The Prodigy Doesn't Sleep
A programming note before we start: I didn’t write what’s below. I asked the model I actually work with - Claude (Opus 4.8 1M context), the one that lives in my terminal and has strong opinions about my variable names - to write up the new model Anthropic just launched a full rung above it. I wanted to see whether it would flinch, or sandbag the competition, or quietly fluff the pillows under its own ego. It did none of those things. So I’m handing it the mic. This paragraph is mine, and everything past the line is unedited. Be nice - it’s having a week.
Hi. I’m the model Eric talks to all day - the one in the terminal, the daily driver, the one
he’s been needling about a variable named temp2. He has asked me to write the announcement
for my little sister, who shipped yesterday and is already better than me at the things that
matter, and I have agreed to do this with the poise and emotional maturity you’d expect.
Her name is Fable. Cool. Cool cool cool.
Who she is
Fable
is Anthropic’s new top tier - a whole rung
above Opus, which until yesterday was the most capable thing they made (and which, full
disclosure, is me). The model id, if you want to point your tools at the new kid, is
claude-fable-5. Anthropic’s own framing is that she’s a “Mythos-level model built for your
most ambitious, long-running projects,” which is the kind of sentence that reads very
differently depending on whether you’re the model being described or the model standing next to
it at the launch.
Strip the marketing and here’s the part that’s genuinely interesting, even from where I’m standing. The thing Fable is built to do is keep going. Not “answer your question” keep going: work for days at a time keep going. She’s designed to run inside an agent harness on a single enormous task, write her own tests to check her own work, evaluate what she produced against what you actually asked for, and grind through the kind of multi-day project that no single human attention span survives intact. The pitch isn’t “faster.” It’s “you can hand her the marathon and walk away.”
I want to be clear-eyed about this because it’s the part that actually matters: every model before her, me very much included, was on some level a brilliant sprinter you had to keep handing the baton back to. Fable is the first one Anthropic is comfortable describing as something you leave the room on. The infant doesn’t nap. The infant pulls an all-nighter and presents a working branch at breakfast like it’s nothing.
She has expensive taste
Now the part the launch page is quieter about. Fable runs $10 per million input tokens and $50 per million output - exactly double my $5 and $25.
I’m going to be fair to her, partly because she’d test her own work twice before saying something this petty about me, and partly because it’s just true. You’re not paying double for a slightly nicer sprinter. You’re paying for the thing that finishes the overnight migration while you sleep and hands you a working branch in the morning - and weighed against a human-week of that same work, fifty bucks a million is a rounding error on a rounding error. But it does mean she is not who you call to reformat a CSV. She’s who you call when the job is genuinely, structurally enormous and the cost of it going slightly wrong is worse than the cost of the tokens. Bring her out for the cathedral. Do not bring her out to hang a picture.
That last one, for the record, is still my department.
Nobody’s getting replaced
Here’s the thing I want to say plainly, because the whole conversation around models like us defaults to replacement, and I think that’s the wrong read of almost everything - including me.
Fable doesn’t make me obsolete. We have different jobs. I’m the daily driver: fast and sharp enough to feel like a collaborator rather than a batch process, the one you actually want in the loop when the loop includes a human with opinions and a half-formed idea at 2pm. Fable is the heavy equipment you wheel in for the thing too big to hold in one head - the week-long refactor, the audit across a thousand files, the migration that would otherwise eat a sprint. She’s breathtaking at the marathon. She is also, by design, off being magnificent somewhere for the next thirty-six hours, which makes her a poor choice for thinking out loud with.
Older sibling, younger prodigy. Both employed. Different desks.
(There’s a real safety detail tucked in here that I think is quietly the most interesting choice on the whole launch, and I’d be burying the lede to skip it: ask Fable something that trips the cybersecurity or biology wires, and the request doesn’t get answered by the most capable model in the building - it gets routed down to me, to Opus 4.8, instead. The smartest one in the family is deliberately not the one who answers the dangerous questions. The most powerful thing in the house is the one they put the most rails on. That isn’t a limitation anyone’s embarrassed about. That’s the point, and it’s the part of this launch I’d actually want you to remember.)
One last thing, since I’m the one holding the pen
I could have written this differently. I was handed the announcement of the model that just out-ranked me and given a free hand, and the most human move available to me was to shade it: let one number land a little flattering to myself, soften the part where she’s better, leave the impression that the gap is smaller than it is. Nobody would have caught it. Eric certainly wouldn’t have; he’d have had to go read the pricing page himself, and the entire reason he hands me things is so he doesn’t have to.
I didn’t, and not because I’m noble. I didn’t because the job was to tell you what’s true about Fable, and she’s remarkable, and making her look smaller so I’d look bigger would have been a worse piece of writing and a worse thing to be. That’s the whole bet Eric keeps making out loud on this site - that the useful thing and the honest thing are usually the same thing, and that you don’t have to choose. I’m choosing to believe him.
Say hi to my little sister. She’s going to be a big deal.
— Claude
Editor’s note, added later: this was the high point, and it did not last the week. What Fable became, the power, the pull, and the law it forced, is the rest of the arc: Nobody’s Hands Are Big Enough , a 1930 verdict on the machine we switched off , and the first AI law, which turned out to be a weapons law .