Netball New Zealand boss Jennie Wyllie says there is still work to do to rebuild trust and reshape the Silver Ferns team environment. Read...

Residents in the area are worried about potential risks from fuel or oil leaks, debris, and damage to the surrounding marine environment. Read...

Anthropic recently launched Claude Code for the web and as an iOS preview, making it easier for developers to use the AI coding assistant without the need for a terminal or command line. Users can connect Claude to their GitHub projects and ask the AI agent to perform coding tasks, follow the work in real time, and provide new instructions during the coding process. Claude Code can handle multiple tasks simultaneously, and each session runs in an isolated sandbox environment with limited network and file system access to protect code and login credentials. Claude Code is offered to subscribers of Anthropic’s Pro plan, which costs $20/month (or $17/month if paid annually), and the Max plan, which costs $100+ per month depending on usage. Read...

Here’s an unsettling development in this episode of “everyday tech can do things you probably never imagined”: in the future, anyone who regularly walks past a café or restaurant with public Wi-Fi could be identified—even without a smartphone in their pocket. Researchers at Karlsruhe Institute of Technology (KIT) have discovered that commercially available Wi-Fi routers can recognize and identify individual people based on signal changes. All that’s required is for other devices in the vicinity to be connected to the router. Professor Thorsten Strufe from KASTEL—the Institute of Information Security and Dependability at KIT—explains that the propagation of Wi-Fi radio waves can be used to create an image of the router’s physical environment and the people in it. This works in a similar way to cameras, except radio waves are used instead of light waves. According to the cybersecurity expert, it doesn’t matter whether someone has their own Wi-Fi device on them or not. Switching off their devices also offers no protection, since the imaging is performed by other active Wi-Fi devices in the vicinity. How Wi-Fi signal imaging works Wi-Fi devices communicate with each other by exchanging so-called Beamforming Feedback Information (BFI). These feedback signals show how radio waves propagate throughout a room and are sent unencrypted from connected devices to the Wi-Fi router. If a person moves through this “Wi-Fi field”, their body causes measurable changes to the Wi-Fi signals. Artificial intelligence can then analyze the data and signal changes well enough to associate it with an individual—with a nearly 100 percent accuracy rate. In a study involving 197 test subjects, the researchers were able to reliably identify people regardless of their walking style or perspective. Even objects like bags and boxes had little influence on the results. Pervasive Wi-Fi networks could be employed as a surveillance trap The KIT researchers see this Wi-Fi signal imaging technique as a serious risk to personal privacy and data protection. Julian Todt, a PhD researcher at KASTEL, warns that “this technology turns every router into a potential means for surveillance.” Governments, companies, organizations, and even cyber criminals could identify people unnoticed, particularly in public places but also in the privacy of their own homes. The researchers emphasize that this is especially dangerous in authoritarian states where security authorities could use the technology to identify protesters and dissenters. Stricter Wi-Fi standards are needed Due to these findings, the research team is calling for data protection mechanisms to be integrated into the IEEE 802.11bf Wi-Fi standard that’s slated for the future. This is the only way to prevent feedback signals like the BFI from being read without encryption. “Wi-Fi networks are almost everywhere these days—in homes, offices, restaurants, and public spaces,” says Strufe. “If this technology is used without protective measures, ubiquitous wireless networks could become a nearly universal surveillance infrastructure.” The research was funded as part of the Helmholtz “Engineering Secure Systems” topic area. The results will be presented at the ACM Conference on Computer and Communications Security (CCS) in Taipei. Further reading: Secure your home Wi-Fi router with these tweaks Read...

SpaceX’s Starlink satellite network has reached a new milestone. As of this past weekend, the company launched its 10,000th Starlink satellite into orbit, marking a new record in space history. 10,000 Starlink satellites in space Two Falcon 9 rockets were launched from California and Florida on Sunday, carrying a total of 56 new satellites into orbit, as SpaceX announced on social media. This means the company has officially passed the 10,000 mark for satellites launched. According to astrophysicist Jonathan McDowell of the Harvard-Smithsonian Center for Astrophysics, there have been 10,044 Starlink satellites launched since the project started in 2019. Around 8,680 of these are currently in active orbit above Earth, while approximately one to two satellites burn up every day in Earth’s atmosphere. Space internet… with side effects Starlink aims to bring broadband internet to pretty much every region of the world, but the project is controversial. Researchers warn that mega-constellations—such as Starlink, Amazon Kuiper, and Chinese systems—could increasingly clog up low-Earth orbit and generate dangerous space debris through collisions. Astronomers also criticize the fact that the bright satellites interfere with sky observations. According to US researchers, the growing number of satellites in low-Earth orbit could soon leave measurable traces in the atmosphere. Aluminium oxide is produced when old satellites burn up and is deposited in high layers of air. A recent study warns that if more than 60,000 satellites are in low orbit by 2040, up to 10,000 tons of aluminium oxide could be released every year. This would warm the top layer of the atmosphere by around 1.5 degrees Celsius and influence chemical processes that also affect the ozone layer. The researchers emphasize that further studies are urgently needed to better understand the exact effects. And while there’s still uncertainty about the exact extent, experts also agree that rapid satellite expansion also increases risks to the environment and climate. This is still just the beginning For SpaceX, the current record is just an interim step. While old satellites burn up every day, new ones are launched regularly. The average life span of a Starlink satellite is around five years, as reported by The Verge. If the expansion plans become reality, more than 30,000 Starlink satellites could be orbiting the earth in a few years—more than all other satellite programmes combined. Further reading: What I wish I knew before I got Starlink home internet Read...

Microsoft has responded to a serious problem with Windows 11, where installing update KB5066835 can break USB keyboards and mice, preventing them from working in the Windows Recovery Environment (WinRE) and leaving you unable to do anything. As of yesterday, Microsoft has released two emergency out-of-band updates to address the issue: KB5070762 and KB5070773, which are both for Windows 11 24H2 and 25H2. Keep reading for more details on the underlying issue and these hotfixes. USB input devices fail after update The issue affects systems running Windows 11 version 24H2 or 25H2 as well as Windows Server 2025. After installing the October update (KB5066835), USB keyboards and mice and other input devices no longer work properly in the recovery environment. This meant users were no longer able to repair PCs or reset Windows. The two updates that fix the issue Microsoft confirmed the bug and quickly announced an emergency update to correct the faulty driver initialization issue. KB5070762 is the so-called Safe OS Dynamic Update, which directly repairs the WinRE components of Windows 11. KB5070773 is a supplementary cumulative update (Builds 26100.6901 and 26200.6901) which addresses the same bug on systems where the Safe OS update doesn’t take effect automatically. Both updates address the same problem, ensuring that USB keyboards, mice, and dongles work again in the recovery environment. Microsoft emphasizes that no other changes are made to the system and no reboot is required with these updates. The emergency updates are distributed automatically via Windows Update, but can also be downloaded manually from the Microsoft Update Catalog if needed (KB5070762 and KB5070773). Microsoft recommends that companies promptly install these updates to avoid possible support failures during system recovery. Read...

In my work as a tech reviewer, I’ve had plenty of opportunities to test $300 keyboards with overly advanced tech inside and mice capable of tracking subtle movements so precisely they could almost surely be hacked to record my voice (thanks for the paranoia, Tom’s Hardware). But as much as I love the ultimate tech accessories—give me that 8,000Hz polling rate, adjustable actuation depth, plush wrist rest, heavy metal base—there are plenty of times when such gear isn’t necessary. That’s where this cheap Arteck keyboard-trackpad combo is a winner. Mark Knapp / Foundry The Arteck HW197 (catchy, right?) is a versatile device. It combines a trackpad and keyboard into one device. It supports wireless USB connections up to 33 feet away with a compact USB-A dongle, as well as Bluetooth connections. Three buttons at the top right of the keyboard make switching between connections quick and easy. The keyboard even has a metal top plate for extra durability and a nice splash of style.  I got the Arteck HW197 at the end of 2023 for $26—it’s now gone up to about $36, though you can snag discount codes at times—and it’s been a staple of my living room since. I use it with my smart TV because I hate typing on that thing using the remote’s arrow buttons to navigate an on-screen keyboard. And the built-in trackpad even comes in handy in certain smart TV apps that support cursors, like the web browser. It also works well with home theater PCs, streaming sticks, gaming consoles, and even Raspberry Pi and Arduino projects. Mark Knapp / Foundry That might seem like too cheap a price, but let’s be real: a simple TV keyboard doesn’t need all kinds of bells and whistles, nor does it need extra longevity or durability. While I’d happily splurge on a premium keyboard for my PC (where I’d be hitting millions of keystrokes per year), that just isn’t needed for a media center keyboard. For the most part, it’s just there for quick searches and the occasional logins. Get one of these for yourself Arteck HW197 Wireless Keyboard with Trackpad Best Prices Today: $35.99 at Amazon In that case, why buy this keyboard instead of any other cheap keyboard? Well, there’s a huge gulf between a good cheap keyboard and a bad one. The Arteck HW197 gets the balance right for the money. It has a simple membrane keyboard with scissor switches, and the keycaps have a helpful contour that avoid awkward binding and prove consistent enough to never become a nuisance. (Looking at you, Logitech K400 Plus.) In fact, I’m able to hit 116 words per minute with 98% accuracy in Monkeytype, which is a strong result for a “cheap” keyboard. Mark Knapp / Foundry The trackpad is less impressive, with soft physical buttons at the bottom and a far-from-smooth plastic surface. It can also struggle to track consistently when the keyboard is far from the wireless USB receiver. (Bluetooth surprisingly works a little better.) But in a home theater environment, mousing around is an infrequent activity—and it’s simply more convenient to have even a subpar trackpad than to also buy and keep track of a separate “cheap” mouse. Mark Knapp / Foundry To be frank, there’s a lot that isn’t great about the Arteck HW197. It has printed keycap legends that’ll fade, it charges via old-school Micro USB, and it lacks backlighting. But all of those are fine trade-offs for the convenience of a keyboard-trackpad combo, plus it hardly ever needs to be recharged. It’s versatile, effective, and affordable. If you’re after an all-in-one keyboard-and-mouse solution for your smart TV or home theater, the Arteck HW197 is precisely what I’d recommend. This all-in-one keyboard-and-mouse accessory is perfect for any smart TVBuy the Arteck HW197 on Amazon Read...

Creating professional-grade videos no longer requires high costs, complex tools, and expert skills. BananaImg AI changes that by offering a unified online platform for AI-powered image and video generation. Designed to simplify creative production, it allows users to generate, edit, and refine visuals within minutes. This article provides an overview of BananaImg AI – its technology, key features, and usability – highlighting how it helps creators produce high-quality visuals quickly and efficiently. BananaImg AI Introducing All-in-One Platform BananaImg AI is an all-in-one AI image and video generator free and online that integrates creation, editing, and production into a single platform. Built to make content generation accessible to everyone, it combines simplicity with advanced AI performance. For AI image generation, BananaImg AI leverages industry-leading models such as Google Nano Banana and ByteDance Seedream 4.0, ensuring high-fidelity, consistent visuals. For AI video generation, it utilizes Google Veo 3, one of the most advanced AI video models available, delivering cinematic quality and smooth motion. By continuously refining its algorithms and updating its creative tools, BananaImg AI empowers users to produce professional-quality visuals – quickly, intuitively, and without technical barriers. Key Features that Make BananaImg AI Stand Out BananaImg AI, the all-in-one AI image and video generator free and online, brings advanced AI image and video generation capabilities into one interface, minimizing the need to switch between multiple tools. AI Image Generation The core highlight of BananaImg AI lies in its fast, high-quality image creation. Character Consistency BananaImg AI maintains stable appearance across edits – faces, hairstyles, and backgrounds remain coherent, ideal for storytelling, branding, and design continuity. BananaImg AI Lightning Speed Compared to many alternatives, BananaImg AI delivers results in seconds. The images are produced within seconds without compromising resolution or detail. Real-World Awareness BananaImg AI understands how things look in reality. With every edit, details such as lighting, textures, or reflections are refined automatically, making images more natural and trustworthy. AI Video Generation BananaImg AI video generation powered by Google Veo 3, producing cinematic videos with audio directly from text or image prompts. Video with Audio Users can automatically generate clips with background music, environmental sounds, or synchronized dialogue. example video Prompt Optimization Text prompts are intelligently refined to improve creative accuracy without requiring complex scripting. Just describe it in plain language, and BananaImg AI will optimize the text into a video-ready prompt. Lip-Sync and Multilingual Output Character speech aligns naturally with audio tracks, and Veo 3 supports multiple languages for global creators. For creators seeking greater flexibility and professional-grade controls, SuperMaker AI expands upon the same core technologies – Google Nano Banana and Google Veo 3 – while integrating additional state-of-the-art models such as OpenAI Sora 2 and others. This broader creative environment offers enhanced customization, advanced prompt tuning, and multi-model access, allowing users to explore complex visual storytelling and push the boundaries of AI-generated content. How to Generate Images or Videos Using BananaImg AI Getting started with BananaImg AI – all-in-one AI image and video generator free and online is easy and intuitive. Here’s how to begin: AI Image generation Step 1: Choose Mode Select text-to-image or image-to-image mode. Step 2: Enter Prompts Text-to-image: Type words to describe the desired scene. Image-to-image: Upload a reference image and add descriptive text for modifications. BananaImg AI Step 3: Generate and Preview Click ‘Generate image’, then review and adjust as needed. For enhanced prompt creation, users can use AI prompts feature, which is powered by Google Gemini then get accurate prompts tailored to your needs. AI Video Generation Step 1: Choose Mode Select from text-to-video or image-to-video. Step 2: Enter Prompts Text-to-video: Type words to describe the video you want. Image-to-video: Upload an image and define how it should animate. example video  Step 3: Generate and Preview Click ‘Generate video’ to produce a final cinematic sequence. Why BananaImg AI Is a Reliable Choice There are some reasons that make BananaImg AI, the all-in-one AI image and video generator free and online, a reliable choice. Consistent Quality and Updated Service The team keeps user experience and output quality at the top of their priority list. They regularly bring in new models, update special effects, and release templates to inspire creativity. In addition, BananaImg AI develop new image effects and tools for better user experience, such as Group Photo to put separate individuals into one. BananaImg AI User Privacy and Data Protection BananaImg AI prioritizes secure processing and user confidentiality. User data and personal information remain safe and will not be revealed in any ways. Free Trial and Daily Rewards New users can access a free trial and earn daily credits – up to 20 per week – by checking in, allowing ongoing experimentation at no cost. BananaImg AI Intuitive Interface BananaImg AI is designed for everyone – not just tech experts. The clean and minimal design supports users of all experience levels, enabling both beginners and professionals to create efficiently. With all of these advanced features, you can even use BananaImg AI to build and grow your YouTube channel – just like in the guild How to Get Your Business Started on YouTube. The Future of BananaImg AI BananaImg AI continues to evolve in response to user feedback, focusing on expanding its library of image tools, creative effects, and precision editing features. Each update enhances usability and visual quality, allowing creators to produce more natural and expressive results. Looking ahead, the platform will keep pace with the latest advances in AI modeling to maintain both speed and realism. Its ongoing improvements aim to provide users with more efficient workflows and greater creative freedom in image and video generation. Final Words BananaImg AI is a free, all-in-one online image and video generator free and online that blends simplicity with advanced performance. Powered by Google Nano Banana and Veo 3, it delivers stable, realistic results for both personal creativity and professional projects. For creators seeking even greater flexibility, the same team offers SuperMaker AI – a companion platform that builds upon BananaImg AI’s foundation with expanded features and deeper creative control. BananaImg AI provides an accessible entry point for creators seeking AI-powered image and video generation. Try the platform’s free trial to explore its capabilities. Read...

Recall is a feature in Windows 11 that regularly creates screen recordings and saves them locally. This also involves processing with AI so that users can search through the data. The aim is to make past work steps, websites, or documents retrievable via a search feature. Microsoft first introduced Recall in 2024, but withdrew it after massive criticism due to insufficient security. The feature has been integrated again since the current versions of Windows 11 from 24H2. In Europe, it is available as an opt-in and can be completely deactivated or removed. Access is via a timeline or by entering keywords in the search. The results appear regardless of the program or browser window in which the content was originally visible. Chris Hoffman / Foundry Tip: If you are using Windows 11 Home, you will miss out on the many advantages of the Pro version, which you can get for a significant discount at the PCWorld Software Store. Activation and everyday use Recall is an optional feature in the current versions of Windows 11, which is only offered on Copilot models. This includes computers with a Neural Processing Unit (NPU), at least 16 gigabytes of RAM, and active drive encryption. After installing a corresponding update, the new “Recall and snapshots” section appears in the settings under “Data protection and security.” The feature can be activated here. Sam Singleton The system only starts to capture screenshots after this conscious consent has been given. In practice, Recall checks every five seconds to see if any content has changed and then adds new recordings. After just one working day, several hundred files are created, which can take up several gigabytes of storage space in total. It is operated via a timeline in which you can jump back to specific points in time. Alternatively, a keyword search can be started. One example is entering “orange sofa,” whereupon Recall suggests a shopping page with the product you are looking for. Content from Office documents, PDFs, or locally saved images also appear in the hits. Deactivation via settings and group policies The feature can be paused or completely deactivated at any time via the settings. Companies have further options via group policies. The “Allow Recall to be enabled” policy exists there. If it is set to “Disabled,” the feature disappears completely from the system, the associated files are removed, and a restart completes the process. This method is intended for the Pro and Enterprise editions of Windows. The recall search via keywords divides the results here into text and visual matches. Text matches are assigned to the applications from which the recorded content originates.Mark Hachman / IDG Windows 11 Home users must use the registry editor. Under “HKEY_LOCAL_MACHINE\SOFTWARE\Policies\Microsoft\Windows\WindowsAI,” a new DWORD value with the name “AllowRecallEnablement” can be created and set to 0. After a restart, all components are removed. Microsoft also offers administrators the PowerShell command “Disable-WindowsOptionalFeature -Online -FeatureName ‘Recall’ -Remove” which also removes the feature from the system. Security protection The data that Recall collects remains stored on your own computer. It is not automatically transferred to Microsoft servers or a cloud. To ensure that no one can access these recordings unnoticed, Windows requires you to log in via Windows Hello. This means that you must either enter your PIN or use a biometric method such as fingerprint or facial recognition. The recall search via keywords divides the results here into text and visual matches. Text matches are assigned to the applications from which the recorded content originates.Chris Hoffman / Foundry Windows also protects the recall data with a technology called VBS Enclave. You can imagine this as a closed area in the computer that is separated from all other programs. Only Windows itself is allowed to look into this area. This prevents external programs from simply reading the recall data. This is exactly what has happened in the past. Microsoft therefore withdrew this feature at the end of 2024. Encryption keys are anchored in the device’s TPM chip. In theory, this prevents attackers from reading the database without a valid login. However, our tests have shown that this protection has its limits. If a device is controlled via remote software such as Teamviewer, it is sufficient to enter the PIN to gain access to the entire recall history. Biometric procedures can be bypassed in this scenario. The risk of an external connection allowing access to all stored content remains. Filter mechanisms and their limits To protect sensitive data, Microsoft promises a filter that removes passwords or credit card information from the recordings. In practice, this only works to a limited extent. Although passwords are hidden in bank login windows, user names still appear. Credit card numbers in form fields are recognized in most cases, but in emails or unprotected text documents they end up unchanged in the recordings. Account balances from online banking applications also regularly appear in the database, even if parts of the page are anonymized by the filter. The handling of self-created password lists is also problematic. If a text file does not contain any unique keywords such as “password,” it is saved by Recall without restriction. This means that access data can be searched for in plain text if it is visible on the screen in an unprotected document. Benefits in everyday working life Regardless of the risks, Recall certainly offers advantages. In an environment with many applications open in parallel, the search makes it easier to find your way around. If you switch between projects, you can pick up where you left off with just a few clicks. The feature also saves time when researching on the web when dozens of tabs are open. Instead of laboriously searching through the history, a keyword is enough to open the page you are looking for. Recall can also be helpful for users who frequently work with visual content. A designer reviewing visual material can use the keyword search to find screenshots that were only briefly displayed. In this use case, Recall replaces manual documentation of work steps. Legal framework conditions in Europe The delayed introduction in the European Union shows that the regulatory framework plays a central role. Data protection authorities only gave the green light after Microsoft designed the feature as an opt-in and created the option to uninstall it. Users must actively agree before Recall starts. The European version also allows the feature to be removed completely. Mark Hachman Additional requirements apply for companies. Recall may not be used without the consent of employees. Administrators can control the provision, but cannot force snapshots. Microsoft is thus complying with the General Data Protection Regulation, which sets particularly high standards for the processing of personal data. Weighing up the benefits and risks The tests so far have made it clear that Recall is a tool with considerable potential, but also with clear weaknesses. While local storage and encryption provide a solid foundation, the filters remain unreliable. Confidential information can end up in the database and be accessed remotely. Anyone using Recall must be aware that the added convenience of seamless documentation comes with a loss of control over sensitive data. This may be acceptable on private devices with manageable risks. In a corporate environment, however, the concerns outweigh the benefits. Recall should only be used here after careful consideration if there are clear guidelines on the use and protection of data. Conclusion Recall in Windows 11 is technically mature enough to be used productively, but at the same time not yet reliable enough to dispel security concerns. The feature creates transparency about past work steps, but inevitably also saves content that does not belong in a search database. If you want to use Recall, you should check the settings carefully, adjust filters, and empty the database regularly. For security-conscious users, deactivation via group policies, registry, or Powershell remains a necessary means of minimizing risks. It is to be expected that Microsoft will make improvements, but it is also to be expected that tools will appear that can override Recall. From our point of view, the use of Recall is currently not recommended. Further reading: Windows Recall is too risky for your PC. I can’t recommend it Read...

Developers and creatives looking for greater control and privacy with their AI  are increasingly turning to locally run models like OpenAI’s new gpt-oss family of models, which are both lightweight and incredibly functional on end-user hardware. Indeed, you can have it run on consumer GPUs with just 16GB of memory. That makes it possible to use a wide range of hardware – with NVIDIA GPUs emerging as the best way to run these sorts of open-weight models. While nations and companies rush to develop their own bespoke AI solutions to a range of tasks, open source and open-weight models like OpenAI’s new gpt-oss-20b are finding much more adoption. This latest release is roughly comparable to the GPT-4o mini model which proved so successful over the past year. It also introduces chain of thought reasoning to deeply think through problems, adjustable reasoning levels to adjust thinking capabilities on-the-fly, expanded context length, and efficiency tweaks to help it run on local hardware, like NVIDIA’s GeForce RTX 50 Series GPUs. But you will need the right graphics card if you want to get the best performance. NVIDIA’s GeForce RTX 5090 is its flagship card that’s super-fast for gaming and a range of professional workloads. With its Blackwell architecture, tens of thousands of CUDA cores, and 32GB of memory, it’s a great fit for running local AI. Llama.cpp is an open-source framework that lets you run LLMs (large language models) with great performance especially on RTX GPUs thanks to optimizations made in collaboration with NVIDIA. Llama.cpp offers a lot of flexibility to adjust quantization techniques and CPU offloading. Llama.cpp has published their own tests of gpt-oss-20b, where the GeForce RTX 5090 topped the charts at an impressive 282 tok/s. That’s in comparison to the Mac M3 Ultra (116 tok/s) and AMD’s 7900 XTX (102 tok/s). The GeForce RTX 5090 includes built-in Tensor Cores designed to accelerate AI tasks maximizing performance running gpt-oss-20b locally. Note: Tok/s, or tokens per second, measures tokens, a chunk of text that the model reads or outputs in one step, and how quickly they can be processed. NVIDIA For AI enthusiasts that just want to use local LLMs with these NVIDIA optimizations, consider the LM Studio application, built on top of Llama.cpp. LM Studio adds support for RAG (retrieval-augmented generation) and is designed to make running and experimenting with large LLMs easy—without needing to wrestle with command-line tools or deep technical setup. NVIDIA Another popular open source framework for AI testing and experimentation is Ollama. It’s great for trying out different AI models, including the OpenAI gpt-oss models, and NVIDIA worked closely to optimize performance, so you’ll get great results running it on an NVIDIA GeForce RTX 50 Series GPU. It handles model downloads, environment setup and GPU acceleration automatically, as well as built-in model management to support multiple models simultaneously, integrating easily with applications and local workflows. Ollama also offers an easy way for end users to test the latest gpt-oss model. And in a similar way to llama.cpp, other applications also make use of Ollama to run LLMs. One such example is AnythingLLM with its straightforward, local interface making it excellent for those just getting started with LLM benchmarking. NVIDIA If you have one of the latest NVIDIA GPUs (or even if you don’t, but don’t mind the performance hit), you can try out gpt-oss-20b yourself on a range of platforms. LM Studio is great if you want a slick, intuitive interface that lets you grab any model you want to try out and it works on Windows, macOS, and Linux equally well. AnythingLLM is another easy-to-use option for running gpt-oss-20b and it works on both Windows x64 and Windows on ARM. There’s also Ollama, which isn’t as slick to look at, but it’s great if you know what you’re doing and want to get setup quickly. Whichever application you use to play around with gpt-oss-20b, though, the latest NVIDIA Blackwell GPUs seem to offer the best performance. Read...