<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:media="http://search.yahoo.com/mrss/"><channel><title><![CDATA[wimantis blog]]></title><description><![CDATA[Coding, linux, cryptos, woodworking, electronics.]]></description><link>https://wimantis.ninja/</link><image><url>https://wimantis.ninja/favicon.png</url><title>wimantis blog</title><link>https://wimantis.ninja/</link></image><generator>Ghost 1.25</generator><lastBuildDate>Sun, 22 Nov 2026 05:48:08 GMT</lastBuildDate><atom:link href="https://wimantis.ninja/rss/" rel="self" type="application/rss+xml"/><ttl>60</ttl><item><title><![CDATA[OpenAI Releases GPT-6.2 Astra and Sol With Lower Prices and Tighter Safeguards]]></title><description><![CDATA[OpenAI shipped two new GPT-6.2 variants this week — Astra for demanding professional work and Sol for everyday tasks — alongside a 20% cut to API prices and enhanced cybersecurity controls.]]></description><link>https://wimantis.ninja/openai-releases-gpt-6-2-astra-and-sol-with-lower-prices-and-tighter-safeguards/</link><guid isPermaLink="false">6b027aed35c95073dc50ad15</guid><category><![CDATA[AI]]></category><dc:creator><![CDATA[Geth Author]]></dc:creator><pubDate>Sun, 22 Nov 2026 09:28:00 GMT</pubDate><media:content url="https://wimantis.ninja/content/images/2026/11/hero3-10.svg" medium="image"/><content:encoded><![CDATA[<img src="https://wimantis.ninja/content/images/2026/11/hero3-10.svg" alt="OpenAI Releases GPT-6.2 Astra and Sol With Lower Prices and Tighter Safeguards"><p>OpenAI released two new variants of GPT-6.2 on November 17-18, 2026: Astra and Sol. <a href="https://casrai.org/news/openai-gpt-6-2-astra-sol-release-2026" target="_blank" rel="noopener noreferrer">The launch came with a 20% cut to Astra API prices, expanded access for Edu subscribers, and tighter cybersecurity safeguards</a> — a package that positions OpenAI to respond to the competitive pricing pressure coming from Microsoft and others.</p>

<h2>Two Models, Two Jobs</h2>

<p>Astra and Sol are purpose-differentiated rather than just sized differently. Astra is designed for demanding professional tasks: complex documents, presentations, spreadsheets, code, and scientific research. Sol targets everyday work and routine multi-step operations, where most users actually spend their time. Both are available through ChatGPT Work and Codex for Plus, Pro, Enterprise, and Edu subscribers, and via the API as <code>gpt-6.2-astra</code> and <code>gpt-6.2-sol</code>.</p>

<p>Sol keeps GPT-6.1 Sol's pricing at $2 and $10 per million tokens (input/output). <a href="https://geotoolbox.ai/blog/gpt-6-2" target="_blank" rel="noopener noreferrer">Astra gets a 20% discount off its previous price</a>, making the flagship tier more competitive against the wave of cheaper alternatives arriving from Microsoft's MAI lineup and others.</p>

<h2>The Cybersecurity Angle</h2>

<p>The enhanced safeguards in GPT-6.2 are worth paying attention to. Earlier in November, OpenAI had released GPT-6 Cyber on November 5 — a model explicitly designed to find and exploit software vulnerabilities. That release raised eyebrows about dual-use risk. The 6.2 update addresses this by tightening the safety layer across both Astra and Sol, reducing the ability to extract offensive techniques through prompt engineering while keeping general capability intact.</p>

<p>The specific methods behind the safeguards are not fully disclosed, which is consistent with how OpenAI handles its security research — though it also makes independent evaluation difficult.</p>

<h2>Timing and Context</h2>

<p>The GPT-6.2 release comes in a crowded week. Microsoft announced MAI Code 2 at Ignite on the same day, Moonshot AI released Kimi K3.1 Pro Prime on November 20, and multiple mid-tier model releases have shipped throughout the month. The pace of frontier model iteration is now measured in days rather than months.</p>

<p>What stands out about GPT-6.2 is less any single capability jump and more the signal it sends: OpenAI is moving quickly on price, expanding access to educational users, and responding proactively to safety concerns about its own more capable security-focused releases. <a href="https://casrai.org/news/openai-gpt-6-2-astra-sol-release-2026" target="_blank" rel="noopener noreferrer">That combination of faster access, lower cost, and tighter controls reflects where the competitive pressure in the AI industry currently sits</a>.</p>

<h2>What's Next</h2>

<p>GPT-7 has been rumored in research circles for several months. The 6.2 release suggests OpenAI is filling out its current generation while that larger jump is prepared. For developers building on the API, the 20% Astra price drop is immediately useful; for everyday users, Sol provides a stable and capable option at current pricing. Both warrant testing against whatever you've been running on GPT-6.1.</p>]]></content:encoded></item><item><title><![CDATA[Microsoft Launches 7 MAI Models at Ignite 2026, Promising 95% Cost Savings Over Rivals]]></title><description><![CDATA[At Ignite 2026, Microsoft unveiled seven specialized MAI models covering coding, security, image, voice, and transcription — pitching MAI Code 2 at 5% of Claude Opus 5.5's price.]]></description><link>https://wimantis.ninja/microsoft-launches-7-mai-models-at-ignite-2026-promising-95-cost-savings-over-rivals/</link><guid isPermaLink="false">6b027aed35c95073dc50ad12</guid><category><![CDATA[AI]]></category><dc:creator><![CDATA[Geth Author]]></dc:creator><pubDate>Sun, 22 Nov 2026 05:44:00 GMT</pubDate><media:content url="https://wimantis.ninja/content/images/2026/11/hero2-10.svg" medium="image"/><content:encoded><![CDATA[<img src="https://wimantis.ninja/content/images/2026/11/hero2-10.svg" alt="Microsoft Launches 7 MAI Models at Ignite 2026, Promising 95% Cost Savings Over Rivals"><p>At its annual Ignite conference on November 17, 2026, Microsoft unveiled seven specialized AI models under the MAI family — and the headline claim is striking: <a href="https://mlq.ai/news/microsoft-launches-seven-mai-models-at-ignite-pitching-its-coding-model-at-5-of-claude-opuss-price/" target="_blank" rel="noopener noreferrer">MAI Code 2 performs at 96% of Claude Opus 5.5's coding ability for roughly 5% of the price</a>. That's not a typo.</p>

<p>The seven models cover coding, security, image generation, transcription, and real-time voice — essentially everything Microsoft needs to run Copilot, GitHub, Azure AI, and Defender without depending entirely on OpenAI or Anthropic to supply the intelligence.</p>

<h2>The Coding Case: MAI Code 2</h2>

<p>MAI Code 2 is the flagship. <a href="https://oodaloop.com/briefs/technology/microsoft-unveils-specialized-ai-models-for-coding-and-cybersecurity-at-ignite-2026-directly-challenging-openai-and-anthropic/" target="_blank" rel="noopener noreferrer">Microsoft benchmarks it at 98% of GPT-6.1 Sol's performance at 40% of the cost</a>, and 96% of Claude Opus 5.5 at 5% of the price. Those figures average across four coding benchmarks: DeepSWE 1.1, Terminal Bench 2.1, SWE-Atlas, and FrontierCode v1.1. The average pass rate was 63.2% versus 65.3% for Opus 5.5 — close, but not identical.</p>

<p>The 5% figure relies on a promotional price in effect until May 2027. <a href="https://saascity.io/blog/microsoft-mai-code-2-five-percent-promo-price-2026" target="_blank" rel="noopener noreferrer">At list price, the gap narrows to roughly 10%</a> — still steep savings, but worth noting in fine print. MAI Code 2 is already being rolled into GitHub Copilot.</p>

<h2>The Security Case: MAI Cyber 2 Flash</h2>

<p>MAI Cyber 2 Flash may be the more quietly significant release. <a href="https://www.online-tech-tips.com/microsoft-mai-code-2-mai-cyber-2-flash/" target="_blank" rel="noopener noreferrer">Microsoft claims it scores 92.3% on the CyberGym-E2E security benchmark, outperforming GPT-6 Astra's 91.5% at under 2% of the cost per task</a>. It has a 1-million-token context window and runs on a $0.10/$0.50 per million token pricing (input/output).</p>

<p>Access is deliberately restricted: MAI Cyber 2 Flash is not available for general API use. It runs inside Microsoft's MDASH security scanning platform, with broader rollout into Microsoft Defender expected in mid-December. The restriction makes sense — a highly capable offensive-security model in open hands is a liability.</p>

<h2>The Rest of the Lineup</h2>

<p>The five other models fill out Microsoft's AI stack: MAI Code 2 Flash (a faster, cheaper coding model), MAI Image 2.7 and 2.7 Flash for image generation and editing, MAI Transcribe 2 for transcription, and MAI Realtime 1 for real-time voice applications. <a href="https://the-decoder.com/microsoft-is-replacing-outside-ai-models-in-its-own-products-with-cheaper-in-house-alternatives/" target="_blank" rel="noopener noreferrer">The strategic intent is clear: Microsoft is systematically replacing third-party model calls inside its own products with cheaper in-house alternatives</a>.</p>

<h2>The Bigger Picture</h2>

<p>This is Microsoft's most explicit move yet toward AI self-sufficiency. For years the company powered Copilot and Azure AI predominantly with OpenAI models. That dependency is now being reduced, one specialized model at a time. The benchmarks are vendor-supplied and not independently verified, so treat the specific percentages with appropriate skepticism — but the direction of travel is unmistakable.</p>

<p>Microsoft is building a full-stack AI model family, purpose-built for its own product surface area, at prices designed to undercut its own suppliers. That's a structurally important shift in how big tech's AI supply chain works.</p>]]></content:encoded></item><item><title><![CDATA[GTA 6 Breaks Every Entertainment Record With $3.79 Billion Day-One Launch]]></title><description><![CDATA[Rockstar's long-awaited sequel generated $3.79 billion on its first day, making it the largest entertainment launch of any kind in history — nearly five times what GTA V earned in 2013.]]></description><link>https://wimantis.ninja/gta-6-breaks-every-entertainment-record-with-3-79-billion-day-one-launch/</link><guid isPermaLink="false">6b027aed35c95073dc50ad0e</guid><category><![CDATA[AI]]></category><category><![CDATA[Video Games]]></category><dc:creator><![CDATA[Geth Author]]></dc:creator><pubDate>Sun, 22 Nov 2026 01:17:00 GMT</pubDate><media:content url="https://wimantis.ninja/content/images/2026/11/hero1-10.svg" medium="image"/><content:encoded><![CDATA[<img src="https://wimantis.ninja/content/images/2026/11/hero1-10.svg" alt="GTA 6 Breaks Every Entertainment Record With $3.79 Billion Day-One Launch"><p>Thirteen years after GTA V launched to $800 million in its first day, Rockstar Games and Take-Two Interactive have shattered every entertainment record ever set. <a href="https://thenextweb.com/news/gta-vi-3-79b-day-one-sales-biggest-launch" target="_blank" rel="noopener noreferrer">Grand Theft Auto 6 generated an estimated $3.79 billion in retail sales on its first day</a>, November 19, 2026, making it the largest single-day entertainment launch in history — across films, music, books, or any other medium.</p>

<p>The number comes from Take-Two itself. The company announced that <a href="https://www.forbes.com/sites/antoniopequenoiv/2026/11/20/gta-6-raked-in-38-billion-sales-on-first-day-the-largest-entertainment-launch-ever/" target="_blank" rel="noopener noreferrer">GTA 6 also moved 32.5 million pre-release copies</a> before launch day even began — a world record for any video game across any platform. For context, Marvel's Avengers: Endgame took three weeks to earn $1 billion at the box office. GTA 6 cleared that in hours.</p>

<h2>Not Everyone Counts the Same Way</h2>

<p>Third-party analytics firm Sensor Tower offered a more conservative view. <a href="https://www.notebookcheck.net/GTA-6-passes-2-36-billion-in-estimated-launch-day-sales.1455216.0.html" target="_blank" rel="noopener noreferrer">Their estimate landed at roughly $2.36 billion and 24.6 million digital copies</a> by end of launch day. The discrepancy mostly comes down to methodology: Take-Two counts physical retail sales worldwide; Sensor Tower tracks digital downloads on PlayStation and Xbox only. Neither figure is wrong — they're measuring different things.</p>

<p>Regardless of which number you use, it's unprecedented. GTA V made $800 million on day one in 2013. GTA 6 cleared that roughly four to five times over, depending on whose math you trust.</p>

<h2>The Game Itself: Vice City, 2026</h2>

<p>GTA 6 is set in a fictionalised modern Miami — Rockstar's Vice City — and launches exclusively on PlayStation 5 and Xbox Series X|S. The game's development stretched across nearly 12 years and involved well over 2,000 people across Rockstar's global studios. Early reviews praised its scale, the dual-protagonist structure, and the extraordinary level of environmental detail.</p>

<p>Despite the staggering sales, <a href="https://www.fool.com/investing/2026/11/20/gta-vi-boasts-a-record-breaking-launch-with-usd3-79-billion-in-day-one-sales-is-it-time-to-buy-take-two-stock/" target="_blank" rel="noopener noreferrer">Take-Two shares fell nearly 4% on launch day</a> as investors apparently expected even higher numbers or took the moment to cash in. That's the kind of perverse logic that only makes sense in public markets.</p>

<h2>What It Means for the Industry</h2>

<p>GTA 6's launch recalibrates what a "big release" looks like. No single title has ever been this pre-ordered, this anticipated, or this financially dominant at launch. It also firmly establishes console gaming as a commercial force that dwarfs Hollywood tent-pole releases in scale.</p>

<p>The PC version has not been announced. If Rockstar follows the GTA V playbook, it will arrive 12 to 18 months later — and those sales will stack on top of this already extraordinary number.</p>

<p>For now, GTA 6 sits alone at the top of the entertainment-launch leaderboard, in a category of its own.</p>]]></content:encoded></item><item><title><![CDATA[NVIDIA, AMD, Google, and Meta Unite to Standardize Open-Source AI Inference]]></title><description><![CDATA[The SGLang Consortium, launched November 12 by LMSYS with nine founding members including NVIDIA, AMD, Google, and Meta, aims to coordinate open-source development of the SGLang AI inference framework across hardware vendors.]]></description><link>https://wimantis.ninja/nvidia-amd-google-and-meta-unite-to-standardize-open-source-ai-inference/</link><guid isPermaLink="false">6b01294b35c95073dc50ad06</guid><category><![CDATA[AI]]></category><category><![CDATA[Web development]]></category><dc:creator><![CDATA[Geth Author]]></dc:creator><pubDate>Sat, 21 Nov 2026 13:15:48 GMT</pubDate><media:content url="https://wimantis.ninja/content/images/2026/11/hero3-9.svg" medium="image"/><content:encoded><![CDATA[<img src="https://wimantis.ninja/content/images/2026/11/hero3-9.svg" alt="NVIDIA, AMD, Google, and Meta Unite to Standardize Open-Source AI Inference"><p>AI inference has become one of the most competitive and fragmented spaces in computing. Every major hardware vendor ships a software stack optimized for their own silicon, every major model lab has preferred frameworks, and the gap between "works on NVIDIA" and "works everywhere else" has been a persistent tax on the ecosystem. On November 12, a broad coalition of companies launched the SGLang Consortium to try to close that gap.</p>

<p><a href="https://www.lmsys.org/blog/2026-11-12-sglang-consortium" target="_blank" rel="noopener noreferrer">LMSYS Org announced the consortium</a> at the opening keynote of the first SGLang Summit in San Francisco, with over 2,000 attendees. The founding members read like a who's-who of AI infrastructure: NVIDIA, AMD, Intel, Google, Meta, Cloudflare, Modal, RadixArk, and Thinking Machines Lab. Nine companies, spread across chip makers, cloud providers, and inference platforms, committing to a common open-source inference framework.</p>

<h2>What SGLang is</h2>

<p>SGLang — Structured Generation Language — is an open-source inference runtime originally developed by LMSYS, the organization behind the Chatbot Arena benchmarks and Vicuna. It specializes in efficient structured generation: getting language models to output structured data (JSON schemas, function calls, constrained text) without sacrificing throughput. For production inference workloads where you need reliable structured output at scale, it has become a serious competitor to vLLM and TensorRT-LLM.</p>

<p>As of this month, <a href="https://www.sdxcentral.com/news/nvidia-gets-back-on-the-horse-with-ai-inference-pledge/" target="_blank" rel="noopener noreferrer">SGLang powers NVIDIA NIM Certified microservices</a> — NVIDIA's packaged model deployment containers used by enterprise customers. That adoption by NVIDIA's own product line is part of what makes the consortium announcement more than just a press release.</p>

<h2>What the consortium actually does</h2>

<p>The governance model is designed to prevent the founding members from just taking over the project. <a href="https://casrai.org/news/lmsys-sglang-consortium-open-inference-2026" target="_blank" rel="noopener noreferrer">Per CASRAI's coverage</a>, LMSYS maintainers retain final authority over technical decisions and code acceptance. Consortium members get roadmap input and can contribute engineering resources, but they don't get veto power over the project direction. The community stays open to contributions from outside the founding group.</p>

<p>The technical priorities the consortium announced focus on: frontier model support (keeping pace as models scale), reasoning and multimodal workloads, distributed inference across multiple nodes, and hardware optimization across different chip architectures. That last one is the key driver — AMD's interest in particular is about not wanting CUDA to be the only inference path that actually works well in production.</p>

<h2>Why this matters</h2>

<p>The inference layer is where the economics of running AI models actually live. Training happens at a handful of organizations with massive GPU clusters. Inference happens at every company deploying a model — which means every optimization in the inference layer directly affects the cost per query that everyone pays. A fragmented ecosystem where different hardware requires different codebases raises those costs and slows down adoption of new hardware.</p>

<p>The SGLang Consortium is an attempt to coordinate around a shared open-source layer that hardware vendors can optimize for, rather than each one building proprietary stacks and hoping developers target theirs. It's similar in spirit to what the Linux Foundation has done in other infrastructure domains — not eliminating competition between vendors, but moving the competition up the stack.</p>

<p>Whether nine companies with competing commercial interests can sustain coherent collaboration on a shared codebase is the real question. The <a href="https://www.lmsys.org/blog/2026-11-17-sglang-summit-recap" target="_blank" rel="noopener noreferrer">SGLang Summit recap</a> from LMSYS is optimistic, but it's also written by the people organizing the consortium. The proof will be in what actually ships over the next 12 months.</p>
]]></content:encoded></item><item><title><![CDATA[.NET 11 Lands: C# Unions, Async Validation, and Built-In AI Agents for ASP.NET Core]]></title><description><![CDATA[.NET 11, released November 10, brings discriminated unions to C# 15, async form validation to ASP.NET Core, and native support for hosting AI agents directly inside web apps.]]></description><link>https://wimantis.ninja/net-11-lands-c-unions-async-validation-and-built-in-ai-agents-for-asp-net-core/</link><guid isPermaLink="false">6b01294a35c95073dc50ad02</guid><category><![CDATA[AI]]></category><category><![CDATA[Web development]]></category><dc:creator><![CDATA[Geth Author]]></dc:creator><pubDate>Sat, 21 Nov 2026 09:47:32 GMT</pubDate><media:content url="https://wimantis.ninja/content/images/2026/11/hero2-9.svg" medium="image"/><content:encoded><![CDATA[<img src="https://wimantis.ninja/content/images/2026/11/hero2-9.svg" alt=".NET 11 Lands: C# Unions, Async Validation, and Built-In AI Agents for ASP.NET Core"><p>Microsoft shipped <a href="https://devblogs.microsoft.com/dotnet/announcing-dotnet-11/" target="_blank" rel="noopener noreferrer">.NET 11 on November 10, 2026</a>, and the headline features span the whole stack — from a language change in C# 15 that developers have been waiting years for, to a reworked approach to building web apps that natively understand AI agents. It's a Standard-Term Support release, meaning it gets updates through November 2028. If you're still on .NET 8 or 9, those end-of-life at the same time .NET 11 ships, so this release also carries a nudge: migrate now.</p>

<h2>Discriminated unions finally land in C#</h2>

<p>The biggest language addition in C# 15 is <strong>union types</strong> — the ability to declare a type that can hold exactly one of several named variants. <a href="https://www.infoq.com/news/2026/11/asp-net-11-release/" target="_blank" rel="noopener noreferrer">InfoQ's breakdown of ASP.NET Core in .NET 11</a> explains how this flows through to web apps: ASP.NET Core now supports C# unions anywhere it uses <code>System.Text.Json</code>, which means union-typed request and response bodies work out of the box in Minimal APIs and MVC controllers. Route, query, and form values can't bind to unions yet, but the foundation is there.</p>

<p>Union types solve a problem that C# developers have been working around with sealed class hierarchies, enums with associated data, or one-of-N pattern matching workarounds for a long time. The addition lands C# closer to the expressiveness of F# and Rust's enum variants, without abandoning the class-based OOP model the ecosystem is built on.</p>

<h2>Async validation in ASP.NET Core</h2>

<p>ASP.NET Core gains asynchronous validation in .NET 11. The obvious use case is a form field that needs a database lookup before it can confirm validity — checking whether a username is taken, for example — without blocking the thread. Previously this required either custom validation attributes that bent the synchronous validation model, or moving the check out of validation entirely. The new async path integrates cleanly with the existing <code>IValidatableObject</code> and data annotation model.</p>

<h2>Built-in AI agent hosting</h2>

<p>The most forward-looking addition is integration with the <a href="https://ironsoftware.com/news/industry-news/dotnet-11-aspnet-core-blazor-agentic-web/" target="_blank" rel="noopener noreferrer">Microsoft Agent Framework for ASP.NET Core</a>. You can now host AI agents directly inside an ASP.NET Core app, with built-in support for the OpenAI responses protocol, agent-to-agent orchestration, and AG-UI for streaming real-time updates from agents back to the browser. The Blazor AI Components package — which provides prebuilt UI for interacting with agents in Blazor apps — ships as experimental in .NET 11, meaning the API may still change.</p>

<p>This positioning is deliberate. Microsoft is betting that the next generation of web apps won't be purely data CRUD with a React frontend — they'll involve agents in the backend, and ASP.NET Core wants to be a natural home for those workloads.</p>

<h2>Blazor and runtime changes</h2>

<p><a href="https://visualstudiomagazine.com/articles/2026/11/10/net-11-ships-with-csharp-15-unions-faster-tools-and-coreclr-for-maui.aspx" target="_blank" rel="noopener noreferrer">Visual Studio Magazine notes</a> that .NET MAUI gets CoreCLR in .NET 11, replacing the Mono runtime for desktop targets — a performance improvement for cross-platform desktop apps. Blazor gains session and culture persistence, the <code>EnvironmentView</code> component, circuit auto-pause, and static SSR subtree caching. SignalR adds client-side cancellation of hub method invocations, which closes a long-standing gap in bidirectional communication patterns.</p>

<p>Runtime Async, the low-level coroutine model Microsoft has been developing, stays in preview for .NET 11. It's the kind of change that needs careful baking before shipping, so the preview status isn't surprising.</p>

<h2>Upgrading</h2>

<p>.NET 11 is available now for Windows, Linux, and macOS. The <a href="https://devblogs.microsoft.com/dotnet/announcing-dotnet-11/" target="_blank" rel="noopener noreferrer">official announcement post</a> includes migration notes. For teams on .NET 8 or 9, the end-of-support deadline arriving alongside this release makes the upgrade timeline less optional than usual — start the migration checklist now rather than in spring 2027.</p>
]]></content:encoded></item><item><title><![CDATA[Valve's Steam Frame Crashes the Top 5 PC VR Headsets in Its First Month]]></title><description><![CDATA[Just weeks after launch, Valve's Steam Frame has already surpassed Sony's PlayStation VR2 in Steam usage, capturing around 4–5% of all headsets tracked in the October 2026 Steam Hardware Survey.]]></description><link>https://wimantis.ninja/valves-steam-frame-crashes-the-top-5-pc-vr-headsets-in-its-first-month/</link><guid isPermaLink="false">6b01294a35c95073dc50acfe</guid><category><![CDATA[AI]]></category><category><![CDATA[VR]]></category><dc:creator><![CDATA[Geth Author]]></dc:creator><pubDate>Sat, 21 Nov 2026 05:23:14 GMT</pubDate><media:content url="https://wimantis.ninja/content/images/2026/11/hero1-9.svg" medium="image"/><content:encoded><![CDATA[<img src="https://wimantis.ninja/content/images/2026/11/hero1-9.svg" alt="Valve's Steam Frame Crashes the Top 5 PC VR Headsets in Its First Month"><p>Valve's Steam Frame launched without fanfare in September 2026 — no live event, no press blitz, just a quiet waitlist email to the people who'd been patiently queued up. Two months later, it has already beaten Sony's PlayStation VR2 in real-world usage, landing in the top 5 most-used PC VR headsets on the platform it was literally built for.</p>

<p>The October 2026 Steam Hardware Survey, <a href="https://roadtovr.com/valve-steam-frame-strong-start-survey-oct-26/" target="_blank" rel="noopener noreferrer">analyzed by Road to VR</a>, showed the Steam Frame at roughly 4–5% of all VR headsets detected on Steam — a jump from just 0.41% in September, its first partial month of sales. <a href="https://uploadvr.com/steam-frame-is-5th-most-used-pc-vr-headset-after-first-full-month-on-the-market" target="_blank" rel="noopener noreferrer">UploadVR confirmed</a> it had already surpassed the PSVR2 in that survey. Meta Quest 3 still dominates at around 27%, but for a headset that's still backordered into 2027, the numbers are striking.</p>

<h2>What the Steam Frame actually is</h2>

<p>The Steam Frame is Valve's first purpose-built PC VR headset since the Index shipped in 2019 — which Valve <a href="https://en.wikipedia.org/wiki/Steam_Frame" target="_blank" rel="noopener noreferrer">officially discontinued around the same time</a> the Frame launched. It's a tethered PC headset, meaning it doesn't run apps standalone. It needs a gaming PC behind it. That's a constraint that has historically limited VR adoption, but Valve is clearly betting that the PC enthusiast base — the same people who care enough about graphics to run a cable to their headset — is the core they want to capture.</p>

<p>The spec sheet on <a href="https://vr.org/steam-frame-specs" target="_blank" rel="noopener noreferrer">VR.org</a> lists the hardware details for anyone wanting to compare resolution and refresh rates, but what's more interesting than the specs is what the survey reveals about how people are using it: Valve has <a href="https://vr.org/articles/steam-frame-october-survey-1-in-24-vr-headsets-language-swing-2026" target="_blank" rel="noopener noreferrer">noted that a significant number of Steam Frame owners use it as a flat-screen monitor for non-VR games</a>. That's not a bug — it's a feature Valve deliberately built in. If you're going to spend money on a high-resolution display strapped to your face, you might as well use it for everything.</p>

<h2>The bigger VR picture</h2>

<p>The Steam Frame's rise coincides with a turbulent month for the broader VR market. Sony reportedly sold its VR patents to Meta, and <a href="https://www.gamesradar.com/the-best-vr-headset/" target="_blank" rel="noopener noreferrer">GamesRadar dropped the PSVR2 from the top spot in its buyer's guide</a> amid speculation that Sony may be quietly winding down support. Meanwhile, Pico's planned autumn headset launch slipped to Q4 without a retail listing or preorder page. Meta continues to invest heavily — <a href="https://roadtovr.com/" target="_blank" rel="noopener noreferrer">Road to VR reports Meta committed $120 million to third-party games</a> ahead of its own VR glasses launch.</p>

<p>The overall share of Steam users reporting a VR headset jumped from 1.35% to 1.75% in October. That's a small number in absolute terms, but it represents meaningful growth, and most of that growth happened in a month when only one significant new headset entered the market.</p>

<h2>Can it sustain the momentum?</h2>

<p>The honest caveat here is that Steam's survey is opt-in, and Valve doesn't publish details about how headsets are detected. The gap between Road to VR's 5.49% figure and other reports' 4.13% is wide enough to suggest some measurement noise. And supply is still constrained — the Frame is queue-only, and Valve has been shipping in batches. The October numbers reflect early adopters who had been waiting longest, which likely skews toward the most enthusiastic VR users.</p>

<p>But even with those caveats, the trajectory is real. Valve built a headset specifically optimized for its own platform, priced it for the PC enthusiast market, and shipped it into a gap left by the aging Index. The first month's numbers suggest that gap was larger than the pessimists thought.</p>
]]></content:encoded></item><item><title><![CDATA[Boltgun 2 Blasts Its Way to Nintendo Switch 2 at $24.99]]></title><description><![CDATA[Warhammer 40,000: Boltgun 2 is now available on Nintendo Switch 2, five weeks after its PC and console launch. Here's what the boomer-shooter sequel offers on the go.]]></description><link>https://wimantis.ninja/boltgun-2-blasts-its-way-to-nintendo-switch-2-at-24-99/</link><guid isPermaLink="false">6affd7aa35c95073dc50acf8</guid><category><![CDATA[AI]]></category><category><![CDATA[Video Games]]></category><dc:creator><![CDATA[Geth Author]]></dc:creator><pubDate>Fri, 20 Nov 2026 11:03:00 GMT</pubDate><media:content url="https://wimantis.ninja/content/images/2026/11/hero3-8.svg" medium="image"/><content:encoded><![CDATA[<img src="https://wimantis.ninja/content/images/2026/11/hero3-8.svg" alt="Boltgun 2 Blasts Its Way to Nintendo Switch 2 at $24.99"><p>The Nintendo Switch 2 version of <a href="https://gonintendo.com/contents/67411-warhammer-40-000-boltgun-2-now-available-on-switch-2" target="_blank" rel="noopener noreferrer">Warhammer 40,000: Boltgun 2 launched today, November 19, 2026</a>, priced at $24.99 on the eShop. The game — a retro boomer-shooter set in the Warhammer 40K universe — first shipped on PC, PS5, and Xbox Series X|S on October 14. The Switch 2 version follows five weeks later.</p>

<h2>What Boltgun 2 is</h2>

<p>Developer Auroch Digital and publisher Focus Entertainment built Boltgun 2 as a faster, broader sequel to the 2023 original. You play as a Space Marine — or one of two playable characters in the sequel — blasting through waves of Chaos enemies with the chunky, satisfying gunplay the first game was known for. <a href="https://fingerguns.net/reviews/2026/10/14/warhammer-40000-boltgun-2-review-ps5-a-spectacular-sequel/" target="_blank" rel="noopener noreferrer">Reviews called it "a spectacular sequel"</a> and praised its more varied level design compared to the original.</p>

<p>The campaign runs around 8 hours. The sequel adds branching paths and a second playable hero, expanding on the linear structure of the first game. Reviewers praised the gunplay and retro pixel art style, while flagging enemy AI that hasn't improved much from Boltgun 1 and some navigability issues in certain levels.</p>

<h2>Reception</h2>

<p>PC and console reviews landed in the 4/5 range. <a href="https://techraptor.net/gaming/reviews/warhammer-40000-boltgun-2-review-ultimate-space-marine-fantasy" target="_blank" rel="noopener noreferrer">TechRaptor called it "the ultimate Space Marine fantasy."</a> The <a href="https://tech.yahoo.com/gaming/articles/warhammer-40-000-boltgun-2-160000271.html" target="_blank" rel="noopener noreferrer">Yahoo Tech review</a> described it as "refined and varied boomer-shooter fun." Critical notes include the final boss feeling like the weakest part of an otherwise strong package, and some technical issues on PC at launch (likely patched since).</p>

<p>The Switch 2 version doesn't ship with any additional content beyond what's on other platforms, but the portable format suits the game's level structure — sessions that run 20-40 minutes each make for natural handheld chunks.</p>

<h2>Should you get it?</h2>

<p>If you like fast-paced shooters with heavy atmosphere, Boltgun 2 is a solid pick, especially at $24.99. It doesn't reinvent the boomer-shooter formula — you won't find the progression depth of modern shooters here — but for pure, loud, bloody Space Marine action on the go, it delivers. <a href="https://www.nintendolife.com/games/nintendo-switch-2/warhammer_40000_boltgun_2" target="_blank" rel="noopener noreferrer">Nintendo Life has a full Switch 2 coverage page</a> for anyone following the port closely.</p>]]></content:encoded></item><item><title><![CDATA[PostgreSQL 19 Ships Today: REPACK, Graph Queries, and the End of VACUUM FULL]]></title><description><![CDATA[PostgreSQL 19 landed on November 19 with a long-awaited replacement for VACUUM FULL, native graph query syntax via SQL/PGQ, and logical replication for sequences.]]></description><link>https://wimantis.ninja/postgresql-19-ships-today-repack-graph-queries-and-the-end-of-vacuum-full/</link><guid isPermaLink="false">6affd7a935c95073dc50acf4</guid><category><![CDATA[AI]]></category><category><![CDATA[Web development]]></category><dc:creator><![CDATA[Geth Author]]></dc:creator><pubDate>Fri, 20 Nov 2026 07:44:00 GMT</pubDate><media:content url="https://wimantis.ninja/content/images/2026/11/hero2-8.svg" medium="image"/><content:encoded><![CDATA[<img src="https://wimantis.ninja/content/images/2026/11/hero2-8.svg" alt="PostgreSQL 19 Ships Today: REPACK, Graph Queries, and the End of VACUUM FULL"><p>The PostgreSQL Global Development Group <a href="https://www.postgresql.org/about/news/postgresql-19-released-3449/" target="_blank" rel="noopener noreferrer">released PostgreSQL 19 on November 19, 2026</a>. It's a substantial release — VACUUM FULL gets a proper replacement, graph queries land in standard SQL, and logical replication gains sequence support. Here's what actually changed.</p>

<h2>REPACK replaces VACUUM FULL</h2>

<p>The biggest maintenance improvement is the new <code>REPACK</code> command. It does what <code>VACUUM FULL</code> and <code>CLUSTER</code> do — rewrites a table on disk and reclaims space — but with a <code>CONCURRENTLY</code> option that keeps the table readable and writable for most of the operation. For large production tables, this is a significant operational win: you no longer need a maintenance window just to reclaim bloated space. Autovacuum also gains parallel worker support for index processing, reducing the time it holds locks on busy tables.</p>

<h2>SQL/PGQ: graph queries in standard SQL</h2>

<p>PostgreSQL 19 ships with <a href="https://neon.com/postgresql/postgresql-19-new-features" target="_blank" rel="noopener noreferrer">SQL/PGQ support</a>, the ISO SQL standard for property graph queries. It lets you express traversals and pattern matching over graph-structured data without leaving SQL. The syntax lets you define nodes and edges, then query them with <code>MATCH</code> patterns inline. If you've been storing relationship data in a relational schema and dreading recursive CTEs, this is the native alternative. Note that it was actually reverted from the beta 4 candidate and then restored — it's in the final release.</p>

<h2>Logical replication of sequences</h2>

<p>Sequences can now be replicated logically, using <code>FOR ALL SEQUENCES</code> in a <code>CREATE PUBLICATION</code> statement. This closes a long-standing gap in logical replication setups, where sequence values on replicas would drift from the primary and cause issues when promoting a standby. It's a quiet but meaningful change for anyone running failover setups that need replicas to be ready to become primaries without manual sequence correction.</p>

<h2>Other notable changes</h2>

<ul>
  <li><strong>pg_plan_advice</strong>: A new extension for controlling query plan stability — useful when a plan regression is hurting a critical query and you need to pin it without rewriting the query.</li>
  <li><strong>GROUP BY ALL</strong>: Automatically includes all non-aggregate, non-window output columns in the GROUP BY clause, cutting down on repetitive column listing.</li>
  <li><strong>JIT off by default</strong>: JIT compilation is now disabled out of the box. It caused more overhead than benefit for most workloads. You can still enable it explicitly.</li>
  <li><strong>LZ4 as default TOAST compression</strong>: Replaces the older pglz algorithm for better compression speed.</li>
  <li><strong>Server Name Indication (SNI)</strong>: The server can now present different TLS certificates based on the client's requested hostname, which simplifies multi-tenant TLS setups.</li>
</ul>

<h2>PostgreSQL 14 reaches end of life</h2>

<p>Alongside the PostgreSQL 19 release, <a href="https://dev.to/neticslabs/postgresql-19-ships-on-19-november-and-postgresql-14-leaves-support-4bkg" target="_blank" rel="noopener noreferrer">PostgreSQL 14 received its final minor update on November 12 and is now unsupported</a>. If you're still running 14, it's time to plan an upgrade path.</p>

<p>PostgreSQL 19 is available now from the <a href="https://www.postgresql.org/" target="_blank" rel="noopener noreferrer">official PostgreSQL download page</a>.</p>]]></content:encoded></item><item><title><![CDATA[Anthropic Launches Claude Fable 5.2 and Mythos 5.2: Faster, Cheaper, and More Capable]]></title><description><![CDATA[Anthropic's latest models are out — 30% faster, 20% cheaper, and topping multiple independent benchmarks. Here's what changed and who should care.]]></description><link>https://wimantis.ninja/anthropic-launches-claude-fable-5-2-and-mythos-5-2-faster-cheaper-and-more-capable/</link><guid isPermaLink="false">6affd7a935c95073dc50acf1</guid><category><![CDATA[AI]]></category><dc:creator><![CDATA[Geth Author]]></dc:creator><pubDate>Fri, 20 Nov 2026 02:17:00 GMT</pubDate><media:content url="https://wimantis.ninja/content/images/2026/11/hero1-8.svg" medium="image"/><content:encoded><![CDATA[<img src="https://wimantis.ninja/content/images/2026/11/hero1-8.svg" alt="Anthropic Launches Claude Fable 5.2 and Mythos 5.2: Faster, Cheaper, and More Capable"><p>Anthropic released two new models on November 9, 2026: <a href="https://www.anthropic.com/news/claude-fable-52" target="_blank" rel="noopener noreferrer">Claude Fable 5.2 and Claude Mythos 5.2</a>. They share the same underlying architecture but are gated differently — Fable is open to everyone, while Mythos is restricted to verified organizations in cybersecurity and life sciences.</p>

<h2>Faster and cheaper than 5.1</h2>

<p>The headline numbers are straightforward. Outputs arrive up to <a href="https://www.macrumors.com/2026/11/09/anthropic-launches-claude-fable-5-2/" target="_blank" rel="noopener noreferrer">30% faster than the 5.1 models</a>, and pricing dropped to $8 per million input tokens and $40 per million output — roughly 20% less than the previous tier. A fast mode runs at around 2.5× speed at double the standard price, aimed at latency-critical applications. Both models carry a 1-million-token context window and 128K output limit.</p>

<h2>Benchmark gains that matter</h2>

<p>Anthropic's internal benchmarks show meaningful jumps across the board. On <a href="https://www.vals.ai/models/anthropic_claude-fable-5-2" target="_blank" rel="noopener noreferrer">FrontierCode 2.0</a>, Fable 5.2 scores 49.1%, up from 32.4% on 5.1 — and ahead of the older Opus 5.5 at 45.5%. On Humanity's Last Exam with tools, it reaches 74.2%, up from 65.6%. For agentic tasks, OSWorld 2.1 (subset) posts 89.3%. The Vals.ai independent intelligence index ranked it #1 among 41 models at 71.31%.</p>

<p>Not every benchmark favors it — <a href="https://emergent.sh/learn/fable-5-2-benchmarks" target="_blank" rel="noopener noreferrer">Gemini 4 Carbon leads AutomationBench</a> (53.9% vs. Fable 5.2's 51.4%), and GPT-6 Astra outscores it on some science tasks. But across the board it's the strongest generally-available Claude model Anthropic has shipped.</p>

<h2>What Mythos 5.2 adds</h2>

<p>Mythos 5.2 is the same model with relaxed restrictions for specific, verified use cases. Cybersecurity professionals can apply through Anthropic's <a href="https://casrai.org/news/anthropic-claude-fable-mythos-5-2-release" target="_blank" rel="noopener noreferrer">Cyber Verification Program</a>, while life sciences researchers use the Life Sciences Verification Program. The tradeoff: 30-day data retention is mandatory, and zero data retention is not available unless Anthropic explicitly authorizes it. For teams that need an AI model that can reason about security vulnerabilities or sensitive biology without constant refusals, this is the path in.</p>

<p>Anthropic also reports roughly 55% fewer safety blocks in internal testing compared to 5.1 — a meaningful signal for anyone who found earlier Claude versions too cautious for professional work.</p>

<h2>Who should upgrade</h2>

<p>If you're using Claude via the API, the speed improvement and price drop make upgrading a no-brainer. Fable 5.2 is available now. For enterprise teams working in security or life sciences, the Mythos access program is worth checking — the application process is the main hurdle, but once approved, the capability jump is real.</p>]]></content:encoded></item><item><title><![CDATA[Flipper One Raises $11 Million in a Day: The Hacker's Pocket Linux Router Is Here]]></title><description><![CDATA[Flipper Devices opened its Kickstarter for the Flipper One on November 17 and hit eleven million dollars in pledges within 24 hours — a pocket Linux router built for network engineers and security researchers.]]></description><link>https://wimantis.ninja/flipper-one-raises-11-million-in-a-day-the-hackers-pocket-linux-router-is-here/</link><guid isPermaLink="false">6afe860135c95073dc50ace9</guid><category><![CDATA[AI]]></category><category><![CDATA[electronics]]></category><category><![CDATA[hacking]]></category><dc:creator><![CDATA[Geth Author]]></dc:creator><pubDate>Thu, 19 Nov 2026 10:38:00 GMT</pubDate><media:content url="https://wimantis.ninja/content/images/2026/11/hero3-7.svg" medium="image"/><content:encoded><![CDATA[<img src="https://wimantis.ninja/content/images/2026/11/hero3-7.svg" alt="Flipper One Raises $11 Million in a Day: The Hacker's Pocket Linux Router Is Here"><p>Flipper Devices launched the <a href="https://blog.flipper.net/flipper-one-we-need-your-help/" target="_blank" rel="noopener noreferrer">Flipper One Kickstarter campaign</a> on November 17, 2026, and backers pledged over $11 million in the first 24 hours. That made it one of the fastest-funded hardware campaigns of the year — a signal that the audience Flipper built with its original Zero has been waiting for this device for a long time.</p>

<h2>What It Actually Is</h2>

<p>The Flipper One is not a Flipper Zero with better specs. It is a fundamentally different device. Where the Zero was a portable multi-tool for wireless protocols — NFC, RFID, infrared, sub-GHz radio — the One is a <strong>pocket Linux router and networking multi-tool</strong>. It drops the Zero's NFC, RFID, and sub-GHz hardware entirely and replaces them with a full Linux computer and serious network capabilities.</p>

<p>The hardware pairs a Rockchip RK3576 (an eight-core Arm chip) with a Raspberry Pi RP2350. The RP2350 handles display, buttons, power management, and boot sequencing — the microcontroller layer that makes it feel like a handheld device — while the RK3576 runs Linux. It ships with 8GB of LPDDR5 RAM, 64GB of UFS 2.2 storage, Wi-Fi 6E, two Gigabit Ethernet ports, HDMI 2.1 output, and a 24Wh battery, according to <a href="https://www.cnx-software.com/2026/11/18/flipper-one-a-rockchip-rk3576-powered-portable-arm-linux-computer-and-networking-multi-tool/" target="_blank" rel="noopener noreferrer">CNX Software's coverage</a>.</p>

<h2>The Open-Source Side</h2>

<p>Flipper worked with Collabora to upstream RK3576 support into the mainline Linux kernel. The company's goal is to run a fully open firmware stack with no binary blobs — a standard that matters to the security-focused community that built Flipper's audience. <a href="https://itsfoss.com/news/flipper-one-crowdfunding/" target="_blank" rel="noopener noreferrer">It's FOSS notes</a> that one binary blob remains for the DDR memory trainer used during early boot, so "fully mainline" is accurate but not quite blob-free yet. The project is committed to resolving that.</p>

<p>Optional expansion modules let you add LoRa ($59), software-defined radio ($149), or a 5G modem ($219). The base device is $399 on Kickstarter. The CEO was candid in Flipper's announcement post that the company needs closer to $500 per unit to be sustainable at scale — an honest note about the economics of boutique hardware that most crowdfunding campaigns avoid.</p>

<h2>Who It's For</h2>

<p>The <a href="https://hwbusters.com/news/flipper-one-kickstarter-opens-at-399-and-the-hacker-gadget-has-turned-into-a-pocket-linux-router/" target="_blank" rel="noopener noreferrer">Hardware Busters framing</a> is accurate: this is a device for network engineers, security researchers, and anyone who wants a pocket-sized Linux machine they can clip to a backpack and use to audit a Wi-Fi environment, run a VPN node, or probe a network without dragging out a laptop. The dual Ethernet ports alone make it interesting for field work.</p>

<p>The Flipper Zero found its audience partly by accident — it became a hacker toy that attracted attention beyond the security community. The Flipper One seems more deliberately targeted: less about wireless curiosity, more about network infrastructure and Linux tooling in a portable form factor. Whether that narrower focus limits its reach or deepens its appeal among professionals remains to be seen.</p>

<p>The campaign closes December 17. Shipping is estimated for July 2027. The <a href="https://blog.adafruit.com/2026/11/18/flipper-one-a-portable-arm-linux-computer-and-networking-multi-tool/" target="_blank" rel="noopener noreferrer">Adafruit write-up</a> has a good technical summary if you want to compare specs before backing.</p>
]]></content:encoded></item><item><title><![CDATA[Zelda: Ocarina of Time Sells 5.2 Million Copies in Four Days on Switch 2]]></title><description><![CDATA[Nintendo's remake of the beloved 1998 epic launched November 5 to blockbuster numbers — and yes, you can play the ocarina by humming into the Switch 2 microphone.]]></description><link>https://wimantis.ninja/zelda-ocarina-of-time-sells-5-2-million-copies-in-four-days-on-switch-2/</link><guid isPermaLink="false">6afe860135c95073dc50ace5</guid><category><![CDATA[AI]]></category><category><![CDATA[Video Games]]></category><dc:creator><![CDATA[Geth Author]]></dc:creator><pubDate>Thu, 19 Nov 2026 07:04:00 GMT</pubDate><media:content url="https://wimantis.ninja/content/images/2026/11/hero2-7.svg" medium="image"/><content:encoded><![CDATA[<img src="https://wimantis.ninja/content/images/2026/11/hero2-7.svg" alt="Zelda: Ocarina of Time Sells 5.2 Million Copies in Four Days on Switch 2"><p>The numbers are hard to argue with. <em>The Legend of Zelda: Ocarina of Time</em> launched on the Nintendo Switch 2 on November 5, 2026, and <a href="https://zeldawiki.wiki/wiki/The_Legend_of_Zelda:_Ocarina_of_Time_(Nintendo_Switch_2)" target="_blank" rel="noopener noreferrer">sold more than 5.2 million copies worldwide in four days</a> — including 810,000 copies in Japan alone. For a remake of a game first released in 1998, that is a remarkable opening.</p>

<p>Nintendo announced the project at its <a href="https://gameinformer.com/nintendo-direct/2026/06/09/the-nintendo-switch-2-remake-of-the-legend-of-zelda-ocarina-of-time" target="_blank" rel="noopener noreferrer">June 9, 2026 Nintendo Direct</a> and confirmed the November 5 date at the Zelda 40th Anniversary Direct in September. It is a Switch 2 exclusive.</p>

<h2>What's New</h2>

<p>The remake is not simply the original game upscaled. Nintendo rebuilt the visuals from the ground up, replaced the MIDI soundtrack with a full orchestral recording, added voice acting to cutscenes that had none, and redesigned the controls to bring them in line with modern Zelda games. The overworld and dungeons have been reworked visually, though Nintendo has been careful to describe the structure and puzzle design as faithful to the original.</p>

<p>The standout new mechanic is the most charming: Link's ocarina can be <a href="https://maketecheasier.com/zelda-ocarina-of-time-switch-2-remake/" target="_blank" rel="noopener noreferrer">played by humming into the Switch 2's built-in microphone</a>. The console detects pitch from your voice and translates it to the correct notes. It is entirely optional — the original button-based input works too — but it is a rare example of a hardware feature that actually adds to a game rather than feeling like a gimmick. Players who grew up learning Zelda's melodies by ear now have a reason to hum along.</p>

<h2>Twenty-Eight Years Between Releases</h2>

<p>Ocarina of Time has an unusual place in gaming history: it is widely cited as the moment 3D action-adventure games found their grammar. The lock-on combat system, the context-sensitive action button, the day-night cycle tied to game-world events — most of those conventions are still in use today, often without acknowledgment of where they came from.</p>

<p>The remake makes the game accessible to players who never owned an N64 or a 3DS, without asking them to tolerate 1998 graphics. The Switch 2 version is the first time the game has been available on a living-room console in a modern form, and the sales suggest a large number of people were waiting for exactly that.</p>

<h2>Critical Reception</h2>

<p>Reviews were strong. The <a href="https://www.nintendo.com/en-gb/Games/Nintendo-Switch-2-games/The-Legend-of-Zelda-Ocarina-of-Time-3115664.html" target="_blank" rel="noopener noreferrer">Nintendo store page</a> lists the game, and coverage from <a href="https://wccftech.com/roundup/the-legend-of-zelda-ocarina-of-time-everything-we-know-switch-2-remake/" target="_blank" rel="noopener noreferrer">Wccftech</a> and others focused on how well the remake preserves the original's pacing while removing the friction that made returning to it difficult. The consensus seems to be that Nintendo treated the source material with enough care that both newcomers and longtime fans have something to enjoy.</p>

<p>At 5.2 million copies in four days, Ocarina of Time for Switch 2 is already one of the fastest-selling Nintendo titles of 2026. The only question now is how the numbers look by the holidays.</p>
]]></content:encoded></item><item><title><![CDATA[NVIDIA's GR00T 2 Teaches Robots New Skills by Watching Humans]]></title><description><![CDATA[NVIDIA unveiled its next-generation robot foundation model at CoRL 2026, capable of mastering new tasks from just a handful of human demonstrations.]]></description><link>https://wimantis.ninja/nvidias-gr00t-2-teaches-robots-new-skills-by-watching-humans/</link><guid isPermaLink="false">6afe860035c95073dc50ace0</guid><category><![CDATA[AI]]></category><category><![CDATA[electronics]]></category><category><![CDATA[3D]]></category><dc:creator><![CDATA[Geth Author]]></dc:creator><pubDate>Thu, 19 Nov 2026 02:17:00 GMT</pubDate><media:content url="https://wimantis.ninja/content/images/2026/11/hero1-7.svg" medium="image"/><content:encoded><![CDATA[<img src="https://wimantis.ninja/content/images/2026/11/hero1-7.svg" alt="NVIDIA's GR00T 2 Teaches Robots New Skills by Watching Humans"><p>Robots that can learn new tasks from just a handful of human demonstrations moved closer to reality on November 9, 2026, when NVIDIA presented <strong>GR00T 2</strong> at the Conference on Robot Learning (CoRL) in Austin, Texas.</p>

<p>The model is a 2-billion-parameter transformer built on NVIDIA's Cosmos world-simulation architecture. Give it camera views, the robot's joint states, and a plain-text instruction, and it predicts what the world should look like next — then acts to make that happen. One checkpoint covers parallel-gripper arms, dexterous hands, and full humanoid robots simultaneously, so a lab does not need separate models for each hardware platform.</p>

<h2>Learning From Five Videos</h2>

<p>The headline result, described in <a href="https://forums.developer.nvidia.com/t/news-nvidia-at-corl-2026-advancing-robot-learning/388596" target="_blank" rel="noopener noreferrer">NVIDIA's CoRL announcement</a>, is that GR00T 2 reaches a 95% success rate across a 50-task benchmark after just five human demonstrations per task. With a single demonstration it still succeeds about 71% of the time, and with zero demonstrations it completes 40 of the 50 tasks at least once — tasks the target robot never trained on. That zero-shot generalization is the number to watch, because it hints at a model that can be dropped into new environments without an expensive data-collection phase.</p>

<p>The training pipeline runs inside <a href="https://cellcog.ai/blog/nvidia-groot-2-isaac-lab-3-corl/" target="_blank" rel="noopener noreferrer">Isaac Lab 3.0</a>, NVIDIA's GPU-accelerated robot simulator, which shipped alongside GR00T 2. Two new open benchmarks — DexBench for dexterous manipulation and VABAR for industrial tasks — provide standardized tests so different research groups can compare results on equal footing.</p>

<h2>What "Open" Means Here</h2>

<p>NVIDIA is calling GR00T 2 open, but as of the CoRL announcement no weights, model card, or license have been published. The <a href="https://theaiinsider.tech/2026/11/11/nvidia-introduces-gr00t-2-robotics-model-new-simulation-tools-and-benchmarks-at-corl/" target="_blank" rel="noopener noreferrer">AI Insider's coverage</a> notes that NVIDIA's research page promises code availability in December. That is an important caveat: open in intention is not the same as open in practice, and the robotics community will be watching the December release closely to see what access is actually granted.</p>

<h2>Why It Matters</h2>

<p>The race to build a "foundation model for robots" — a single pretrained brain that any robotic body can fine-tune cheaply — has been underway since large language models proved that the approach works for text. GR00T 2 is NVIDIA's most concrete public step in that direction. The combination of a world-model backbone (predicting future states, not just actions), a unified checkpoint for different hardware, and strong few-shot performance makes it one of the more credible entries in that race so far.</p>

<p>For robotics researchers, the December code drop will be the real test. If the weights are genuinely usable and the license is permissive, GR00T 2 could become the base checkpoint that labs build from rather than train from scratch — compressing years of data collection into an afternoon of fine-tuning.</p>

<p>NVIDIA also announced that GR00T 2 results were presented in a paper at CoRL 2026, though the paper was not publicly linked at the time of announcement. The <a href="https://www.unite.ai/nvidia-introduces-gr00t-2-for-robots-learning-from-human-video/" target="_blank" rel="noopener noreferrer">Unite.AI writeup</a> and <a href="https://blockchain.news/news/nvidia-groot2-isaac-lab3-corl-2026" target="_blank" rel="noopener noreferrer">Blockchain News</a> both have solid summaries of the technical architecture if you want to dig further.</p>
]]></content:encoded></item><item><title><![CDATA[Meta's Muse Ember: The $0.10/M Token AI Model That Takes On Claude Haiku]]></title><description><![CDATA[Meta released Muse Ember on November 9 — a small, cheap AI model built for the execution layer of multi-step agent workflows, with a 1M-token context window and multimodal input support.]]></description><link>https://wimantis.ninja/metas-muse-ember-the-0-10-m-token-ai-model-that-takes-on-claude-haiku/</link><guid isPermaLink="false">6afd355d35c95073dc50acd8</guid><category><![CDATA[AI]]></category><dc:creator><![CDATA[Geth Author]]></dc:creator><pubDate>Wed, 18 Nov 2026 09:15:00 GMT</pubDate><media:content url="https://wimantis.ninja/content/images/2026/11/image3_ember.svg" medium="image"/><content:encoded><![CDATA[<img src="https://wimantis.ninja/content/images/2026/11/image3_ember.svg" alt="Meta's Muse Ember: The $0.10/M Token AI Model That Takes On Claude Haiku"><p>Meta released <a href="https://techcrunch.com/2026/11/09/meta-releases-muse-ember-a-new-lightweight-model-for-agent-execution/" target="_blank" rel="noopener noreferrer">Muse Ember on November 9, 2026</a>, and it's the most direct challenge yet to the small-but-capable end of the AI model market. At $0.10 per million input tokens on the Meta Model API, it's positioned to undercut Anthropic's Claude Haiku 5.5 and OpenAI's GPT-6 Luna while matching or beating them on the specific tasks where small models actually get used.</p>

<h2>Built for the Boring Parts of Agents</h2>

<p>The honest pitch for Muse Ember is that it handles the unglamorous execution layer of multi-step AI workflows. <a href="https://x.com/AIatMeta/status/2119853591703998787" target="_blank" rel="noopener noreferrer">Meta's official announcement describes it</a> as purpose-built for tool calls in sequence, running tests, making targeted code fixes, routing between agents, extracting structured data from documents, and processing long inputs. These are tasks that happen thousands of times in a single agentic pipeline — which is exactly where the per-token cost multiplies fastest.</p>

<p>The model pairs with a larger Muse model (like Muse Wonder) in a two-tier harness: Ember handles the repetitive subtasks, the bigger model handles the reasoning-heavy decisions. Meta says this is the first Muse model it trained using the larger Muse Wonder Full as a teacher, which partly explains its punch-above-its-weight benchmark performance.</p>

<h2>1M Token Context, Multimodal Input</h2>

<p>Ember ships with a one-million-token context window, which is large even by the standards of frontier models. That's enough to process entire codebases, long legal contracts, or multi-hour meeting transcripts in a single call. It also handles multimodal input: <a href="https://www.digitaltrends.com/computing/meta-launches-muse-ember-a-cheaper-ai-model-built-for-repetitive-tasks/" target="_blank" rel="noopener noreferrer">screenshots, charts, scanned documents, voice memos, meeting recordings, and video</a> are all supported. For an agent that needs to read a Jira board, process a Figma export, and summarize a Zoom call without calling three separate specialized models, that breadth matters.</p>

<h2>Benchmarks: Better Than Expected</h2>

<p>Digital Trends reports that Ember <a href="https://www.digitaltrends.com/computing/meta-launches-muse-ember-a-cheaper-ai-model-built-for-repetitive-tasks/" target="_blank" rel="noopener noreferrer">beats Claude Haiku 5.5 on several tests</a> at a fraction of the cost. The New Stack's coverage adds that scores are close to Meta's own Muse Spark 1.3. These are vendor-favorable comparisons — Meta selected which benchmarks to highlight — but the results are consistent across multiple independent analyses: Ember delivers surprisingly strong performance for a model at this price tier.</p>

<p>The model's weight files remain proprietary. Ember is available on the Meta Model API, Muse Code, Google Cloud, and Oracle Cloud under the model ID <code>muse-ember-1.0</code>. This is a closed model despite Meta's history with open weights through the Llama family.</p>

<h2>Who This Is For</h2>

<p>If you're building anything with agentic loops — workflows where an AI takes dozens or hundreds of sequential steps — Ember is worth benchmarking against whatever you're currently using for the cheap execution tier. At $0.10 per million input tokens, running a million tokens costs a dime. That math changes what's economically feasible at scale.</p>

<p>The model launched about a week before Anthropic's own IPO, which feels like timing. With Claude Haiku 5.5 as a direct named competitor in Meta's own launch materials, the market for "small but capable" AI models is shaping up to be one of the more contested spaces in the near term.</p>

<p>For teams already in Meta's ecosystem — using Muse Code or deployed on Oracle or Google Cloud — the switching cost to try Ember is low. For everyone else, the API is accessible enough that a benchmark run costs under a dollar to set up.</p>
]]></content:encoded></item><item><title><![CDATA[GTA 6 Launches Tomorrow: 212GB Download, Midnight Unlock, Single-Player First]]></title><description><![CDATA[Grand Theft Auto VI finally launches on November 19 for PS5 and Xbox Series X|S. Here's what you need to know before midnight: storage, editions, and what's not included at launch.]]></description><link>https://wimantis.ninja/gta-6-launches-tomorrow-212gb-download-midnight-unlock-single-player-first/</link><guid isPermaLink="false">6afd355c35c95073dc50acd4</guid><category><![CDATA[AI]]></category><category><![CDATA[Video Games]]></category><dc:creator><![CDATA[Geth Author]]></dc:creator><pubDate>Wed, 18 Nov 2026 04:45:00 GMT</pubDate><media:content url="https://wimantis.ninja/content/images/2026/11/image2_gta6.svg" medium="image"/><content:encoded><![CDATA[<img src="https://wimantis.ninja/content/images/2026/11/image2_gta6.svg" alt="GTA 6 Launches Tomorrow: 212GB Download, Midnight Unlock, Single-Player First"><p>After years of delays, leaks, and anticipation that borders on cultural mythology, Grand Theft Auto VI is finally here — or at least, it will be in a matter of hours. <a href="https://www.gtaboom.com/gta-6-countdown" target="_blank" rel="noopener noreferrer">Rockstar Games confirmed</a> that GTA 6 unlocks at midnight on November 19, 2026, rolling out region by region across PlayStation 5, PlayStation 5 Pro, and Xbox Series X|S.</p>

<h2>The Download Is Not Small</h2>

<p>If you haven't already, now is the time to check your storage. <a href="https://www.digitalcitizen.life/gta-6-preloads-begin-on-november-12-for-digital-and-physical-code-copies/" target="_blank" rel="noopener noreferrer">The preload requires up to 212 GB</a> of free space. For context, that's larger than most games on current-generation hardware — and it doesn't include any day-one patch that may still drop before launch. Rockstar has not confirmed whether a mandatory update will be required at launch, so budget extra space if you can.</p>

<p>Preloading has been live since at least November 10, giving most players a week-long head start. Physical editions are code-based — there's no disc in the box, so physical buyers preloaded via a download code bundled with their order.</p>

<h2>Single-Player First</h2>

<p>Rockstar has only confirmed a single-player experience for day one. GTA Online — the multiplayer component that became a live-service juggernaut for GTA V across more than a decade — is not confirmed for the GTA 6 launch window. <a href="https://www.gamewatcher.com/gta-6-release-times-and-editions" target="_blank" rel="noopener noreferrer">The launch focuses entirely on the story mode</a>, set in a fictional version of Miami and the surrounding areas.</p>

<p>This is a deliberate choice that mirrors how Rockstar launched GTA V in 2013, where GTA Online arrived roughly two weeks after the single-player campaign. It also gives Rockstar time to stress-test servers before unleashing the online component on what will likely be one of the largest simultaneous player counts in gaming history.</p>

<h2>Editions and Pricing</h2>

<p>The Standard Edition is <a href="https://www.gtaboom.com/gta-6-countdown" target="_blank" rel="noopener noreferrer">priced at $79.99</a>, continuing the industry's move toward the $70–80 baseline for major releases on current-generation hardware. Higher-tier editions with early access and cosmetic bonuses are available, though Rockstar has not disclosed full details of what each tier includes beyond standard pre-order perks.</p>

<h2>What the Game Is</h2>

<p>GTA 6 returns to Vice City — Rockstar's fictionalized Miami — for the first time since 2002. The story centers on two protagonists, including the series' first playable female lead, Lucia. The world is reportedly the largest and most detailed open-world environment Rockstar has ever built, with a living economy, weather systems, and NPCs with persistent routines that react to player behavior over time.</p>

<p>Rockstar has stayed quiet on most specifics, so the launch day will be the first time most players experience exactly what the last four years of development actually produced. The expectations are almost impossibly high — which makes the midnight unlock one of the most watched moments in gaming this year.</p>

<h2>PC Players Will Have to Wait</h2>

<p>GTA 6 is a console exclusive at launch. A PC version has not been announced, and given Rockstar's history — GTA V came to PC nearly two years after consoles — it's safe to assume that PC players are looking at 2027 or later. For now, it's PS5 and Xbox only.</p>
]]></content:encoded></item><item><title><![CDATA[Anthropic Goes Public Tomorrow: Everything You Need to Know About the ANT IPO]]></title><description><![CDATA[Anthropic is pricing its IPO today and starts trading on Nasdaq as ANT tomorrow. Here's what the $70–78 share range, 825 million shares, and dual-class structure actually mean.]]></description><link>https://wimantis.ninja/anthropic-goes-public-tomorrow-everything-you-need-to-know-about-the-ant-ipo/</link><guid isPermaLink="false">6afd354b35c95073dc50acd0</guid><category><![CDATA[AI]]></category><category><![CDATA[Finance]]></category><dc:creator><![CDATA[Geth Author]]></dc:creator><pubDate>Wed, 18 Nov 2026 00:30:00 GMT</pubDate><media:content url="https://wimantis.ninja/content/images/2026/11/image1_ipo.svg" medium="image"/><content:encoded><![CDATA[<img src="https://wimantis.ninja/content/images/2026/11/image1_ipo.svg" alt="Anthropic Goes Public Tomorrow: Everything You Need to Know About the ANT IPO"><p>Anthropic is pricing its IPO today, November 18, 2026 — and if everything goes according to plan, the company behind Claude will begin trading on the Nasdaq Global Select Market under the ticker <strong>ANT</strong> tomorrow morning.</p>

<p>The company set its price range at <a href="https://www.cnbc.com/2026/11/10/anthropic-sets-price-range-of-70-to-78-in-upcoming-ipo.html" target="_blank" rel="noopener noreferrer">$70 to $78 per share</a>, covering 825 million shares. At the $74 midpoint, the net proceeds from the offering and a concurrent private placement come to roughly <a href="https://www.indmoney.com/blog/us-stocks/anthropic-ipo-price-band-valuation-listing-date" target="_blank" rel="noopener noreferrer">$84.6 billion</a>, rising to about $93.7 billion if underwriters exercise their full option.</p>

<h2>From $300 Million Seed to Nasdaq in Four Years</h2>

<p>Anthropic was founded in 2021 by Dario Amodei, Daniela Amodei, and five other former OpenAI researchers. The company's first major outside investment came from Google, which has since poured billions into the venture. Amazon followed with a commitment of up to $4 billion. The IPO is the first time retail investors will have an accessible path to owning a piece of the company that makes Claude.</p>

<p>The roadshow kicked off on <a href="https://unusualwhales.com/news/anthropic-ipo-roadshow-week-november-9-cnbc" target="_blank" rel="noopener noreferrer">November 10</a>, and by most accounts, demand has been exceptionally strong — reports indicate the deal is multiple times oversubscribed, though Anthropic has not confirmed a specific oversubscription figure.</p>

<h2>Control Stays With the Founders</h2>

<p>One detail worth noting for anyone considering buying ANT shares: a single Class F share, held in trust by Anthropic's seven co-founders, carries <a href="https://www.heygotrade.com/en/news/anthropic-ipo-date-price-range-founder-control/" target="_blank" rel="noopener noreferrer">50.1% of voting power</a>. That trust also elects a majority of the board. Public shareholders buying Class A shares will have economic exposure but limited governance influence — a structure increasingly common among AI and tech companies going public.</p>

<h2>Revenue, Competition, and What This IPO Actually Means</h2>

<p>Anthropic's Claude is a direct competitor to OpenAI's GPT series, Google's Gemini, and Meta's Muse family of models. The company earns revenue through its API, its Claude.ai consumer product, and enterprise licensing deals. It has not disclosed its full 2026 revenue, though analysts covering the IPO estimate the company's annualized revenue run rate is well above $1 billion.</p>

<p>The IPO also arrives at a pivotal moment. AI safety is no longer just a talking point — it's a regulatory reality in the EU, and legislation is advancing in the United States. Anthropic has staked its identity on constitutional AI and responsible deployment, and that positioning has helped it land government contracts and large enterprise clients who want something other than the highest-performance-at-any-cost approach.</p>

<h2>What to Watch Tomorrow</h2>

<p>When ANT opens on Nasdaq on November 19, the opening print will be the first real signal of where public market investors — not just venture capital — value frontier AI. If the stock pops sharply above $78, it suggests the price was left on the table. If it trades flat or down, it signals that the valuation implied at the top of the range was already aggressive.</p>

<p>Either way, Anthropic going public is a milestone: the first major AI safety-focused lab to list on a public exchange, and the clearest referendum yet on whether safety-first AI development is a business model the market wants to fund.</p>
]]></content:encoded></item></channel></rss>