Happy Friday wizards,
Here’s what’s brewing in AI.
The big thing
Fable 5 is back. And you can test it until Tuesday.
After three weeks of being banned, Anthropic’s Mythos-class model Fable 5 returned worldwide, after the US Commerce Department finally lifted the export controls that pulled it and it’s bigger brother, Claude Mythos 5.
If you’re a paid Claude user, you have Fable included in your subscription until July 7. After that, it moves to pay-as-you-go usage credits. And we all know what that means, right? Pricey! Specifically: it costs 2x of what Claude Opus does through the API. Prohibitively expensive for most use cases.
So, you have a short time window to really try the most capable model on the market on something genuinely hard. If you don’t have plans this weekend, how about a little vibe coding?
Why it matters First of all, don’t get FOMO. This is primarily a marketing stunt from Anthropic to give people a taste of the model before they paywall the thing hard. Better models will come, and they’ll be cheaper too.
That said, if you’re genuinely curious about knowing where the frontier of LLMs is at this moment in time, here’s what I recommend doing*:
Spec out a task using Opus 4.8 with Max reasoning. Go through a workflow you’re struggling with at work, an app you want to create, whatever as long as it’s a hard problem. Have it create a plan for you, then give that plan to Fable 5; ask it to implement it end-to-end and test everything after. This way you’ll see what it can do, and you might also get something useful that you’ll keep, even after Anthropic has put its money-meter on the model.
*to be fair, you probably need to be on the Max plan ($100/mo or $200/mo) to get enough usage to actually finish the job.
And what about after Fable gets put on usage-based pricing?
Then every builder will be back to the same question: which of these models are actually worth their price?
Well — ZenMux, today’s sponsor, just launched something built to answer exactly that…
IN PARTNERSHIP WITH ZENMUX
AI benchmarks only tell part of the story. The real question is: which model would you actually keep using?
ZenMux just launched Token Economics Arena, where you can compare 10+ popular AI models across coding, agents, long-context reasoning, multilingual tasks, and content creation.
What's even better is that this isn't just an experiment—it also makes it much easier for everyone to try leading Chinese AI models like GLM 5.2, Kimi K2.7, Seed 2.1 Pro and Qwen 3.7. During the campaign, many of these models are available at up to 80% off, making real-world testing much more affordable.
Instead of relying on benchmark scores, ZenMux lets real usage decide.
Every 1M tokens consumed = 1 vote
Test different models on your own tasks
Vote for the model that actually delivers the best experience
At the end of the campaign, ZenMux will award the Most Used Model and People's Choice Model, with the winners receiving a crown badge on the platform.
NEWS NEWS NEWS ❦ NEWS NEWS NEWS
All the small things
Models
Anthropic’s other release, Sonnet 5, is a real upgrade for free users. The new model just replaced Sonnet 4.6 as the default on Claude’s Free and Pro plans, and it’s Anthropic’s biggest Sonnet-leap yet. On everyday knowledge work Sonnet 5 is claimed to be up there with Opus 4.8, trailing it only on hard coding. The flip side, if you’re paying by the token: Sonnet 5’s intro API price is near 40% of Opus, but Artificial Analysis found it burns so many more tokens that a typical task runs about 15% more than Opus 4.8. Cheaper sticker, bigger bill?
Industry moves
OpenAI offered the US government a 5% stake in itself. To ease the political heat, OpenAI floated letting Washington own about 5% of the company (≈$42 billion), which Sam Altman framed as sharing AI’s upside with the public, an “Alaska Permanent Fund for AI” he wants every big lab paying into. Two things stand out: it landed six days after Washington delayed GPT-5.6, and a regulator that owns billions of your stock has every reason to want you to win.
AWS and Microsoft are hiring armies to babysit your AI rollout. Within days of each other, AWS put $1 billion into a “forward-deployed engineering” unit and Microsoft unveiled a $2.5 billion group called Frontier, both built to embed their own engineers inside customer companies and get the AI running in production. Company AI implementations die on messy real data, compliance, and undocumented legacy systems. Around 70% of enterprise AI pilots never reach production. So the labs themselves are staffing companies with people whose job is to make sure they do.
New tools & product features
Anthropic released Claude Science, a new research app (beta on Mac and Linux) that pulls a scientist’s scattered tools into one workspace. It can natively draw 3D protein structures, genome tracks, and chemical diagrams. Anthropic also started its own drug-discovery program aimed at neglected diseases the big pharma companies tend to skip.
Research
The heaviest AI spenders are hiring more people, according to a study. Ramp and Revelio Labs analysed 21,000 US companies and found that the firms spending most aggressively on GenAI grew headcount about 10% and entry-level roles about 12% in the two years after they started.
❦
I wish you an excellent weekend.
You are a delight.
Dario
What's your verdict on today's email?
See which AI use cases are paying off with Context Windows Pro
Most companies pick AI use cases by brainstorming internally. 90% of those initiatives fail.
I’ve created Context Windows just so you can pick the winners.
🟦 Find high-performing use cases from 2,000+ companies at contextwindows.ai, or book a demo with me
Disclosure: To cover the cost of my email software and the time I spend writing this newsletter, I sometimes work with sponsors and may earn a commission if you buy something through a link in here. If you choose to click, subscribe, or buy through any of them, THANK YOU – it will make it possible for me to continue to do this.