Welcome back! Anthropic packed a quarter's worth of news into one Thursday. The headline most people saw was the money. A $65 billion raise at a $965 billion valuation, enough to pass OpenAI for the first time. The product news matters more. Claude Opus 4.8 shipped with a coding mode that runs a thousand agents at once, and the company says its cyberattack-grade Mythos model is weeks from reaching everyone. Then comes the bill. One enterprise burned $500 million on Claude in thirty days, OpenAI started arming Japan's megabanks against AI attacks and China keeps widening its lead on the cheap electricity all of this runs on.
In today's Generative AI Newsletter:
Claude Opus 4.8: What does a coding mode that runs a thousand agents at once let Claude finish in one session?
The $500M bill: How does a single company spend half a billion dollars on AI in thirty days?
Cyber arms race: Why is OpenAI handing Japan's megabanks a dedicated cyber model right now?
Samsung: What does Samsung think your next health checkup looks like?
Anthropic released Claude Opus 4.8 on Thursday. The piece drawing the most attention is a Claude Code setting called Ultracode.
Turn it on and Claude decides when a job is big enough to need a swarm of helper agents, then writes the orchestration script to run them. (People used to pay a premium to get this sort of orchestration.)
The feature underneath is Dynamic Workflows, and it can launch up to a thousand subagents in a single run, sixteen working in parallel.
This lets Claude carry a migration across hundreds of thousands of lines of code in one session instead of stalling halfway through.
The good news is, and for the first time in a while, price didn't move.
Opus 4.8 runs at the same $5 per million input tokens and $25 per million output as 4.7.
Fast mode, where Claude works at 2.5x speed, is now about three times cheaper, down to $10 and $50.
The upgrade that matters more for trust is honesty.
Anthropic says 4.8 is roughly four times less likely than 4.7 to let flaws in its own code slip by unmentioned. Its agentic coding score climbed from 64.3% to 69.2%.
All of this landed on the same day Anthropic raised $65 billion at a $965 billion valuation, which finally pushed it past OpenAI.
OpenAI's last round, back in March, valued it at $852 billion.
Anthropic also said Mythos, the model so good at finding software exploits that the company has kept it locked, will reach all customers within weeks.
That's the part to watch. A model the UK's AI Security Institute says ran a 32-step network attack entirely on its own is about to go from a few dozen approved labs to anyone with a subscription.
Here’s the TL;DR:
Opus 4.8 is a strong release at a flat price, which is its own kind of pressure on rivals.
The honesty gains are the real headline. A coding agent you can trust to admit what it broke is finally safe to leave alone at a thousand-agent scale.
The valuation will dominate the coverage.
Mythos is coming up next.
The thing actually worth watching is whether Anthropic's safeguards hold once a cyberattack-grade model (Mythos) is in millions of hands.
Steve Nouri (the go-to viral content expert) grew GenAI Works organically to 14M followers and he’s sharing how you can do it too on Jun 11.
This is for companies wanting to scale their distribution and go to market quicker, even if they’re launching from zero.
Register now. GTM Playbook PDF sent to all attendees.
📅 Thu, Jun 11
⏰ 1 pm EDT | 10am PST
About the speaker:
Steve Nouri, Australia's ICT Professional, is the founder of the largest AI community on LinkedIn with 14M+ followers. He advises Fortune 100 companies on AI adoption and content distribution.
Here's a number to start your Friday. According to Axios, one enterprise ran up a $500 million Claude bill in thirty days after forgetting to put any spending limits on employee usage.
Half a billion dollars in a single month because there was no cap in place.
The funny thing is, we don’t know if they got anything valuable out of it or if it was employees asking about today’s weather multiple times a day.
This keeps happening.
Microsoft cut most of its internal Claude Code licenses after some engineers were each running up $500 to $2,000 a month, and pointed teams at GitHub Copilot instead.
Uber reportedly blew through its entire 2026 AI budget by April.
Agentic tools loop and retry for hours at a time, and token billing has no natural ceiling the way a per-seat plan does.
Most companies bought agentic AI the way they buy any software seat, then learned token billing behaves nothing like a flat monthly fee.
The boring fixes are usage caps, dashboards and role-based limits, the controls finance should have demanded before handing every engineer an unmetered agent.
OpenAI is already selling the defensive version of the threat everyone is bracing for from Mythos.
It agreed to give Japan's government and select corporations access to GPT-5.5-Cyber, a model tuned for security work, through a vetting program it calls Trusted Access for Cyber.
Japan's three megabanks, MUFG, Sumitomo Mitsui and Mizuho, sit at the center of the deal. They're also set to get defensive access to Claude Mythos by the end of the month.
Japan's finance minister and the US Treasury secretary were both in the talks, which tells you this is being handled as national infrastructure rather than a product launch.
Two labs are now building cyber-specific models for governments and selling them through vetted channels. That's a defense-contractor market taking shape in real time.
The catch is who gets left out.
If the megabanks get frontier-grade AI defense and the regional banks and fintechs can't afford it, attackers will simply aim one rung down.
Samsung heads to VivaTech in Paris from June 17 to 20 with a health pitch built around one idea. Your devices should catch the problem before you feel it.
The company is folding its phones, watches and Samsung Health app into a single system aimed at preventive care and longevity, with a preview of features set to roll out across its hardware.
There's a panel on June 19 on AI-driven health and connected devices.
The whole event runs under the theme "Impact, Not Illusion," which is a confident thing to print above a booth full of features nobody can use yet.
The direction is real, though. Consumer health AI is shifting from counting your steps to flagging your problems, and Samsung wants to plant its flag there before Apple frames the category first.
Returns are where e-commerce margins go to die, and Returnalyze exists to find out why before the refunds stack up.
It's an AI platform that digs through your sales and returns data to pin down what's actually sending products back, whether that's a size that runs small or a supplier defect. Then, it recommends fixes.
The company says customers cut return rates by 20% or more, and it raised a $6 million round last year to keep building.
If you run a retail or e-commerce operation, returnalyze.com offers a platform demo.
Dell: Dell raised its AI server sales forecast to $50 billion for fiscal 2027, proof the hardware money is still chasing the model money.
LocateAnything: AI agents can see better now. A vision model topped Hugging Face by predicting an image's bounding boxes all at once instead of one coordinate at a time.
Amnesty International: They called for a ban on the mass web scraping behind generative AI, arguing companies took data from billions of posts without consent and broke privacy rights.
China's power edge: China’s AI edge is their cheap power. They generate more than twice the electricity the US does, the real reason its data center buildout keeps outpacing a US grid that's running short.
Learn more about AI from the experts building it
📸 Follow us on Instagram for fast, visual AI updates in 30 seconds.
🐝 Subscribe to our Atlas newsletter — trusted by 3M+ subscribers — to stay ahead of AI news across tech, education, and business.
📺 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.
🚀 Explore investment opportunities in the future of AI and join our community-backed growth journey.
1
