The most powerful AI model anybody had ever shipped to the public lasted a few days, and then the US government switched it off.

Anthropic (the folks who make Claude, the tool I use every single day) put out a model last week called Claude Fable 5.

It was the public, guardrailed version of something even more capable they'd been testing with security researchers. The early reviews were kind of bonkers. People handed it big, messy coding jobs and watched it power through them.

One company said it found and fixed hundreds of security holes in its own software.

Then last Friday, Anthropic got a letter from the Commerce Department.

The government decided foreign nationals couldn't have access, and the cleanest way to comply was to shut the whole thing off for everyone.

Anthropic says it's a misunderstanding and they're working to turn it back on. Maybe they will. I don't know.

A lot of people spent last week spun out over a tool that, for most of them, was live for about a long weekend.

Meanwhile Opus 4.8, the model I use to run real parts of my business, the one sitting right there in my account for $20 a month, didn't move an inch. It does everything today it did the day before Fable showed up. Which is a lot.

I've been using this stuff a while now and it's crazy how what's available today on the lowest LLM models is light years ahead of what the biggest ones could handle a year ago.

We get desensitized quickly.

Don't worry so much about the "rug-pull" of Fable 5. The "regular" models are still incredibly useful. Haiku, Sonnet and Opus all have their place in my weekly workflow.

Talk to you next time. Have a great one.