.

“I paid them a fee to color match the speaker box from inside my Mac Color Classic,” Strosnider wrote in a Tinkerdifferent forum post on Tuesday. “In exchange, I asked them to release the color to the public so anyone can use it.”

Endangered classic Mac plastic color returns as 3D-printer filament Read More »