Apple’s 1TB M5 13-inch MacBook Air is down to $1,399 at Amazon thanks to a $200 month-end discount.

The lowest price since Prime Day is available now on Apple’s M5 13-inch MacBook Air with an upgrade to a 1TB SSD. The $200 discount is hosted by Amazon, bringing the cost of the Sky Blue model with a 10-core GPU and 16GB of unified RAM down from $1,599 to $1,399.

Buy 1TB MacBook Air for $1,399

According to our 13-inch MacBook Air M5 Price Guide, this is the steepest discount across the entire product line.

Prime members can get delivery as soon as tomorrow, so you won’t have to wait long to use your new laptop.

Lake Ontario is no more on Google Maps in the US. Google changed the name of easternmost of the Great Lakes to “Lake America” late Saturday, days after President Donald Trump ordered its rebranding in official government databases.

Trump issued the executive order on Thursday as part of an ongoing feud over tariffs with Canada, whose capital province is called Ontario. It followed a similar order he issued in January 2025 to rename the Gulf of Mexico as the Gulf of America as part of a fight with the US trading partner to the south.

Google, Apple, Microsoft, and other maps providers adopted the Gulf of America naming as soon as the US Department of the Interior updated its records, which came a couple of weeks after Trump’s order. So it was no surprise that when those records—which are used to determine names on roadway signs and in government literature—reflected Lake America, digital maps from Google also quickly took on the new name.

Apple Maps and Bing Maps have not yet reflected the change but are expected to. In fact, the map for the federal government’s Geographic Names Information System database, which informs these changes, still showed Lake Ontario as of Sunday morning, with a note that the maps are “in the process of being updated to reflect this name change.”

While Google and other private organizations that provide maps are not bound by law to use official names, they generally follow government databases like GNIS. The location names a user sees depends on their own geography.

“People using Maps in the US will see ‘Lake America,’ those in Canada will continue to see ‘Lake Ontario,’ and those outside of the US and Canada will see both names,” Google said in an unsigned statement. “These updates follow our long-standing policy for bodies of water with names that vary from country to country, and are starting to roll out now.”

Trump’s efforts to rewrite maps have generated considerable public debate in open-source mapping communities such as OpenStreetMap, whose technology is used by companies such as Uber and Lyft. Ultimately, OpenStreetMap recognizes both the US and foreign names and gives organizations using its maps the choice over which to show.

One map provider has said it would decline to make the change is MapQuest, which said in a social media post in response to Trump’s order that “we’re not changing it.”

What free gifts does Xbox give for birthdays?

Screenshot by Dave Meikleham (Microsoft)

How to get free gifts from Xbox Rewards on your birthday

Microsoft

As RAM prices surge and RAMageddon headlines multiply, Google posted new requirements for Android app developers earlier this week. Starting in 2027, apps distributed via Google Play must work within strict memory thresholds to “help protect the overall user experience from applications using excess memory and causing system-wide slowdowns.” 

The idea is to prevent a single misbehaving app from glitching the device, hogging RAM and degrading your entire smartphone experience. Not surprisingly, Google explicitly cited the supply shortage in memory — now a scarce, expensive resource amid booming AI demand — as one of the motivations behind the new rules. 

And these are rules, not just suggestions. Google threatens that apps that don’t comply “will be slowed down and may be terminated.” Compared with a lot of past updates, it also seems a pretty tight time frame for compliance. 

Far from punishing innovation, these rules codify what responsible development should look like — efficient memory use, faster launches and smoother multitasking experiences — and once they’re in the codebase, they might hang around for years. That means our phones may no longer slow down with every OS update. 

So, for everyday Android users, that’s a win: Not only could everything run better on today’s lower-end phones and computers, but it might also, hopefully, prolong the life of older hardware by slowing the upward climb in system requirements for upgraded software. And anything that runs well on low-end hardware generally runs faster and more smoothly on the best devices.

Who has the headache now?

Operating system and application developers have gotten accustomed to prioritizing what I think of as the most expensive common denominator — showy, resource-hogging features that highlight the high-margin devices manufacturers most want to sell, with lots of memory, storage and powerful processors. 

But the component supply crunch and skyrocketing prices have suddenly made low-end configurations of phones and computers critical to sustaining sales in those contracting device categories. 

The crunch is really for anything created in a semiconductor fab, though the effects are most visible in memory supplies and prices. Fab resources have been, or are expected to be, reallocated toward whatever’s ultimately most profitable. 

Right now, that means anything critical to AI, especially data centers. Memory producers and packagers can make a lot more money from high-bandwidth memory and SRAM for advanced processing rather than from the DDR incorporated into consumer devices. 

So there’s a lot more incentive for developers to optimize operating systems and application performance for less memory and lower-power hardware. Even if the memory shortage eases in 2027 (though that’s unlikely), the benefits of making our smartphones less sluggish should continue for some time.

RAM shortages affect Windows, too

Then there’s Windows. 

I can’t remember when it last ran decently with 8GB of RAM (and I’d love to know in what universe it can run on 4GB, the minimum requirement on paper). In March 2026, Microsoft finally announced it was taking Windows 11’s poor memory management seriously, “lowering the baseline memory footprint for Windows.” 

The company’s June update on its progress implied that part of the problem is the current interface framework applications are built with (WinUI 3), which rolled out in 2021, as well as the web components integrated into the OS (that’s why Edge is everywhere). 

After five years, shortages finally made it a Microsoft problem, not just a “you need to upgrade” solution.

It’s not completely fair to accuse the operating systems of resource bloat. MacOS and iOS only have to run on Apple hardware and have a far smaller application ecosystem to worry about, but systems like Android and Windows have to support a wide variety of hardware and an anarchic ecosystem of apps and games. 

That means a lot more overhead, including resources that must remain in memory to keep everything communicating. And to date, they’ve had the luxury of being able to back-burner, or at least deprioritize, efforts to make the OSes run more efficiently with less memory. 

Hopefully, no longer. When platforms start treating memory as a scarce resource and push developers to respect that scarcity, the result could be fewer surprise slowdowns and longer lifespans for existing devices. 

The AI irony

One of the reasons for the increase in memory requirements over the past few years has been the AI push. On-device AI features, such as local LLMs, image generation and advanced photo editing, require bursts of memory to run smoothly or to complete tasks with fewer errors. If ordinary apps are maxing out RAM, there’s no room for these features without constant throttling or crashing. 

Another irony? As demand for AI server memory pushes DRAM prices higher, we can’t afford or get the devices that companies have been pushing, and the models have been getting smaller as well.

Manufacturers have responded by incorporating less RAM or even cutting it in new phones. Google’s own Pixel 11 Pro base models, for example, drop from 16GB to 12GB compared to prior generations

Hardware prices might never return to pre-RAMageddon levels. But by squeezing waste out of everyday apps now, Google is carving out space needed for next-generation AI experiences on the same hardware. 

Lori Grunin

Senior Editor / Computers and Gaming Hardware

I’ve been reviewing hardware and software, devising testing methodology and handing out buying advice for more than 25 years.

See full bio

Advertisement

The fintech has joined a growing list of companies and banks pushing for a stake in the global stablecoin market.

Revolut has announced its first stablecoin, a euro-backed token called EURR.

The fintech commenced the phased roll-out of the stablecoin today (26 August), and said it will be first available to a “select group of customers” in Denmark, Poland and Portugal.

A broader launch across the European Economic Area is expected later this year, subject to product, operational and regulatory readiness.

EURR is issued by stablecoin platform Bridge – which was acquired by Stripe in 2024 – and is fully integrated into the Revolut retail app, according to the company.

The stablecoin is designed to maintain a value of €1, and is backed by reserves held and managed by Bridge.

Revolut said that EURR will give eligible customers a “euro-denominated, on-chain rail” to move between euros and crypto, and that the launch marks Revolut’s “next step in becoming a bridge between fiat [government-issued money] and crypto”.

“Revolut initially eliminated hidden fees and friction in currency exchange – now we are doing the exact same thing for crypto,” said Iman Olya, product owner of stablecoin at Revolut.

“EURR completely removes the pain of moving on- and off-chain, becoming a new seamless and instantaneous bridge between fiat and crypto.”

Revolut is the latest company to join the global stablecoin push, following the likes of fintech giants Stripe, Klarna and Visa, and numerous global banks.

While stablecoins have become more prevalent in the international financial landscape, the market is currently majorly denominated in US dollars, with European authorities making a push for euro-backed stablecoins to challenge US dominance.

“The growing argument is that to remain relevant, Europe must respond by promoting euro-denominated stablecoins of its own,” said Christine Lagarde, president of the ECB, at the Banco de España LatAm Economic Forum in May.

“Otherwise, it faces a future of digital dollarisation and a loss of monetary sovereignty.”

Earlier this year, AIB and Bank of Ireland joined a consortium of European banks – called Qivalis – working towards issuing a euro-denominated stablecoin to expand Europe’s financial infrastructure and compete with US-backed versions, which make up the overwhelming majority of stablecoins in circulation.

While stablecoins, once confined purely to crypto trading, have risen in mainstream popularity in recent years, statistics suggest that the popularity of the cryptocurrency may be slowing.

Statistics from DefiLlama show that the combined market value of stablecoins has levelled off in 2026, while data from Visa has shown that usage of the cryptocurrency has dropped this year.

Don’t miss out on the knowledge you need to succeed. Sign up for the Daily Brief, Silicon Republic’s digest of need-to-know sci-tech news.

An overblown new report claims that Apple has been carefully orchestrating executive retirements for years to not spook shareholders. There is absolutely nothing to support the premise.

According to Bloomberg, Cook’s retirement is one of very many exits that are being scheduled to avoid the appearance of a mass exodus that Ternus will have to deal with.

This is similar to Bloomberg‘s previous reporting of multiple people quitting Apple and how the company will struggle without them. So the new report is predicting that Apple will soon be unrecognizable, but of the 11 people on Apple’s leadership page, it only actually expects two more to go.

The two said to be focusing on their 401k are retail head Deirdre O’Brien, who is only 60, and chief hardware officer Johny Srouji, 61.

The argument for O’Brien going is just that she’s rumored to want to.

The argument for Srouji leaving is that Bloomberg once reported he had tried to quit for another role. To be generous, that report was unlikely. As well as relying on that report as established fact, the new report is using it to argue a different case.

Bloomberg originally said that Srouji was quitting to join some other company. Now it’s saying this is evidence that actually he wants to retire.

There were others

The report is specifically about who will be leaving Apple soon, and therefore who John Ternus will have to rush to replace. But it actually makes a stronger case for how Cook’s retirement is practically the end of the changes for now.

There is still Jennifer Bailey, head of Apple Pay, who is retiring in October. But her retirement has already been announced.

Otherwise, it’s claimed that since 2024, Tim Cook has been timing the retirements of key figures so that shareholders were never spooked by mass departures. The claim is that this is why long-standing Chief Financial Officer Luca Maestri stepped down in October 2024, although Maestri didn’t retire and still hasn’t.

Then Chief Operating Officer Jeff Williams did retire, although it was more than a year later in November 2025. Retirement was then announced for both environment head Lisa Jackson and general counsel Kate Adams in December 2025.

Finally, Tim Cook announced in April 2026 that he was stepping down as CEO. Although he also isn’t retiring.

So that’s three people who did retire in the last two years, and a claim that two more will in something like the next few years.

Of the others, it’s presumed that Eddy Cue will be thinking of his succession plan. There is no mention at all of Craig Federighi.

It’s true that there are more Apple executives reaching retirement age now than when Cook became CEO. But the report admits that during Cook’s time, chief designer Jony Ive, hardware head Bob Mansfield, iOS lead Scott Forstall, and not one but two retail chiefs left.

It is also true that Apple is changing and it’s fair to say that the current changes are because of executives reaching retirement age. But the company has always been changing and it has always been replacing key people.

The pace of retirements and departures hasn’t really changed. What has, is the visibility.

Like Tim Cook retiring.

Enterprises trying to feed PDFs, slides and scanned documents into AI pipelines keep running into the same wall: the tools either miss the structure — tables, charts, layout — or cost too much to run at scale.

Cohere released Parse 5 on Thursday, positioning it on price-to-performance, not raw accuracy — the right cost-capability mix for enterprise scale. Parse 5 is a 2.3-billion-parameter vision language model built to convert PDFs, slides and images into structured Markdown at enterprise scale. 

Cohere’s own published benchmark comparison puts Parse 5 behind three larger, general-purpose frontier models on accuracy. GPT-5.5, Opus 4.8 and Gemini 3.5 Flash all score higher than Parse on the three ParseBench dimensions Cohere reports. Cohere is not claiming the top score. It is claiming the best price for a score close to the top.The company priced the model at $1.50 per 1,000 pages through its API, with Model Vault, Cohere’s secure, single-tenant platform for managed inference, available for higher-volume deployment.

“Document parsing isn’t solved because the hard part isn’t reading text, it’s preserving structure and meaning,” Nils Reimers, VP of AI Search at Cohere, told VentureBeat. “Enterprise documents mix tables, diagrams, charts, and formatting that change the interpretation of the data. Most tools still drop structure or hallucinate content, and even frontier models break on layout‑heavy pages.”

Inside the single-pass architecture

Parse 5 takes a page as an image, runs it through a single vision-language model pass and returns structured Markdown, collapsing the OCR-plus-model pipeline most tools run as separate steps.

Architecture. It is a 2.3-billion-parameter vision language model built on Cohere Labs’ North-Micro-Vision-Instruct architecture, with an 8,192-token context window and roughly a 4.6-gigabyte footprint. It accepts a PDF, PowerPoint or JPEG page as a base64-encoded image and returns Markdown in reading order, with tables rendered as HTML, image descriptions and bounding box coordinates for tables and images.

Language coverage. Arabic, English, French, German, Italian, Japanese, Korean, Portuguese and Spanish get stable accuracy, with lower-accuracy zero-shot support elsewhere.

Output modes. The default output returns a Markdown string per page. A blocks mode returns typed elements, where each table carries its own HTML, bounding box and description, the format Cohere positions as what makes citation-level traceability possible for agents.

Availability. Parse 5 is generally available now through the Cohere API, Model Vault, Microsoft Foundry and AWS SageMaker.

The benchmark shows a trade-off, not a win

ParseBench is a benchmark that scores document-parsing tools against human-verified enterprise pages. Cohere reports Parse 5 scoring 79.2 across three dimensions: tables, content faithfulness and semantic formatting. That puts Parse 5 behind GPT-5.5 (84.4), Opus 4.8 (84.3) and Gemini 3.5 Flash (81.8), and ahead of LlamaParse’s Cost Effective tier (78.3), Mistral OCR 4 (74.5), Databricks AI Parse (72.4) and Azure Document Intelligence (69.3).

Cohere’s table notes two excluded dimensions, Layout and Chart, and attributes both to product scope rather than a performance gap. Parse 5 returns reading-order Markdown instead of per-element bounding boxes for text, and describes charts rather than extracting their underlying data, with chart-data extraction planned for a future version.

Reimers said that design choice reflects where agentic workflows actually break.

“For charts, for example, we provide a general description of the chart together with an indicator, how Agentic AI can visually inspect the chart,” Reimers explained. “Other solutions try to extract the data from the chart, but then miss out critical information (for example, the color or the pattern of a line) that leads to hallucinations in Chat and Agentic AI applications.”

Cost is where Cohere makes its real case. Reimers pointed to a workflow the company modeled for a large financial services firm.

“We ran the numbers for a large financial services workflow that processes 750 million documents a year and showed that choosing Parse 5 over a large general‑purpose model like GPT‑5.5 would reduce costs by more than 98 percent.”

That figure is Cohere’s own estimate for a single modeled workflow, not an audited deployment.

Where Parse 5 sits against the field

There is no shortage of options for enterprises looking at parsing solutions.

General-purpose frontier models, GPT-5.5, Opus 4.8 and Gemini 3.5 Flash, top the accuracy comparison but carry the cost and latency of a large model on every page. 

Then there are specialized parsers, including Mistral OCR 4, LlamaParse and open-weight options like Chandra OCR 2 and RedNote’s dots.mocr.

Hyperscaler document intelligence services, AWS Textract, Google Document AI, Azure Document Intelligence and Databricks AI Parse, compete more on ecosystem convenience than on raw parsing quality, and score lowest in Cohere’s own comparison.

Kevin Petrie, VP of Research at BARC US, said document parsing sits at the center of enterprise AI adoption right now.

“We’re completing a survey now that shows document analysis is by far the #1 use case for AI, with 62% adoption rates among organizations we polled,” Petrie told VentureBeat. “Documents and other unstructured objects, including images and so on, hold the proprietary context that organizations need to differentiate their agentic AI initiatives.”

Petrie added that only time will tell how Cohere’s cost-performance stacks up against frontier models, but strategically his view is that Cohere has the right focus. 

Stephanie Walter, Practice Leader for AI Stack at HyperFRAME Research, sees Cohere Parse 5 as sitting in a good spot between legacy OCR and using an expensive frontier model on every page. 

“Its potential advantage is delivering structure, spatial provenance and private deployment at a price suitable for high-volume ingestion,” Walter told VentureBeat. “It does not need to win every benchmark. It needs to make reliable enterprise-scale parsing economical.”

The real test is downstream, not on the benchmark

“Parsing is the first quality gate in the enterprise AI stack,” Walter said. “If tables, headings, images, or reading order are lost during ingestion, better embeddings and larger models cannot recover that missing structure.”

A benchmark score isn’t the only input that matters here. “Enterprises should test parsers against their own most difficult documents and measure downstream retrieval and task accuracy, not how clean the extracted text looks,” Walter said. “The right question is not ‘Did it read the PDF?’ but ‘Can the agent now use the information correctly?’”