Blogroll

You now have a major streaming alternative to The Roku Channel, Pluto TV, and Tubi if you can't get enough free TV—provided you have the right console. Sony has launched Live TV on PS5 in the U.S., promising ad-supported movies and shows so long as you're signed in to your PlayStation account.

USB-C is often touted as the universal standard for a wide range of devices, and while there's certainly more compatibility than there used to be, the reality is a lot more complicated.

Many Americans don't like data centers.

At the very least, a majority of U.S. citizens don't want one of the dozens of proliferating AI infrastructure projects to end up in their backyard, as complaints of environmental destruction, noise pollution, and water constraints reach a fever pitch.

But President Donald Trump doesn't really care. In a recent Truth Social post, the country's leader called data centers a "golden goose" that Americans would be remiss to push back on.

"The only reason that communities throughout the U.S.A. should not want Data Centers is if they want to end up being backwards and poor," the President wrote. "If they want to be successful and rich, with far lower taxes and jobs all over the place, let Data Reign."

SEE ALSO: Trump administration acted illegally in fight against Anthropic, judge rules Trump doubles down on data centers

The inflammatory post echoed similar sentiments shared earlier this month, as well as arguments he made in posts from July. In fact, Trump has steadfastly backed large-scale data center projects, despite his own party backing away from data center commitments on the campaign trail.

"If I were the mayor of a town or the governor of a state, and I had a chance to get a big plant in an AI plant or a data center, I would absolutely want it because the jobs are enormous, and the money paid, the taxes paid are just enormous," said Trump at a recent convening of cryptocurrency leaders at the White House.

An anonymous insider told the Washington Post that Trump's hardline on leaving AI and data centers unfettered by federal regulation is a "personal" subject for him. "You just can’t change his mind about it," the source told the publication.

Behind the scenes, the Trump administration has been edging open the floodgates to Big Tech-backed data centers as part of his mission to establish a nationwide AI infrastructure to rival China.

In fact, Trump has cited the foreign power regularly in arguments for data center investment. "China could not be happier with this anti Data Center movement. Actually, they can’t believe it is happening!" said Trump in the Aug. 31 post.

A recent post by the Global Government Affairs team at Elon Musked-owned X alleged that the platform had uncovered a bot farm network made up of 200,000 accounts posting anti-data center sentiment. The claims appeared to be based on a recent OpenAI investigation that found several bot accounts using OpenAI's tools to produce such content, but the report also indicated these accounts were sharing genuine news reports about energy concerns.

How the federal government is enabling data centers

Trump's Big Tech allies have benefited from his pro-America approach to tech advancement. Last year, Trump signed an executive order fast-tracking federal permitting for AI data centers. The administration has also lobbied for expedited electricity grid connections and pushed for government approvals of new chemicals used for data center manufacturing and facilities.

In an attempt to quell the ensuing community-driven uproar against these projects, the federal government announced a Ratepayer Protection Pledge with several of AI's biggest players already signed on, committing to offset energy grid costs for local communities.

Still, Big Tech's centers have faced mounting legal pressure, including several lawsuits that claim companies like SpacexAI (formerly xAI) and Microsoft violated pollution-permitting laws in the construction of their facilities. Just last week, non-profit newsroom Floodlight accused a Microsoft-backed data center in New Jersey of violating federal law by operating unpermitted gas-powered generators. The center, known as DataOne, is the first in the state to build its own power plant, the Guardian reported. Many centers are shifting to self-sustained power as states crack down on energy grid connections.

SEE ALSO: The NAACP is fighting back against AI data centers

In turn, the federal government has interceded in efforts to enforce environmental law, including levying threats that such attempts pose a "national security threat" for potentially impacting the Department of War's AI capabilities.

Meanwhile, the Environmental Protection Agency (EPA) has been planning to scrap a federal rule requiring public notice and time for comment when applying for such air permits. The move would effectively shift responsibility for public notice to local agencies and could legitimize the legal violations that communities are invoking to fight back, experts say.

Labor Day is right around the corner, and that means we can all look forward to excellent deals on all of our favorite tech, tools, and more. If you need a few more Ryobi battery packs or are gearing up for fall DIY projects, Home Depot has huge discounts you won't want to miss.

Netflix is set to flip its calendar over to September tomorrow, and with it comes a flood of new catalog movies to get into—including the Alien films, the whole Resident Evil pile, and dozens more. If you're a U.S. subscriber to the service, however, you might still be salivating over the 26-minute Grand Theft Auto VI preview or the chilling crime thriller Facing El Chapo.

Netflix is set to flip its calendar over to September tomorrow, and with it comes a flood of new catalog movies to get into—including the Alien films, the whole Resident Evil pile, and dozens more. If you're a U.S. subscriber to the service, however, you might still be salivating over the 26-minute Grand Theft Auto VI preview or the chilling crime thriller Facing El Chapo.

Claude has had SKILL.md files for a while now, and they’re one of the most useful things about it. That said, skills are only as good as the person writing them. If you describe your workflow poorly in a skill, Claude will also do the work poorly — it’s basically garbage-in, garbage-out.

Samsung Galaxy phones have a dirty secret: the official thickness measurements do not include the camera bump. Most smartphones have this same secret, but Samsung’s design choices make it a particularly annoying problem. I set out to fix it.

The Tim Cook era of Apple has come to an end.

Today, August 31, 2026, is Tim Cook's final day as CEO of Apple, and the outgoing Apple chief had some emotional parting words to share on social media.

"Sending lots of love to the Apple community on my last day as CEO," Cook posted on the social media platform X. "My title changes tomorrow, but the love I have for the Apple community never will. Thank you for being a constant source of inspiration. My gratitude is endless, and I’m excited for the next chapter!"

This Tweet is currently unavailable. It might be loading or has been removed.

Tim Cook held the position at Apple for 15 years, taking over as CEO after Steve Jobs departed in August 2011, shortly before his passing the following October. Starting Sept. 1, Cook's new position within the company will be as executive chairman of Apple's board of directors.

Cook will be succeeded by incoming Apple CEO John Ternus, the company's current Senior Vice President of Hardware Engineering. At 50 years old, Ternus is the same age Cook was when he took over as Apple CEO.

• Guess what Apple will release in September in our Big Guessing Game: Apple edition. Enter now for a chance to win an Apple Watch!

Under Cook, Apple saw the launch of some of its most popular product categories, such as AirPods and the Apple Watch. Cook also led the company as it completely relaunched its computer product lines, ranging from MacBooks to desktop Macs, with Apple's own Silicon chipset.

Apple became a trillion-dollar company under Tim Cook's leadership as well.

Cook also had to navigate the company through difficult global geopolitical issues, which led to some interesting moments, such as when President Donald Trump referred to the Apple CEO as "Tim Apple." Cook responded at the time by changing his profile name on then-Twitter to "Tim ."

Ternus's first big Apple event at the helm will take place on Sept. 9, when the company is expected to launch its new line of iPhones, including its first-ever foldable device.

The European Commission announced Sunday that it has formally designated ChatGPT as a Very Large Online Search Engine, and both Reddit and Roblox as Very Large Online Platforms, under the EU's Digital Services Act (DSA).

According to the Commission's press release, all three services reported reaching at least 45 million average monthly users in the EU, clearing the threshold that triggers the designation.

The new classification means that the platforms have until Dec. 31 to meet the DSA's additional requirements for platforms of this size. These include assessing and mitigating risks tied to illegal content, harm to minors, users' physical and mental wellbeing, fundamental rights, elections, and public security, per the Commission.

SEE ALSO: EU says Meta hasn't done enough to prevent minors under 13 from using Instagram and Facebook

The Commission explained its reasoning for each service individually. ChatGPT, described as an AI system capable of responding to prompts and searching the web, qualifies as a "hybrid service" that meets the DSA's definition of an online search engine.

Roblox was described as an immersive gaming and creation platform where users build, publish, and play games made by other users. Reddit was categorized as a discussion-based platform where users create and interact with content inside topic-focused communities. Because both enable the public spread of third-party content, the Commission classified them as online platforms rather than search engines.

With these designations, the Commission gains new investigative authority to examine the functionality of these services and any related systems, and will oversee DSA compliance in partnership with national regulators where each company is legally established in the EU. For ChatGPT and Reddit, it'll be Ireland's Coimisiún na Meán; for Roblox, it's the Netherlands' Authority for Consumers and Markets.

"Reddit is committed to complying with the EU Digital Services Act," a spokesperson for the company told Mashable through email. "We’ve been preparing to meet the VLOP requirements and look forward to coordinating with EU regulators while providing transparency for Reddit users."

All three services were already subject to the DSA's baseline rules for platforms and search engines before Sunday's announcement. The new designation adds them to a growing list of 28 very large online platforms and search engines that have now been designated under the law overall.

Disclosure: Ziff Davis, Mashable’s parent company, in April 2025 filed a lawsuit against OpenAI, alleging it infringed Ziff Davis copyrights in training and operating its AI systems.

Windows is still the most popular desktop wos.While Linux may be obscure on the desktop, there are some prominent users already adopting it. Governments around the world are doing it to be independent or sometimes cut off from Western software. Here are some of the governments building their digital infrastructure around Linux.

Blush Novelties Labor Day Sale: Best dildo deal Blush Novelties Neo Elite Neon Pink 7.5-Inch Dildo $29.99 (save $5) Get Deal Best rabbit vibrator deal Blush Simone Rabbit Vibrator $49.99 (save $28) Get Deal Best butt plug deal Blush Naughty Candy 3.5-inch Butt Plug $14.99 (save 44) Get Deal

If you need a new vibrator (I mean, who doesn't?), there are tons of Labor Day sales going on right now where you can get your hands on some of the best vibes on the market for a fraction of the price. (That's right, Labor Day isn't just for discounted mattresses and kitchen appliances!)

SEE ALSO: I've tested 100+ sex toys. Here are the 15 most mind-blowing toys I've ever owned.

Blush Novelties, an OG brand in the business, just shared that it's offering up to 30% off from Sept. 2 through Sept. 8 (no code required). You can shop brands under the BN name on the official Blush website or via Amazon. I honestly think Amazon is probably the better bet here, since you'll get fast, free shipping with Prime.

Even though the sale hasn't officially kicked off just yet, a handful of toys are already discounted on Amazon and the website. You can browse the site's "sale" section (you'll find up to 50% off select items there now), or hold out until Sept. 2 to see if a better deal comes along.

A quick note before you add anything to your cart: Blush is a legacy brand, so its catalog is just a smidge outdated and includes some older jelly or TPE (porous) toys. Definitely DO NOT buy those. You only want to use body-safe, non-porous materials on, in, and around your intimate bits.

Milwaukee tools are built to take a beating, but caked-on grease, sawdust, and other messes will quietly wreak havoc on your gear. You're probably thinking, "Who cleans their power tools?" but it's pretty common and more important than you think.

Did you know that a number of smart TVs come equipped with cameras? While that idea may raise privacy concerns for many people, and rightfully so, these cameras offer several neat features you can explore.

Best Labor Day 2026 deals at a glance: Best Apple deal Apple iPad Air, 11-inch (M4, WiFi, 128GB) $649 (save $100) Get Deal Best TV deal TCL 65-inch QM8L SQD-Mini-LED QLED 4K TV $1,397.99 (save $1,102) Get Deal Best Amazon device deal Echo Show 11 $189.99 (save $60) Get Deal Best headphones deal Bose QuietComfort Ultra $249 (save $200) Get Deal Best home deal Ruggable save up to 30% sitewide Get Deal Best outdoor deal Ninja FrostVault Cooler (30 quart) $149.99 (save $50) Get Deal Best robot lawn mower deal Segway Navimow X430 Robot Lawn Mower $2,249 (save $250) Get Deal Best kitchen deal Nutribullet Power Core Blender $84.99 (save $15) Get Deal

Summer is coming to an end. While that seems a bit sad, it also means the temperatures cool off and we still have one final long weekend to enjoy before fall kicks in.

Like most long weekends, big retailers are offering special discounts on some of our favorite products. That's great news if you're in the mood for a fall refresh. Consider getting a mattress upgrade if you haven't been sleeping well or browse the REI Labor Day sale to get new outdoor gear to replace anything that's worn out over the summer. It's also a great time to get a new TV ahead of NHL and NFL season.

After Labor Day, we might not see a major sales until Black Friday. In an era when tech prices continue to rise, it could be smart to shop these deals ASAP.

Best Apple deal Opens in a new window Credit: Apple Apple iPad Air, 11-inch (M4, WiFi, 128GB) $649
$749 Save $100   Get Deal Why we like it

The Apple iPad Air with an M4 chip is something to behold. It's a step up in performance level compared to the previous M3, not to mention it's incredibly lightweight and thin. In addition, it's powerful and smart enough to handle intense programs like Final Cut Pro, making this a great option for creators and artists.

Mashable Tech Editor Timothy Beck Werth reviewed the M4 iPad Air and wrote, "Artists and note-takers will definitely appreciate this one."

More Apple deals

• Apple AirTag (2nd Gen, 4 pack) — $79.99 $99 (save $19.01)

• Apple AirPods Pro 3 — $225 $249 (save $24)

• Apple iPad, 11 inch (A16, WiFi, 128GB) — $399 $449 (save $50)

Best TV deal Opens in a new window Credit: TCL TCL 65-inch QM8L SQD-Mini-LED QLED 4K TV $1,397.99 at Amazon
$2,499.99 Save $1,102.00   Get Deal Why we like it

At a massive 44% off, the 65-inch TCL QM8L could be the perfect upgrade for streaming fall TV shows, movies in the winter, and gaming thanks to the 144Hz refresh rate. While it usually sits at $2,499.99, Amazon has it on sale ahead of Labor Day for $1,397.99 which is the lowest we've ever seen at Amazon.

TCL used their DeepColor System technology in this mode for big contrast and deep blacks. Plus, it comes with Google TV built in for keeping all of your favorite streaming apps neatly organized and easy to access.

If you've ever had a long session with Claude Code or Cowork, you've probably had your context compressed before. Once that happens, Claude starts to lose track of the things you've already worked on. When you reference something in your code, like the "retry logic" section, Claude is going to have a harder time specifically finding that segment of your code. If you pay attention to what it is doing, you'll often see it run multiple commands trying to search for strings that turn up the correct portion of the code.

TL;DR: Watch Barcelona vs. Rayo Vallecano in La Liga for free on CazéTV. Access this free livestream from anywhere in the world with ExpressVPN.

La Liga offers up another big game today, with last year's champions Barcelona in action against Rayo Vallecano.

Rayo Vallecano are yet to win a game this season. And though it's very early days, the talk around Barcelona is whether they can make this another winning campaign, which would make it three league titles in three years. A win against Rayo Vallecano would certainly remind Barcelona's biggest rivals, Real Madrid, that they're in fine early-season form.

If you're interested in watching Barcelona vs. Rayo Vallecano in La Liga from anywhere in the world, we've got all the information you need.

When is Barcelona vs. Rayo Vallecano?

Barcelona vs. Rayo Vallecano in La Liga kicks off at 3.30 p.m. ET on Aug. 31. This fixture takes place at the Camp Nou.

How to watch Barcelona vs. Rayo Vallecano for free

Barcelona vs. Rayo Vallecano in La Liga is available to live stream for free on CazéTV.

CazéTV is geo-restricted to Brazil, but anyone can secure access with a VPN. These tools can hide your real IP address (digital location) and connect you to a secure server in Brazil, meaning you can live stream Barcelona vs. Rayo Vallecano for free from anywhere in the world.

Live stream Barcelona vs. Rayo Vallecano for free by following these simple steps:

1. Sign up for a streaming-friendly VPN (we recommend ExpressVPN)

2. Download the app to your device of choice

3. Open up the app and connect to a server in Brazil

4. Visit CazéTV

5. Watch Barcelona vs. Rayo Vallecano for free from anywhere in the world

Opens in a new window Credit: ExpressVPN ExpressVPN (1-Month Plan) $12.99 only at ExpressVPN (with money-back guarantee) Get Deal

The best VPNs for streaming are not free, but they do tend to offer generous money-back guarantees. By leveraging these offers, you can watch Barcelona vs. Rayo Vallecano without committing your cash. This is obviously not a long-term strategy, but it gives you enough time to stream this fixture (plus Premier League action) before recovering your investment.

What is the best VPN for La Liga?

ExpressVPN is the best choice for bypassing geo-restrictions to stream live sport on CazéTV, for a number of reasons:

• Servers in 105 countries, including Brazil

• Easy-to-use app available on all major devices, including iPhone, Android, Windows, Mac, and more

• Strict no-logging policy so your data is secure

• Fast connection speeds free from throttling

• Up to 10 simultaneous connections

• 30-day money-back guarantee

A two-year subscription to ExpressVPN is on sale for $68.40 and includes an extra four months for free — 81% off for a limited time. This plan includes a year of free unlimited cloud backup and usually offers a generous 30-day money-back guarantee. Alternatively, you can get a one-month plan for just $12.99.

Live stream Barcelona vs. Rayo Vallecano for free from anywhere in the world with ExpressVPN.

A new version of OpenClaw, the open-source AI agent, is available now.

OpenClaw 2.0 was released over the weekend, promising a "simpler setup" and a new browser app. On its website, OpenClaw calls it "by far the largest update in the history of OpenClaw."

The OpenClaw Foundation says that OpenClaw 2.0 was completely crowdsourced with 933 contributors. Of those 933 contributors, 569 of them contributed to the project for the first time.

• Guess what Apple will release in September in our Big Guessing Game: Apple edition. Enter now for a chance to win an Apple Watch!

Here's what we know about OpenClaw 2.0, including how you can try it for yourself.

What's new with OpenClaw 2.0

The new release acknowledges that the original OpenClaw (the AI agent formerly known as Clawdbot, then Moltbot) was more complicated than it needed to be for a first-time users. The update notes say that configurations were moved out of the installation process, simplified, or completely removed in order to make the installation process quick and easy.

The update also includes a completely rebuilt browser in order to simplify OpenClaw. The browser is where users interact with OpenClaw, assign tasks to agents, and track ongoing work and processes.

OpenClaw 2.0 also introduces Shared Cloud Sessions, a feature labeled as a "multiplayer experience." Shared Cloud Sessions allow a user to bring in team members onto ongoing work, allowing multiple people to come in and take over existing tasks with all of the context still attached.

Other new features in OpenClaw 2.0 include a new search function for previous conversations, live session cards that display task progress in real-time, and an assortment of additional new widgets and dashboard features.

There are a number of major security updates in OpenClaw 2.0 as well, including a new private credential request feature, which allows users to securely share credentials with an agent without exposing them in a chat.

OpenClaw quickly went viral when it was released late last year. The self-hosted, open-source agentic AI platform developed such a cult following that founder Peter Steinberger was quickly hired by OpenAI. OpenClaw is now managed by a nonprofit that's supported by OpenAI.

However, OpenClaw has struggled to maintain its position as other AI platforms rolled out similar agentic AI features that were more easily accessible to non-technical users. Perhaps a simplified setup process in this newest version will help OpenClaw retain its positioning among AI users.

How to try OpenClaw 2.0

Since OpenClaw is open-source, developers should have no problem getting it up and running. For less technically-inclined users, you can download OpenClaw 2.0 for Mac or Windows at the OpenClaw website. Simply download the installer and follow the instructions to get started. You can also read all about OpenClaw v2026.8.1 (aka OpenClaw 2.0) in the official release notes.

Be warned — you should understand the cybersecurity risks of handing your device over to an AI agent before installing or using OpenClaw.

Disclosure: Ziff Davis, Mashable’s parent company, in April 2025 filed a lawsuit against OpenAI, alleging it infringed Ziff Davis copyrights in training and operating its AI systems.

These days, most of us share files using methods like email or cloud storage with shareable links. While thumb drives don't seem to be going anywhere, the days of carrying one around with you wherever you go are long over.

At a glance

• The Flash family extends GigaPath and GigaTIME with dramatically improved efficiency, making large-scale pathology research more accessible and practical.
• A distilled pathology foundation model backbone reduces computational requirements without sacrificing performance, enabling repeated analyses across larger patient cohorts.
• These open models support population-scale discovery, helping researchers investigate disease biology, biomarkers, and clinical outcomes across diverse cancer datasets.

GigaPath (opens in new tab) and GigaTIME (opens in new tab) demonstrated how foundation models can support whole-slide analysis and tumor microenvironment modeling from routinely collected pathology data. GigaPath-Flash and GigaTIME-Flash make these capabilities substantially more efficient, enabling researchers to analyze larger cohorts, run more experiments, and move toward population-scale discovery. GigaPath-Flash and GigaTIME-Flash are research models. They are not intended or validated for clinical use, including diagnosis, prognosis, treatment selection, or other patient-care decisions. Performance may vary across datasets, scanners, institutions, populations, and use cases.

The scale opportunity in computational pathology

Histopathology is among the richest and most widely available sources of information in cancer research. Every tissue biopsy produces a whole-slide image that captures cellular morphology at subcellular resolution — and hospitals generate millions of these slides each year. This data contains information relevant to diagnosis, prognosis, treatment selection, and the biology of the tumor microenvironment.

Foundation models have begun to unlock this information at scale. But whole-slide images are large — often exceeding a gigapixel — and applying a foundation model to even a single slide requires processing thousands of image tiles. When a research question involves tens of thousands of patients, the computational cost grows quickly. And population-scale discovery is not a single model run: it requires repeated cycles of feature extraction, statistical analysis, hypothesis testing, and validation across patient subgroups, biomarkers, and clinical endpoints. 

Computational cost limits the number of patients, datasets, tasks, and hypotheses that researchers can study. To realize the full potential of pathology foundation models, we need models that can be applied repeatedly and affordably across large patient populations.

From GigaPath and GigaTIME to the -Flash family

GigaPath (Nature, 2024) (opens in new tab) is a whole-slide foundation model pretrained on large-scale real-world histopathology data from Providence. Unlike models that operate only at the tile level, GigaPath learns contextualized representations of entire slides, capturing both local cellular patterns and global tissue architecture.

GigaTIME (Cell, 2026) (opens in new tab) extends this line of work to tumor microenvironments. Trained on 40 million cells with paired H&E and multiplex immunofluorescence (mIF) data, GigaTIME translates routine H&E images into virtual spatial proteomics maps across 21 protein channels. Applied to over 14,000 cancer patients, it generated a virtual population that uncovered more than 1,200 statistically significant associations between immune cell states and clinical biomarkers.

GigaPath and GigaTIME addressed the scale of pathology data and biological discovery. And now, the Flash family of models addresses scale of experimentation.

Figure 1: Overview of the GigaPath/GigaTIME model family. GigaPath-Flash provides efficient tile and slide encoders at 22M parameters. GigaTIME-Flash predicts spatial proteomics from H&E, replacing the CNN backbone with the distilled ViT-S encoder. Introducing GigaPath-Flash and GigaTIME-Flash

The Flash family shares a core design goal: preserving useful pathology representations while substantially reducing the computational resources required to generate and use them. Both models are built on a common efficient backbone — a compact ViT-S tile encoder distilled from the original billion-parameter GigaPath encoder — and both are released under the Apache 2.0 license. 

GigaPath-Flash

GigaPath-Flash is an efficient foundation model for whole-slide representation learning. It combines a 22M-parameter ViT-S tile encoder with a 21M-parameter LongNet slide encoder. The tile encoder is distilled from the original GigaPath ViT-g teacher, transferring the representational capacity of a billion-parameter model into a backbone that is an order of magnitude smaller. The slide encoder contextualizes all tile embeddings via dilated attention, scaling linearly with the number of tiles.

On slide-level classification benchmarks (PANDA prostate grading and EBRAINS brain tumor subtyping), GigaPath-Flash achieves the lowest inference cost among whole-slide pretrained models while retaining competitive performance — scoring within 3% of the original GigaPath at roughly 50 times less compute.

Figure 2: Efficiency–performance trade-off on whole-slide benchmarks. GigaPath-Flash (red, top-left) achieves competitive performance at substantially lower computational cost than other whole-slide pretrained models. GigaTIME-Flash

GigaTIME-Flash replaces the CNN backbone of the original GigaTIME with the GigaPath-Flash ViT-S encoder, paired with a lightweight convolutional decoder for H&E-to-mIF translation. The model is fine-tuned using LoRA adapters, keeping the pretrained encoder weights largely frozen.

On both in-distribution and out-of-distribution cohorts spanning brain, breast, colon, and lung cancers, GigaTIME-Flash matches or improves upon the original GigaTIME in spatial protein prediction quality. The gains are particularly notable on out-of-distribution data, suggesting that the foundation model backbone improves generalization to previously unseen tissue types.

Figure 3: Mean windowed Pearson correlation for GigaTIME and GigaTIME-Flash on the GigaTIME test set and four out-of-distribution Prov-TMA cohorts. GigaTIME-Flash matches or improves upon the original across all cohorts. Efficiency without giving up the foundation

The efficiency gains of the Flash models are substantial:

ModelTypeEfficiency gainGigaPath-FlashWhole-slide Foundation Model~50× less compute, 97% of predictive performance compared to GigaPathGigaTIME-FlashSpatial Proteomics~6× faster, ~8× less memory, better predictive performance compared to GigaTIME

For a single slide, these differences reduce runtime and hardware requirements. Across tens of thousands of slides, they can determine whether an experiment is practical at all. To illustrate, we estimate the wall-clock time for generating virtual mIF across cohorts of different sizes on a single A100 GPU, assuming approximately 10,000 tiles per slide:

Model1,000 slides100K slides1M slidesGigaTIME-Flash~2 GPU-hours~7 GPU-days~70 GPU-daysGigaTIME~7 GPU-hours~30 GPU-days~300 GPU-daysEstimates assume ~10,000 tiles per slide, batch size 128, single NVIDIA A100 GPU. Actual runtime depends on slide size, tiling resolution, and hardware. Figure 4: GigaTIME efficiency scaling. Left: throughput (tiles/sec) vs. batch size. Right: peak GPU memory (GB) vs. batch size. GigaTIME-Flash scales to over 1,600 tiles/sec while using a fraction of the memory. An open model release

Both GigaPath-Flash and GigaTIME-Flash are released as open-weight models under the Apache 2.0 license. Model weights and code are available on HuggingFace:

• GigaPath-Flash (opens in new tab)
• GigaTIME-Flash (opens in new tab)
• GigaPath-Flash and GigaTIME-Flash: Efficient Pathology Foundation Models for Whole-Slide and Tumor Microenvironment Analysis

This is an early research release. Our current evaluations cover a limited set of benchmarks and cohorts, and broader validation across tasks, scanners, and patient populations is still needed. We expect the most valuable applications of these models to include scientific questions, cohorts, and use cases beyond those in our initial experiments. We welcome community evaluation, and we are equally interested in reports of where these models work well and where they fall short.

Downstream clinical applications will require additional multi-institutional and prospective validation.

Spotlight: Event Series

Microsoft Research Forum

Join us for a continuous exchange of ideas about research in the era of general AI. Watch the latest episodes on demand.

Watch on-demand Opens in a new tab Efficiency as an enabler of discovery

GigaPath and GigaTIME demonstrated what pathology foundation models can learn from whole slides and tumor tissue. GigaPath-Flash and GigaTIME-Flash are a step toward making those capabilities usable across larger populations, more experiments, and a broader research community.

By making pathology foundation models more efficient, we hope to expand the scale of the scientific questions researchers can ask.

Acknowledgements

GigaPath-Flash and GigaTIME-Flash are joint work across Microsoft Research, the University of Washington, and Providence. For technical details, see the paper.

Paper co-authors: Naoto Usuyama, Jeya Maria Jose Valanarasu, Sicong Yao, Hanwen Xu, Jaspreet Bagga, Guanghui Qin, Robert E. Kramer, Cliff Wong, Soohee Lee, Hao Qiu, Theodore Zhengde Zhao, Racheli Ben Shimol, Angela Crabtree, Kevin Matlock, Eduardo Alejandro Lozano Garcia, Naiteek Sangani, Alberto Santamaria-Pang, Maximilian Rokuss, Yashna Hasija, Naisargi Manishkumar Patel, Jason Entenmann, Alexandra Q. Bartlett, Bill J. Wright, Bernard A. Fox, Brian Piening, Sheng Zhang, Sheng Wang, Tristan Naumann, Carlo Bifulco, Hoifung Poon

Opens in a new tab

The post GigaPath-Flash and GigaTIME-Flash: Toward population-scale discovery with efficient pathology foundation models appeared first on Microsoft Research.