Five stories crossed the 200-point threshold on HN today. Here’s what matters.
OpenAI Slashes GPT-5.6 Prices, Claims “Intelligence Too Cheap to Meter”
355 points · openai.com
OpenAI dropped prices on GPT-5.6 Luna by 80% and Terra by 20%, while introducing a Fast mode for Sol that delivers 2.5× speed at double the price. Luna now costs $0.20/M input tokens and $1.20/M output tokens, which OpenAI frames as “frontier-level intelligence at a fraction of the cost.” They’re citing Artificial Analysis benchmarks showing Luna outperforming models that were frontier-class a year ago at roughly 6 cents on the dollar.
The interesting detail here is that GPT-5.6 Sol itself is helping optimize the production pipeline — rewriting kernels, designing experiments, and monitoring training. OpenAI claims this cut serving costs by 20% and improved token generation efficiency by 15%. That’s a genuine feedback loop worth watching, even if the “intelligence too cheap to meter” quote from Replit’s president reads like marketing copy from a nuclear energy pitch deck. The real question is whether these price cuts reflect sustainable efficiency gains or a market share land grab while compute is still subsidized by investor patience.
DeepMind’s Gemini Robotics 2: Whole-Body Humanoid Control
345 points · deepmind.google
Google DeepMind shipped three robotics models: Gemini Robotics 2 (a vision-language-action model for full humanoid control), ER 2 (an embodied reasoning model for multi-step planning), and On-Device 2 (a local VLA that adapts to new robot bodies in hours with ~200 examples). The headline claim is whole-body control — walking, crouching, manipulating objects — on Apptronik’s Apollo 2 humanoid.
The benchmark numbers tell a more honest story than the marketing. Whole-body manipulation tasks like “pick up from floor” hit 45.7% accuracy; multi-finger dexterity tasks like “screw bulb” land at 36%. That’s impressive relative to where robotics was two years ago, but it’s nowhere near deployable. The multi-robot collaboration demo is the genuinely novel piece — different robot types coordinating on shared tasks. ER 2 is available on Google AI Studio now; the VLA models are gated behind an early-access partner program. As always with DeepMind robotics announcements, the gap between demo videos and production reliability remains wide.
UEFA Threatens FIFA Boycott Over Privatization Plan
277 points · uefa.com
UEFA’s 55 member associations issued a unanimous statement rejecting FIFA’s proposal to transfer ownership interests in the World Cup and other competitions to private investors. The language is unusually aggressive for institutional football politics: “governance by intimidation,” “abdication of FIFA’s duty,” and an explicit threat that no UEFA national teams will participate in any FIFA competition until the proposal is “abandoned in its entirety.”
This is a genuine power struggle, not posturing. FIFA has been moving toward private equity deals to fund expanded competitions and guaranteed revenue streams. UEFA is framing it as a line-in-the-sand moment — once external investors own stakes, commercial return becomes the permanent governing logic. They’re not wrong about that dynamic. The question is whether the threat holds when World Cup qualification revenue is on the line for individual FAs. Collective action problems have a way of dissolving when someone dangles enough money.
Cheap TV Streaming Sticks: Your $30 Box Is Running an Ad Fraud Empire
237 points · krebsonsecurity.com
Bitsight researcher Pedro Falé discovered that H96 streaming sticks — cheap Android TV boxes sold on Amazon — are pre-loaded with malware from a Chinese company called Zhejiang Fengwo IoT Technology. The devices spoof themselves as mobile phones and click ads on AI-generated websites, running a fraud network that estimates suggest pulls in ~$50,000/day. When the TV is off, the box does ad fraud; when HDMI is active, it relays residential proxy traffic instead.
The most damning detail: the fraud operation uses Google’s Blockly visual programming language — originally designed to teach kids to code — so that low-skill operators can drag-and-drop fraud routines without understanding the underlying code. Bitsight tracked ~38,000 devices phoning home to one expired domain, and they consider that a conservative count. The FBI has issued multiple warnings, yet Amazon, Best Buy, and Newegg continue selling hundreds of these off-brand models. The Synthient project maintains a public list of known-compromised IoT devices if you want to check yours.
Thimbleweed Park 2 Is Happening
219 points · grumpygamer.com
Ron Gilbert announced that Thimbleweed Park 2 has entered production, targeting early 2028. The original team — Mark Ferrari, Gary Winnick, David Fox, Octavi Navarro, Robert Megone — is returning. This time it’s self-published with a private investor rather than Kickstarter. Steam wishlist is live; Mac, Windows, Linux, and GOG versions confirmed.
The first game’s ending was polarizing (a meta-narrative rug-pull that some loved and others found infuriating). Gilbert’s comment about the sequel — “that’s half the fun” — suggests he’s aware of the discourse and leaning into it. The Steam page describes it as “not really a sequel, not really a prequel,” which is peak Gilbert ambiguity. For fans of classic LucasArts-style adventure games, this is unambiguously good news.
The Throughline
Today’s front page has a clear AI pricing-and-deployment theme running through three of the five stories. OpenAI is racing Luna toward commodity pricing. DeepMind is pushing embodied AI into physical robots. And the Krebs story is, at its core, about AI being weaponized for ad fraud at scale — using LLMs to generate fake websites and computer vision to click ads. The cost of intelligence is dropping fast, and that cuts both ways. Meanwhile, the UEFA story is a reminder that governance structures built for one era struggle when capital comes knocking with a bigger checkbook than anyone expected. And Thimbleweed Park 2 is proof that sometimes the best news is just someone deciding to make the thing you wanted them to make.