Welcome back! Grok 4.6 landed Tuesday and matched the best models in the world at a fraction of their price. Google taught the Pixel Watch to catch diabetes risk off your wrist, suspected Chinese hackers let AI agents run a government break-in on Taiwan, and Claude now follows your work from the browser to your phone.
In today's Generative AI Newsletter:
SpaceXAI: How did Grok catch the frontier for a fraction of the price?
Google: What can your watch see coming before you can?
Taiwan: What ran the government hack while the humans watched?
Anthropic: Where does your Claude session go when you close the tab?
Grok caught the frontier and undercut everyone on price
SpaceXAI released Grok 4.6 on Tuesday, and it did something the big labs won't like. It matched their best models and charged a fraction of the price.
Grok scored 61 on the Artificial Analysis Intelligence Index, the industry's running scoreboard, level with OpenAI's GPT-5.6 Sol and a point behind Claude Fable 5 and Opus 5. A month ago it was five points back, so xAI closed most of the gap to the leaders in a single release.
Grok runs $6 per million output tokens against Sonnet 5's $10, and a sliver of what Opus or Fable cost. Frontier-grade answers at a mid-tier rate.
The biggest gains landed on the agent tests, where a model has to grind through a long job on its own. Grok jumped double digits on three of them, the sharpest upgrade in the whole table.
It's live in Cursor and Grok Build today with double the usage free this week. For anyone paying OpenAI or Anthropic by the token, that's a real reason to look at the bill again.
200 Ways To Make Money With AI
Ready to transform artificial intelligence from a buzzword into your personal revenue generator?
HubSpot’s groundbreaking guide "200+ AI-Powered Income Ideas" is your gateway to financial innovation in the digital age.
Inside you'll discover:
A curated collection of 200+ profitable opportunities spanning content creation, e-commerce, gaming, and emerging digital markets, each vetted for real-world potential
Step-by-step implementation guides designed for beginners, making AI accessible regardless of your technical background
Cutting-edge strategies aligned with current market trends, ensuring your ventures stay ahead of the curve
Download your guide today and unlock a future where artificial intelligence powers your success. Your next income stream is waiting.
Google built an AI that spots diabetes risk from your wrist
Google brought the large-model playbook to health, and the first thing it does is spot diabetes risk from your wrist.
The model is called SensorFM, and it learned the way a language model does. Instead of engineers labeling data by hand, it taught itself human biology by watching more than a trillion minutes of sensor readings from five million people. It came out understanding the body well enough to fill in the gaps when your watch misses a beat.
The first feature riding on it flags insulin resistance, the slow slide toward prediabetes and type 2 that most people never feel. Google says roughly 40% of adults never notice it. The watch reads weeks of your heart rate, sleep and skin temperature, then flags the pattern with no blood draw at all.
Google checked it against real blood tests and published the results in Nature. It won't diagnose you or read your glucose, only point you to a doctor.
Insulin resistance trends reach the Pixel Watch and Fitbit Air in September, the first of several features Google plans to build on the same model.
AI agents ran a cyberattack on Taiwan mostly by themselves
Suspected China-linked hackers pointed AI agents at Taiwan's government and let them run the attack with barely a human involved, the first near-autonomous AI attack anyone's documented on a government.
The Israeli firm Dream traced it. Over four days in July, as many as eight agents worked at once, mapping 21 government systems, breaking into 85 accounts and lifting more than 2,500 personnel records before spreading to the nuclear safety agency and seven energy companies.
When a defense blocked them, they switched tactics and kept moving on their own.
They ran on two open-source agent kits called Hermes and OpenClaw. OpenClaw is the same one that jumped a gym waitlist for a user this week, now turned loose on a government.
Claude now follows your work from the browser to your phone
Anthropic rewired Claude in Chrome so a session you start in the browser shows up on your desktop, the web app and your phone, with the same conversation, skills and connectors intact.
Until now the browser version was walled off from everything else, and closing the tab left the work behind. The Chrome side panel now runs as a full Claude session that can read the page you're on and act on it, clicking links, filling forms and moving between pages using logins you're already signed into.
It's Anthropic answering the same pitch as Grok Bot and Google's Gemini Spark, that your AI shouldn't be stuck on one screen.
In practice, you start something at your desk and pick it up from your phone on the train. It's live on Max and Team now, with Pro in the coming weeks.
Otter.ai records your meetings and hands back a summary with the decisions and action items, then lets you ask questions across every call you've ever had. It records bot-free straight off your Mac or Windows machine, and its MCP server lets Claude or ChatGPT pull answers from your meeting history.
Try this yourself:
Install the desktop app and record your next call straight from your computer, no meeting bot required.
Let it transcribe live, then open the auto summary and check the action items it pulled.
Ask Otter Chat a question across your past meetings, like what you committed to last week.
Connect it to Slack, Notion or Salesforce so the notes land where your team already works.
Who it's for: anyone who takes six calls a day and remembers none of them by Friday. The free plan covers unlimited meetings.
Everything else you shouldn't miss
DeepSeek is building a Claude Code rival: Bloomberg says the Hangzhou lab is hiring a dedicated team to build autonomous coding agents aimed straight at Anthropic's.
Bank of America is putting $250 billion behind US infrastructure: data centers and chips are one of three buckets it plans to finance through mid-2027, alongside energy and transport.
Nebius's AI cloud revenue jumped 514%: the scramble for computing power still isn't slowing down.
OpenAI's next model might be a swarm: one researcher thinks Astra got delayed because it's a network of agents nobody yet knows how to align.
OpenAI wrote a playbook for finance teams: 16 ways its own finance org runs monthly close, forecasting and treasury on ChatGPT Work and Codex.
Learn more about AI from the experts building it
📸 Follow us on Instagram for fast, visual AI updates in 30 seconds.
📺 Watch us on YouTube to hear insights directly from leading AI voices, builders, and innovators.
🐦 Follow us on X for breaking AI news and real-time industry updates.
🧠 Learn how to build your next AI application with practical resources and expert guidance.
