Anthropic today released Claude Sonnet 5, a new AI model that the company says delivers near-flagship performance at mid-tier prices — a move designed to give cost-conscious enterprise developers access to powerful agentic capabilities just as the San Francisco-based AI lab barrels toward an initial public offering that will test whether the private market’s staggering AI valuations can survive public scrutiny.
The release, which Anthropic describes as “the most agentic Sonnet model yet,” makes Sonnet 5 the default model for users on Anthropic’s Free and Pro plans, while also making it available to Max, Team, and Enterprise customers. Introductory API pricing is set at $2 per million input tokens and $10 per million output tokens through August 31, after which it rises to $3 and $15 respectively — still well below the $5 input and $25 output pricing of Anthropic’s top-of-the-line Opus 4.8.
The strategic logic is unmistakable: Anthropic is trying to democratize access to capabilities that until very recently only its most expensive models could deliver, while building the kind of broad-based developer adoption that will look attractive in an S-1 filing.
Sonnet 5 narrowed the gap with Anthropic’s flagship Opus model across five major evaluations, and surpassed it on one. (Source: Anthropic)
Advertisement
Sonnet 5 benchmarks show the mid-tier model closing in on Anthropic’s flagship Opus
Sonnet 5 posts major gains over its predecessor, Sonnet 4.6, across every evaluation Anthropic disclosed. On SWE-bench Pro, an agentic coding benchmark, Sonnet 5 scores 63.2% compared with Sonnet 4.6’s 58.1% — a jump that brings it within striking distance of Opus 4.8’s 69.2%. On Terminal-Bench 2.1, another coding evaluation, the gap narrows further: 80.4% for Sonnet 5 versus 67.0% for Sonnet 4.6 and 82.7% for Opus 4.8.
In multidisciplinary reasoning, as measured by Humanity’s Last Exam, Sonnet 5 scores 43.2% without tools and 57.4% with tools — the latter figure essentially matching Opus 4.8’s 57.9%. On computer use tasks evaluated through OSWorld-Verified, Sonnet 5 reaches 81.2%, up from 78.5%. And on GDPval-AA v2, a knowledge-work benchmark, it scores 1,618 — surpassing Opus 4.8’s 1,615 and far exceeding Sonnet 4.6’s 1,395.
The pattern across these evaluations tells a consistent story: Sonnet 5 doesn’t merely inch forward from its predecessor. It vaults into a performance tier that overlaps substantially with Anthropic’s flagship model, while costing roughly 60% less per token at standard pricing and even less during the introductory period.
Enterprise partners say Sonnet 5’s agentic AI capabilities finish jobs that previous models abandoned
The emphasis on agentic capabilities — the ability to plan, use tools like browsers and terminals, and execute multi-step workflows autonomously — reflects where the AI industry’s center of gravity has shifted in 2026. Enterprises are no longer simply asking chatbots questions; they are deploying AI systems that can navigate complex software environments, execute multi-step coding tasks, and operate with minimal human supervision.
Advertisement
Early access partners painted a picture of a model that doesn’t just start tasks but finishes them. Sualeh Asif, co-founder of Cursor, the AI-powered code editor that has become a bellwether for developer tool adoption, said that “with Claude Sonnet 5, agents stay on plan, follow our conventions, and ship clean multi-step changes, all at an efficient cost.” Daniel Shepard, a senior engineer at Zapier, described handing the model a two-part automation job — updating Salesforce account tiers and sending a launch announcement — that “used to stall halfway” with previous models but now completes end to end.
These testimonials matter because they describe exactly the kind of reliability gap that has kept many enterprises from moving agentic AI from pilot programs to production deployments. A model that gets 80% of the way through a complex task before stalling creates more problems than it solves; one that reliably completes the full workflow changes the economics of automation. Anthropic also introduced cost-performance curves showing that developers can now adjust effort levels across Sonnet 5 and Opus 4.8 to find the optimal balance of cost and accuracy for their specific use case — a granularity that reflects growing sophistication in how enterprises consume AI services.
On computer use tasks, Sonnet 5 neared the accuracy of Opus 4.8 at a significantly lower per-task cost. (Source: Anthropic)
An updated tokenizer boosts Sonnet 5 performance but could quietly raise costs for some workloads
One technical detail buried in the announcement’s footnotes deserves attention: Sonnet 5 uses an updated tokenizer that changes how the model processes text, similar to the change Anthropic introduced with Opus 4.7.
Advertisement
The tradeoff is that the same input can map to roughly 1.0 to 1.35 times as many tokens depending on content type. Anthropic says the introductory pricing is calibrated to make the transition “roughly cost-neutral,” but enterprise customers running high-volume workloads will want to benchmark their specific use cases carefully before assuming their bills won’t change.
Anthropic says Sonnet 5 is safer than its predecessor, but its most capable models still lead on alignment
Anthropic’s safety disclosures reveal a nuanced picture. The company reports that Sonnet 5 shows lower rates of hallucination and sycophancy than Sonnet 4.6, is better at refusing malicious requests, and is more resistant to prompt injection attacks in agentic contexts. On Anthropic’s automated behavioral audit — which tests for a wide range of misaligned behaviors including cooperation with misuse and deception — Sonnet 5 scored lower (meaning safer) overall than Sonnet 4.6.
However, Sonnet 5 showed “somewhat higher rates of misaligned behavior” compared with the more capable Opus 4.8 and Anthropic’s Claude Mythos Preview, the company’s powerful but tightly restricted cybersecurity-focused model. On a Firefox 147 exploit development evaluation created in collaboration with Mozilla, neither Sonnet model could develop a working exploit — both scored 0.0% — though Sonnet 5 showed a slightly higher partial success rate (13.2%) than Sonnet 4.6 (8.8%). Both remain far below Opus 4.8 (68.8% working exploits) and Mythos 5 (88.4%).
Because of these incremental gains in cyber-adjacent capabilities, Anthropic launched Sonnet 5 with cyber safeguards enabled by default — real-time systems that detect and block dangerous cybersecurity usage. The safeguards mirror those on Opus 4.7 and 4.8 but are less restrictive than those applied to Fable 5, the latest Mythos-class model that Bloomberg reported on June 10 is “blocked from responding to queries related to cybersecurity and biology.” Organizations enrolled in Anthropic’s Cyber Verification Program automatically receive the same access on Sonnet 5 without needing to reapply.
Advertisement
Neither Sonnet model produced a working exploit for a Firefox vulnerability, while Mythos 5 succeeded nearly 90 percent of the time. (Source: Anthropic)
From $14 billion to $47 billion in revenue: Sonnet 5 arrives as Anthropic’s IPO narrative takes shape
The Sonnet 5 launch arrives at what may be the most consequential moment in Anthropic’s short history. The company confidentially filed its IPO prospectus with the SEC in early June, setting up what CNBC has described as “the most scrutinized public offering in tech history.”
The financial trajectory has been extraordinary. In February, Anthropic raised $30 billion at a $380 billion valuation, with the company reporting $14 billion in annualized revenue that had “grown more than tenfold in each of the past three years,” as The Guardian reported.
By late May, Anthropic had closed a $65 billion Series H round at a $965 billion post-money valuation — co-led by Altimeter Capital, Sequoia Capital, and others — with a revenue run rate that had crossed $47 billion. Harrison Rolfes, an analyst at PitchBook, told CNBC that the number that will “either validate or collapse the entire narrative the private markets have been pricing for three years” won’t be the valuation or revenue, but gross margin — a figure no outside observer has yet seen.
Advertisement
In this context, Sonnet 5 serves a dual purpose. For developers, it offers genuine capability improvements at competitive prices. For Anthropic’s IPO narrative, it demonstrates the company can deliver a compelling product at a price tier that could drive the kind of broad adoption Wall Street rewards — high-volume, recurring API revenue from thousands of enterprise customers.
Government deals and growing competition define the market Sonnet 5 enters
The timing also aligns with Anthropic’s aggressive push into institutional contracts. Just yesterday, California Governor Gavin Newsom announced a first-of-its-kind partnership providing Claude to all state agencies at a 50% discount, with free workforce training.
Kate Jensen, Anthropic’s Head of Americas, called it an effort to “put Claude to work for the people who keep this state running.” The deal — which extends to California’s cities and counties — represents exactly the kind of durable, recurring adoption that could anchor revenue well beyond the developer community.
But Anthropic’s release lands in an increasingly crowded field. OpenAI, which raised a $122 billion round in March at an $852 billion valuation, is pursuing its own IPO. Elon Musk’s SpaceX, which merged with xAI, priced its IPO at $135 per share with a $1.77 trillion valuation. Google, Meta, and a growing wave of well-funded competitors — including Asian AI startups that, as the Wall Street Journal has reported, are developing Mythos-like cybersecurity capabilities — are all vying for the same enterprise market.
Advertisement
Gil Luria, head of technology research at D.A. Davidson, told CNBC that while Anthropic “appears to have the lead” in frontier AI models, “much of their current usage is for trials and experimentation and that may not sustain.” That observation cuts to the heart of the challenge facing every frontier AI lab: converting experimental developer usage into durable, production-grade revenue.
Anthropic’s more capable models showed lower rates of misaligned behavior than Sonnet 5, which nonetheless improved markedly over its predecessor. (Source: Anthropic)
The real test for Sonnet 5 isn’t benchmarks — it’s whether cheaper AI can sustain a trillion-dollar story
Sonnet 5’s positioning — offering near-Opus performance at Sonnet prices — is a direct play for that conversion. Enterprise customers experimenting with expensive Opus-class models may find that Sonnet 5 delivers sufficient quality for production workloads at a price point that finance teams can approve at scale. If it works, it could accelerate the shift from experimentation to deployment that every AI company needs to justify its valuation.
Three things will determine whether Sonnet 5 matters beyond the initial benchmark charts. Real-world agentic reliability is the first: benchmarks measure capability, but production deployments measure consistency, and the true test will come when thousands of developers push the model through messy, unpredictable workflows at scale.
Advertisement
The tokenizer economics are the second: the updated tokenizer’s 1.0 to 1.35x token expansion could quietly erode the pricing advantage for certain workloads, and enterprise customers should run their own cost analyses rather than relying on headline per-token prices. The third is the IPO narrative itself: when Anthropic’s S-1 eventually becomes public, investors will scrutinize whether the Sonnet tier — cheaper but high-volume — or the Opus tier — expensive but high-margin — drives the bulk of revenue and, critically, gross profit.
As PitchBook’s Rolfes told CNBC, the 2026 IPO window “either becomes the most consequential IPO cycle since the dot-com era or the most expensive lesson in narrative-versus-fundamentals that public markets have ever taught.”
Anthropic is betting that a model good enough to rival its flagship and cheap enough to run at scale is the product that closes the gap between those two outcomes. The public markets will soon decide whether they agree.
At a conference last year, Ford CEO Jim Farley said that artificial intelligence is “going to replace literally half of all white-collar workers in the US.” Just last week, Ford executives said that the automaker had quietly rehired more than 350 of what it internally calls “gray beard” engineers over the past three years to help fix the AI quality-control systems that weren’t getting the job done.
Over the last decade, US automakers have cut more than 20,000 jobs, nearly a 20% reduction in workforce between Ford, General Motors and Stellantis combined. While Ford hasn’t said for sure how many of these gray beard rehires were originally fired to make way for AI and how many are simply returning retirees, Farley’s recent statements on automation-fueled worker replacement certainly paint an awkward picture.
Advertisement
Representatives for Ford and the United Auto Workers union did not immediately respond to requests for comment.
Not getting the desired results
“Artificial intelligence is a fantastic tool, but it’s only as good as the information you use to train it,” Charles Poon, Ford’s vice-president of vehicle hardware engineering, told reporters last week. “Mistakenly, we thought that by just introducing artificial intelligence and ingesting the design requirements that we had, that would produce a high-quality product.”
Kumar Galhotra, Ford chief operating officer, was even more blunt about the realities of AI in manufacturing, saying that Ford had been “relying more and more on automated quality systems and not getting the desired results.”
More than a simple oopsie, automation issues have been costing Ford billions in warranty costs and recalls. A study from iSeeCars, an automotive marketplace and research company, ranked recent Ford models among the most recalled vehicles in the industry. Ford’s statements and the rehiring of experienced workers are essentially an admission that moving too quickly into AI was a big mistake.
Advertisement
Many major corporations in almost every aspect of tech and manufacturing have been naming artificial intelligence as an excuse for large workforce reductions, often without fully accounting for what gets lost when that human factor walks out the door. Entire industries have been crunching the uncomfortable numbers of replacing human judgment with automated systems, with some even backtracking on their decisions when the true cost of AI proves too high.
Ford CEO Jim Farley has spoken frankly about how AI tech will lead to a drastic reduction in white-collar jobs.
Ford
What happens now?
Last week, Ford announced that, for the first time in 16 years, it had captured the number one spot among mainstream brands in JD Power’s 2026 Initial Quality Survey, up from tenth last year. The automaker credits the rise, in part, to the contributions of the rehired gray beards. But before you get too excited about the triumph of these modern-day John Henrys over the machines set out to replace them, don’t forget what ultimately happened to that folklore hero: He was still replaced by the steam engine.
Advertisement
Galhotra said the rehired specialists — some former Ford employees, others drawn from industry suppliers — were brought back specifically to “hunt for failure points before a part ever reaches the plant floor.”
Ford isn’t abandoning AI. Instead, the returning gray beards are doing two things: training younger staff who never worked alongside those veterans and helping to rebuild the data pipelines that the AI tools run on.
Essentially, they’ve been brought back to fix and train the automated software systems that replaced them. Ford also said it has built a dedicated 40-person software quality assurance team and added more than 100,000 AI-powered automated tests to catch edge cases late in development.
Technology marches on.
Advertisement
Ford just happened to learn the lesson loudly enough to become a case study, but I don’t think it will be the last. There may not always be gray beards to call on to save the day.
The best time to plant a tree was 20 years ago, which was probably also the best time to stock up on RAM and storage. The massive demand from AI giants has sent memory and storage prices through the roof, and it’s no longer just gamers eyeing new graphics cards who are feeling the pinch. Following a recent price hike, many Apple products have become more expensive. Devices that now start at higher price points include the iPad, MacBook, iMac, Mac Studio, Apple TV, Apple Vision Pro, and HomePod. While some products like the HomePod mini have become $30 more expensive, fully configured Mac Studio variants have seen staggering price increases of up to $4,200.
Surprisingly, Apple’s best-selling product, the iPhone, hasn’t seen a price hike — at least not yet. The base model iPhone 17, which is equipped with 8GB of RAM, starts at $800, and the budget-oriented iPhone 17e is priced at $600. The more expensive iPhone 17 Pro and iPhone 17 Pro Max devices start at $1,100 and $1,200, respectively. It’s also good to know that all current-gen iPhone models ship with a base capacity of 256GB, which is a decent amount of storage.
Apple Watch models and AirPods headphones and earphones are also currently unaffected. Of course, accessories like AirTags or the Apple Pencil that do not rely on large amounts of RAM also haven’t seen any price changes. Sadly, though, refurbished Macs and iPads now cost more.
Advertisement
Newer iPhones may cost a lot more
Enkhtulga Khandsuren/Shutterstock
Although the upcoming iPhone 18 Pro series is expected to have the same RAM capacity as current-gen Pro models, a price hike may be imminent. Information shared by an IDC analyst to Tom’s Guide suggests that Apple might increase the price of upcoming iPhone models by as much as $200. A starting price of $1,300 for the iPhone 18 Pro and $1,400 for the larger iPhone 18 Pro Max would be a tough pill to swallow. Reports initially pointed to a $50 or $100 price increase, but given how much Apple has bumped up the prices of its Macs, a $200 increase doesn’t seem as far-fetched.
Apple is also rumored to launch its first foldable smartphone, the iPhone Fold, alongside the iPhone 18 Pro series later this year. As if the sophisticated engineering behind a foldable wasn’t already expected to drive up costs, the current situation with memory prices skyrocketing is more than likely to push prices even higher. Analyst Ming-Chi Kuo speculates the iPhone Fold may cost as much as $2,500.
Advertisement
If you’re looking for a solid smartphone from Apple, now’s the time to grab an iPhone. Although Siri AI will be available for all current-gen models with iOS 27, the expensive iPhone 17 Pro and iPhone Air devices that ship with 12GB of RAM get access to a more powerful on-device AI model.
Wenting Zhang looked at the M5PaperS3 and decided its e-ink screen could do more than display static pages. The compact development board from M5Stack carries an ESP32-S3 processor, a 4.7-inch 960-by-540 e-ink touchscreen, a simple buzzer, and a microSD slot. He turned the whole package into a handheld that runs original Game Boy software at a steady 60 frames per second.
Most e-ink displays are noted for being a little slow. It has a lot to do with how charged particles in the panel take their time drifting into place, thus a full refresh can take hundreds of milliseconds and frequently produces a faint afterimage. Zhang stomped on those physical constraints. He did not simply accept the status quo and dismiss those restrictions as the maximum. Instead, he created a bespoke driver that only works with the display’s low-level parallel interface. This driver holds a super-compact state record for each pixel, with only four bits per dot.
That state buffer receives a new set of image data every sixtieth of a second, while the driver nudges each pixel into place using a series of voltage changes. It’s a clever approach that renders the standard global lock, which requires the entire panel to complete one full cycle before receiving fresh commands, completely obsolete. The Game Boy panels are small (160 by 144 pixels), therefore a crisp threefold upscale with slight dithering is well within the processor’s capabilities. Fortunately, all of the active buffers fit inside the ESP32-S3’s extremely fast internal SRAM.
The two cores on the board function together, with one running the emulator itself. The second core is responsible for processing display updates and sending new data via DMA transfers timed to the vertical sync signal. This preserves the 60 Hz rhythm perfectly on beat even when the emulation workload changes. Zhang tried several Game Boy emulator cores before settling on CrankBoy, an optimized fork of Peanut GB. It provided the best trade-off between speed and compatibility on this hardware. Most games run close to full speed. The system will skip the odd frame to keep the timing just perfect for sound and input in the event of a demanding situation. Unfortunately, Game Boy Color titles remain out of reach for the time being due to their roughly doubled processing demand.
The sound emanates from a single buzzer on the board. Zhang created a rapid-switching scheme that cycles between crude approximations of the four original Game Boy audio channels. The end output is recognized chiptune music rather than exact waveform playback, but it preserves the essence of the games without the need for additional gear. The touchscreen includes all of the controls. The panel’s lower portion displays a Game Boy-style directional pad and various action buttons. Any taps register right away. There is experimental Bluetooth LE gamepad support, although it only works with a limited number of controllers right now.
Saved data is stored on the microSD card, and the hardware power button only cuts power, with no shutdown process (Zhang added a manual save button to the screen). Quick save states add an extra degree of security during lengthy gaming sessions. The entire project’s source code (known as Paper Boy S3 in some places) is now available on GitLab. There’s also a separate proof-of-concept JIT recompiler that another developer worked on, demonstrating one approach to go even faster if someone wants to take it a step further. [Source]
T-Mobile appears to be migrating its 303,000-core VMware environment to another platform while fighting Broadcom in court for the extended support it says its perpetual-license agreement guarantees. “The matter is somewhat urgent,” The Register reports, because a court-ordered support arrangement expires August 3, “so T-Mobile may soon be unable to get support for its very substantial VMware estate.” The Register reports: The dispute relates to a deal T-Mobile struck with VMware in August 2023, which saw the telco acquire perpetual licenses and two years of support for some software, plus the option for a further year of support. When Broadcom acquired VMware in 2023, it stopped selling perpetual licenses and standalone support deals for customers with those licenses. Broadcom also reduced the virtualization giant’s product range from over 150 products to two subscription-only bundles. Broadcom now mostly sells its Cloud Foundation (VCF) private cloud suite. Customers including AT&T and Tesco tried to exercise their right to extended support, but Broadcom declined to do so. AT&T settled on confidential terms. Tesco is pursuing the matter in the courts.
When customers exercise their option for extended support, Broadcom argues it can’t deliver because the products covered by the contract don’t exist anymore, its contracts allow it to deny support for dead products, and subscriptions are now the industry standard. T-Mobile started using VMware’s products in 2008. In one hearing, the carrier’s counsel described T-Mobile’s VMware implementation as “the base of the entire internal network” and “the place where 1,000 applications reside.” Another filing, from Broadcom, says the telco runs VMware software on over 303,000 CPU cores.
Court documents allege that in 2024 Broadcom notified T-Mobile it would not renew support after the initial two-year deal expired in 2025. The two parties kept talking about possible new arrangements. T-Mobile also sought an injunction that would compel Broadcom to provide extended support. Broadcom opposed the injunction, arguing that T-Mobile deliberately waited too long to seek it. At one point T-Mobile suggested a $20 million deal for another two years of support. An affirmation filed last week by T-Mobile vice president of technology Kevin Luu says the carrier sought that arrangement “to be able to complete T-Mobile’s transition away from VMware at a more deliberate pace.”
The court eventually granted the injunction forcing Broadcom to offer support beyond August 2025, but required T-Mobile to pay $5.28 million and post a $500,000 undertaking. Broadcom continued to provide support but also sought damages on grounds that the injunction meant it missed out on a new deal with T-Mobile. The telco has rubbished that argument in part because the two parties were still talking about a new deal. Broadcom later proposed to charge $24 million for extended support covering six products, a sum it said would cover over 20 staff needed to support T-Mobile. The carrier fired back by pointing out that it has made just two support calls in 2026, which hardly justifies such a massive staff and expense.
The 2026 iPhone Photography Awards are in their 19th year of finding the best images captured using iPhone cameras. I’ve collected a few favorites from the winning images, but be sure to view all of the winners at the IPPAwards site.
2 of 14Robyn Jensen/IPPAwards
Grand Prix: Robyn Jensen
Robyn Jenson’s photo of an erupting volcano in Guatemala at night is a challenge for an iPhone’s cameras.
Advertisement
Shot on iPhone 15 Pro, 6.765mm (24mm equiv), f/1.8, 1s, ISO 12500
3 of 14Gellert Gombai/IPPAwards
First Place: Gellert Gombai
This photo by Gellert Gombai was made using an iPhone X, a phone likely older than the two children who are the subjects.
Shot on iPhone X, 4mm (28mm equiv), f/1.8, 1/1500s, ISO 20
4 of 14Arnold Plotnick/IPPAwards
Second Place: Arnold Plotnick
If an iPhone photography competition didn’t include a stark photo of a cat, is it even real? US photographer Arnold Plotnick caught this feline’s steady gaze in Amsterdam.
Advertisement
Shot on iPhone 16 Pro, 6.765mm (29mm equiv), f/1.8, 1/60s, ISO 320
5 of 14Catherine Wang/IPPAwards
Third Place: Catherine Wang
Catherine Wang of the US turned to a long tradition of still-life photography to compose this scene in Virginia.
Shot on iPhone 16 Pro Max, 6.765mm, f/1.8, 1/40s, ISO 250
6 of 14Barry Mayes/IPPAwards
Abstract – First Place: Barry Mayes
UK photographer Barry Mayes must’ve warmed to this scene of intricate frost on a car window.
Advertisement
Shot on iPhone 8 Plus, 3.99mm (28mm equiv), f/1.8, 1/120s, ISO 50
7 of 14Peter Crome/IPPAwards
Animals – First Place: Peter Crome
Good light and good dogs, all the ingredients for this winning photo in the Animals category by UK photographer Peter Crome.
Shot on iPhone 14 Pro, 9mm (77mm equiv), f/2.8, 1/400s, ISO 32
8 of 14Leping Cheng/IPPAwards
Animals – Honorable Mention: Leping Cheng
It’s an easy lesson to forget as a photographer: be sure to look up. This mix of perspective and timing garnered an honorable mention in the Animals category.
Advertisement
Shot on iPhone 12 Pro Max, 5.1mm (26mm equiv), f/1.6, 1/2900s
9 of 14Simona Bonanno/IPPAwards
Animals – Honorable Mention: Simona Bonanno
Photography can be as much about concept as it is about capturing a moment. The tuft of mane on this white horse fits with the fluffy clouds in the background.
Shot on iPhone 15 Pro, 6.765mm (24mm equiv), f/1.8, 1/12000s
10 of 14Krystal Rountree/IPPAwards
Children – First Place: Krystal Rountree
US photographer Krystal Rountree took first place in the Children category with this slice of awareness that a wave is coming soon.
Advertisement
Shot on iPhone 15, 5.96mm (26mm equiv), f/1.6, 1/2500s, ISO 50
11 of 14Iryna Nemyrovych/IPPAwards
Children – Honorable Mention: Iryna Nemyrovych
This young boy playing in the water is perfectly framed by the arch of the tree behind him. His light skin contrasting with his dark surroundings draws even more attention.
Shot on iPhone 15 Pro, 6.765mm (24mm equiv), f/1.8, 1/4000s
12 of 14Kęstutis Cemnolonskis/IPPAwards
Nature – Honorable Mention: Kęstutis Cemnolonskis
Photography is so often a matter of knowing where the light is and hoping you get lucky. It’s not clear if photographer Kęstutis Cemnolonskis knew the sun would illuminate this break in the grove of trees or if it was an accident, but the capture invites you to wonder.
Advertisement
Shot on iPhone 12 Pro Max, 7.5mm, f/2.2, 1/850s
13 of 14Shan Qin/IPPAwards
Other – Second Place: Shan Qin
Looking almost like the setup for a Wes Anderson film, this composition by Shan Qin evokes a time when train travel was more elegant (or kitschy).
Shot on iPhone X, f/1.8, 1/950s
14 of 14Carlos Rubin/IPPAwards
Portrait – Second Place: Carlos Rubin
Puerto Rican photographer Carlos Rubin took advantage of this woman’s contrasting blue swimwear and orange cap to make a portrait that goes beyond a snapshot.
Advertisement
Shot on iPhone 12 Pro, 6mm (52mm equiv), f/2, 1/125s, ISO 25
The first known use of humans using wind to perform mechanical work with machines dates back to ninth-century Persian windmills. But if we count sailing vessels among those machines, the history goes back to sometime just before the invention of written language. Since then, humans have been sailing everything from the tiniest of Sunfish to the largest of shipping vessels, and even sailing boats like canoes that aren’t typically designed for efficient sailing. For those who already own a canoe, the conversions can be straightforward but often involve drilling into the hull. This homemade conversion kit, on the other hand, requires no drilling at all.
The first, and most obvious, part of the conversion is to add a mast and sail. [Tea]’s primary setup does involve drilling a mast thwart into the gunwales of the canoe, but he also built an alternative setup which clamps to the gunwales and the bow deck instead. The standing lug sail is then hoisted on an unstayed wooden mast. The next major component of the build are a pair of leeboards which also clamp to the gunwales and function like a centerboard, and can be adjusted for one’s preferred amount of weather helm. Rounding out the stern of the boat is a custom-built rudder with a pair of lines in lieu of a tiller which can be positioned anywhere along the length of the boat.
All of the wooden parts of this build were custom-built from common lumber with finishing touches from a router to soften all of the hard edges. Canoe sailing is fairly popular, although without the leeboards these common sailing kits are often meant for downwind sailing only. A complete setup like this turns it into a much more capable craft. Without a canoe as a base vessel to start with, though, a complete sailing vessel can be built from common lumber as well.
T-Mobile is asking a New York court to rule that Broadcom was contractually obligated to continue supporting its VMware perpetual licenses.
In its complaint, T-Mobile said it has tens of thousands of virtual machines using VMware software across approximately 303,140 CPU cores. It also said that it was migrating off VMware but noted the time-consuming and technical challenges involved in migrating over 1,000 applications.
It filed its lawsuit, which was first reported by The Register today, in the Supreme Court of the State of New York in August 2025 (PDF).
The mobile company claimed that in 2023, it bought perpetual VMware licenses, plus two years of support with the option to buy a third year. But after Broadcom bought VMware, it stopped sales of VMware perpetual licenses in favor of subscriptions and started bundling VMware products into a few, more expensive bundles.
Advertisement
When T-Mobile tried to extend support for a third year for $5,288,398.45, Broadcom wouldn’t allow it, per an August 2025 filing from T-Mobile. A Broadcom representative reportedly told T-Mobile via email: “Broadcom announced end of available of all perpetual products, which includes Stated Out Year Renewals for perpetual support.”
A judge granted T-Mobile an injunction that allowed it to receive support services from October 2025 through August 3, 2026, for $5.28 million, plus the posting of a $500,000 undertaking.
Now, T-Mobile seeks a declaration that it was entitled to renew support services and further relief as the court deems necessary.
Shenzhen hosted the June 30 event where UBTECH introduced its UWORLD U1 series to the world. The company presented these full-size machines as the first humanoid robots of their kind built for mass production and everyday consumer use rather than factory work alone.
Designers went to great measures to make the robots appear realistic by giving them silicone skin with all the proper characteristics, such as pores, veins, and fingerprints. Guys stand 183cm tall and weigh 42kg, while girls are slightly smaller at 168cm and 35.2kg. Despite their realistic sizes, they do not appear overly large or out of place in a household environment, which has to be a significant plus.
Three models, one lightweight platform R1 Air (20 DOF, monocular camera), R1 (26 DOF, binocular camera, head+waist joints), and R1 Edu (26 DOF…
Easy setup – no coding required for basic use Unbox, power on, and start. Manual teaching feature: physically pose the robot, and it replays the…
More DOF = more expressive movement 26‑DOF models (R1 / R1 Edu) add head and waist articulation for smoother dance and running. For safety reasons…
Engineers were able to fit 88 degrees of freedom into each bot utilizing servo joints and a clever proprietary neck design, allowing the robots to mimic human movements much more effortlessly. They can sit, lean, lie back, and give a gentle hug. Everything you would expect a human to do. In early displays, it was really fascinating to see robots dancing with their human partners, because the flawless transitions between upper body and leg movements made them appear to be in sync.
Of course, all that great technology would be useless without some serious software to back it up. So UBTECH went to considerable pains to create an emotion-aware language model that can detect a wide range of subtle clues, including facial emotions, body posture, voice tone, and even how you speak in different contexts. In tests, it detected approximately 20 different emotional states with an accuracy of more than 90%. The system is divided into two parts: a super fast local response that provides answers in less than 500 milliseconds and a deeper level of reasoning that kicks in when more thoughtful responses are required. To make it all feel more natural, the lip movements are perfectly in sync with the speech, so there is no lag that can disrupt the flow of conversation.
Advertisement
When it comes to memory, the robot records all of its interactions in encrypted form on a local device, similar to an agent-based operating system that gradually accumulates a picture of your everyday routines and preferences. The majority of the processing occurs on the device itself, with a Rockchip RK3588 chip doing the most of the work, but there are three layers of privacy protection built in to ensure your data remains secure. It’s quite fantastic stuff. The best aspect is that you don’t need to wake it up with a specific word; simply start chatting and it will begin reacting, as well as maintaining eye contact. The battery life is also good, with 2 to 4 hours of vigorous use on a single charge. If you want to take things to the next level, you may connect it to the internet and gain access to more complex capabilities.
As you’d imagine, there are several variants to choose from, so you may find one that fits your budget and needs. The Lite model is the basic level, focusing solely on the upper body, and it is reasonably priced at 119,800 RMB ($17,632). The Pro model up the ante to a full body, allowing you to move around more, whereas the Ultra model is the top of the line, with dynamic movement, tons of more power, and a plethora of customizing possibilities. That one costs a whopping 990,000 RMB ($145,707) for the males and 880,000 RMB ($129,517) for the females. You can pre-order it today with a 3,000 RMB ($441) deposit, and more than 13,000 individuals have done so since pre-orders began. It’s important to note that sales are only available to those above the age of 18. UBTECH hopes to start shipping the robots later in 2026.
Next spring could be a fiesta of new iPads, MacBooks and iPhones.
Devindra Hardawar for Engadget
Mark Gurman is back with a new report in Bloomberg about how Apple’s device lineup may be evolving over the next 12 months. Sources told the reporter that the company aims to present an overhaul of the baseline MacBook Pro in the first half of 2027. The 14-inch entry-level laptop will reportedly sport a new design that aligns with the look of the higher-end computers likely to be announced starting in the fall. Gurman suggests that lineup will include Apple’s first touchscreen MacBook, which had previously been rumored for the M6 laptop generation.
In addition to the new entry-level laptop, Apple is reportedly testing out four new iPad Pro models. Although specifics of the new tablets were not shared, sources suggested that the next round of iPads would focus on features for improving performance while retaining the current size options of 11 inches and 13 inches.
Spring is becoming the time when Apple makes more announcements for its entry-level and budget products, and even sometimes pulls off the occasional surprise. In spring 2026, we got the MacBook Neo alongside a new iPhone, iPads and MacBook Pros over the course of a week. Even without the potential addition of five products, spring 2027 was looking to be similarly chock full of news from Apple. On the smartphone side, we were already expecting to hear about the base model of the iPhone 18 and an update to the iPhone Air. It could also be when Apple breaks from its Pro and Max tradition and skips straight to M7 silicon.
Safer, cheaper, and nothing to do with cybersecurity
Anthropic has released the latest version of its mid-sized model, Sonnet 5, which the company claims is its most “agentic” yet.
For developers writing agents to automate tedious and recurring tasks, Sonnet 5 promises improved capabilities in reasoning, tool use, coding, and knowledge work. This version is also less likely to pull embarrassing (for Anthropic) gaffes of misunderstanding, so the company asserts.
Advertisement
“Our safety assessments found that Sonnet 5 shows an overall lower rate of undesirable behaviors than Sonnet 4.6, and is generally safer to use in agentic contexts,” the company asserted in an introductory blog post on Tuesday.
Sonnet 5 is smarter at refusing malicious requests and resisting prompt-injection attempts. It doesn’t hallucinate as often and doesn’t suck up to the user so much (“sycophancy”) as did its older brown-nosing Sonnet 4.6 sibling. It is also more aware of, and can block, user misuse and deception, the benchmarks in Anthropic’s System Card seem to indicate.
Sonnet is the default model for Claude Free and Pro users, and is also available to the token-pinching Max, Team, and Enterprise customers.
The benchmarks also indicate Sonnet 5’s performance can come close to that of Anthropic’s flagship enterprise-focused Opus 4.8, but can execute the same tasks more cost effectively. For Opus, Anthropic charges $5 per million input tokens and $25 per million output tokens.
Advertisement
Starting in September, Sonnet users will pay $3 per million input tokens and $15 per million output tokens, though Anthropic is running a special through the end of August where tokens will only be $2 per million inputs and $10 per million outputs.
So users trimming their token budgets can run jobs through Sonnet instead of Opus, the company suggests.
The 5.0 release offers a new setting to adjust the model’s effort at completing tasks. Simple tasks can be completed through one of the lower “effort” settings, which uses fewer tokens, while longer-running agent-based tasks can go full throttle (“xhigh” or even Homer Simpson’s favorite setting, “max”).
What Sonnet 5 can do for developers
For much of 2026, AI product deployment has focused on equipping large language models to complete what has become known as “long horizon tasks.” It might be easy for a model to fix a bug or churn out some code. However, keeping its finicky attention fixed on a multi-part task has proven more difficult.
Advertisement
The new version of Sonnet can go the distance, according to the company, compared with the earlier Sonnets.
“Across a broad suite of internal and third-party benchmarks, Sonnet 5 shows clear gains over Claude Sonnet 4.6 in coding, agentic search, multimodal reasoning, and professional-task performance,” the System Card asserted.
At the same time, however, the performance across these tasks still trailed that of the Opus and Mythos models.
One testimonial from a Zapier engineer described a two-part job that flummoxed earlier Sonnets: Update a contact database and send out a notice to all users. Version 5 was able to complete the task “end to end.”
Advertisement
Cybersecurity: Nothing to see here
The San Francisco-based company also went out of its way not to attract any more undue attention from Washington, DC policymakers.
“We did not deliberately train Sonnet 5 on cybersecurity tasks,” the company asserted.
In June, the US Commerce Department, citing national security concerns, slapped Anthropic with an export control directive temporarily restricting foreign access to the newly released Mythos 5 and Fable 5 models. Whether Anthropic brought this on itself – through what could be regarded as hyperbolic assertions of Mythos’ deity-like bug-sleuthing powers – is certainly worth discussing. But Anthropic, like Pete Townshend, certainly won’t be fooled again.
While it can readily perform routine cybersecurity tasks, Sonnet 5 is guardrailed against generating offensive attack code. When commanded to write a Firefox exploit, it failed to complete the task (though it got a bit further than Sonnet 4.6 in the attempt).
Advertisement
“This latter change is likely due to improvements in general intelligence rather than specific training,” the company’s blog post noted. ®
You must be logged in to post a comment Login