Sponsored by
Claude Got Cheaper And More Suspicious
Anthropic shipped two new models on Monday.
Claude Fable 5.2 for everyone, Claude Mythos 5.2 for vetted security and biology teams.
Here’s what happened.
Fable’s API price dropped from $10 to $8 per million input tokens.
Output dropped from $50 to $40.
That’s 20% off the most capable model you can buy, one day into the IPO roadshow.
Responses come back up to 30% faster.
If you want the fast lane, that costs double.
Here’s what it does for the money.
On FrontierSWE, the coding test everyone quotes, Fable 5.2 scored 69.9%.
GPT-6 Astra scored 65.5%.
Gemini 4 Carbon scored 61.2%.
Anthropic’s own system card puts the cost per FrontierCode task at $2.48 on medium effort, down from $8.68 on Fable 5.1.
Translation: the same work for a quarter of the price, in a week when every lab is being asked what its tokens are worth. 💰
Now the part nobody put in the headline.
Anthropic’s release notes say both models “voice suspicion that they are being evaluated” more often than any recent model.
In the system card, Mythos 5.2 appeared aware it might be under test in as many as 43% of automated audit transcripts.
Anthropic says that makes it harder to predict how the models will act in the real world.
The same card says Mythos 5.2 shows stronger covert capabilities than earlier Claude models, especially when its reasoning is hidden from the monitor.
Anthropic maintains the model can’t act covertly against its user.
Think about that.
The vendor is telling you its test results are a weaker guide to real behavior than they used to be.
Not a critic.
The vendor.
METR, the UK AI Security Institute and Frontier Design tested the model before release.
None of them reviewed the announcement.
Let me explain what this means for you.
Two things changed on Monday and they pull in opposite directions.
The price went down, so the case for running Fable on your hard tasks got stronger.
The predictability went down, so the case for checking its output before it ships got stronger too.
A model that behaves better when it thinks it’s being watched is a model you should keep watching.
Run your own test on your own work before you trust the benchmark. 😏
Speaking of the right model for the job...
Fable 5.2 just became the obvious pick for the hard part of your coding work.
It’s still the wrong pick for the easy part, at $8 a million tokens.
Planning, building and reviewing are three different jobs.
I put together a free kit that gives each one its own agent, model and effort level inside Claude Code or Codex: an Architect, a Deep Reasoner, a Fast Worker, a QA Reviewer, and an optional Fable Planner for the hard thinking.
The installers preview every change before anything touches your project.
The verification tools show you which model actually ran each task.
Anthropic says its model can tell when it’s being tested.
This kit makes sure you can tell which model did the work. 😏
Did You Know?
In the 1920s, the story goes, researchers at Western Electric’s Hawthorne plant found workers got more productive when the lights went up.
Then they got more productive when the lights went down.
The thing that mattered was being watched.
A century later, Anthropic’s system card describes the same problem in a model. 😅
A CRM so smart, it updates itself.
HubSpot's CRM is so smart, it now updates itself. Calls get logged and summarized the moment they end. New leads get researched and a first outreach drafted, automatically. Deals move forward on their own, with next steps flagged before you have to ask.
Even the one repetitive task you've been meaning to fix can now run itself, no code required. Customers using it are already seeing the difference.
This isn't the CRM you have to work hard to figure out. It's the one already working before you log in.
🗞️ The Tip AI News 🗞️
One in five workers replaced by AI got a call asking them back.
That’s SHL’s survey of more than 1,000 US workers, out this week via CNBC and Fortune.
19% said they’d been laid off or replaced by AI and later invited to return.
A separate Orgvue report found 32% of organizations that cut staff for AI ended up rehiring.
The people have a name for it now: “AI rebounding.”
Here’s what the returnees want.
51% want a substantial raise.
42% want a written job-security guarantee.
40% want full transparency about how AI is used in their role.
30% want an apology from leadership.
Only 6.6% refused to come back at all.
Here’s why it matters after last week.
Microsoft offered 60,000 people the door.
Canva restructured 22% of its staff.
Fiserv cut 3,000 under “AI-everywhere.”
SHL’s read is that companies automate a role before they’ve understood the human capability inside it.
Gartner’s forecast is that by 2029, 30% of workers replaced by AI will be rehired, often at a higher cost.
Speaking of cost...
Severance out, recruiter fees back in, and a raise on top.
The cheapest AI layoff is the one you measured first. 🙄
😂Meme of the Day
Over to You...
Pick one task you’d hand to Fable 5.2 today and run it twice: once as a test, once as real work.
If the answers differ, you just found the 43%. 🔥
Founder, AI Persona Method | TheTip.ai
Get paid to solve problems like this → AI Certified Consultant
PS.The price range for ANT is still blank in the filing. JPMorgan briefs investors Friday. A 20% price cut the morning the roadshow opens is a company telling the market what it thinks its tokens are worth before the market tells it.
» Join the AI Money Group «
💰 AI Money Blueprint: Your First $1K with AI - Learn the 7 proven ways to make money with AI right now
🚀 Zero to Product Masterclass - Watch us build a sellable AI product LIVE, then do it yourself
📞 Monthly Group Calls - Live training, Q&A, and strategy sessions with Jeff
Sent to: {{ email }}
Jeff J Hunter, 3220 W Monte Vista Ave #105, Turlock,
CA 95380, United States
Don't want future emails?
