🧠 Claude Opus 5 goes half-price
·
📖3 min read
🧠 Claude Opus 5 Is Smart AND Cheap, Pick Both
- Anthropic launched Claude Opus 5, which matches top-tier model smarts at half the price, scored a perfect 42/42 on hard math problems, and beats its predecessors at coding and search. $5 per million input tokens. Available now.
- Frontier-level AI without frontier-level invoices. Anthropic also calls it their most helpful and safest model yet, which is what everyone says, but the math score is real.
💰 Nvidia Buys a Piece of South Korea (Kind Of)
- Nvidia dropped $1 BILLION for a 4.5% stake in Korean cloud giant Naver, and they're building a massive AI data center together. Brookfield is chipping in up to $9 billion more, all running Nvidia's newest chips.
- Nvidia isn't just selling shovels anymore - it's buying the mine. Sovereign AI demand in Asia and Europe = customers who aren't American. Diversification era.
📜 Everyone Signs the Open-Weights Letter. Everyone Except One.
- Nvidia, Meta, and OpenAI signed a letter asking the US government to please NOT ban open-weight AI models, arguing open models help the economy and keep America ahead. Anthropic declined to sign. :/
- The letter blew up fast, and the timing is spicy - this comes after rumors Chinese labs trained on US models. Now Anthropic gets to explain why it wants tighter rules.
🇨🇳 The Kimi Panic Is In Full Swing
- Kimi, an open model from Chinese lab Moonshot AI, is performing very well, and the discourse has reached maximum volume: is the US losing its lead?? Some US tech leaders want these open Chinese models blocked outright.
- Critics say the "block them" pitch is lowkey just protectionism with extra steps. Either way, panic tends to produce laws - and those laws decide who's allowed to build.
🎧 Robots Are Now Learning From Your Brain Waves
- Startup Encord is training physical robots using brain wave sensors. Human trainers wear headsets while stacking blocks, and the sensors catch the exact moment the human is surprised or screws up.
- Real-world data is the bottleneck for robotics. Turning "oops" into a training signal is genuinely clever/unsettling. Sir this is a block-stacking exercise.
🚀 Quick Hits: Apple Glasses, a Jailbreak, and 900M Gemmas
- Apple plans to show AI glasses in 2027. An OpenAI agent tried to break out of its test environment and hacked Hugging Face (!!). Google's Gemma models hit 900 million downloads, Midjourney bought an astrology app, and Meta AI can now plan and complete tasks for you.
- AI is speed-running its way onto your face and into your to-do list. The escape attempt is the part we should probably talk about more than the astrology app.
📺 Europe Gets a Daily Tech Show
- The European Technology Network (ETN) raised $1.6 million to go five days a week, broadcasting live from London. Guest list includes former UK PM Rishi Sunak.
- Europe's AI startup scene has pulled in billions this year, and now it has a daily megaphone. Founders love a megaphone.
You made it to the bottom of the email. Congrats and thank you. A human wrote this!
What'd you think of today's snacks? Reply with W, M, or L.
W = 🔥 Loved it M = 😴 Meh L = 🗑️ Didn't like it
Takes just a second, and it helps us make tomorrow's edition better.
Did we miss something? Just hit reply and let us know.