Watch now or save to watch later:
Fable is back. But not for long.
Anthropic has restored access to Claude Fable 5, and for a few days the expensive thing is sitting inside normal paid plans for up to 50% of weekly usage limits. That window closes on July 7, 2026.
After that, it moves to usage credits and the API price is a cool $10 per million input tokens and $50 per million output tokens. If you don’t use the API and have no idea what it means I’ll translate: effing expensive!
So yes, this is time-sensitive. It closes
This is basically Oracle of Delphi energy. You get to walk up to the very clever machine and ask it the big question…
And most people are still going to ask it to tidy an email. Please don't! Hell if you are going to do this sell me your usage instead ha!
Do the 20% test
OK so what should you ACTUALLY use your (limited) Fable on?
This is the filter I would use: Ask Fable things where being 20% smarter changes the answer entirely.
Is it actually 20% smarter than all the other models? Maybe. Doesn’t matter. Use this as a mental framework. Basically think of it as the smartest person you’ve ever talked to…what would you ask them?
Judgment calls. Tricky trade-offs. Untangling messy situations. The question behind the question. The hard stuff!
Do not use it for boilerplate code. Do not use it for formatting. Do not use it to summarise something Opus or GPT-5.5 can already summarise perfectly well. That is using a freight train to pick up your shopping.
Total waste.
If you have a messy business decision, give it that. If you have a huge project repo and you cannot see the shape of the thing anymore, give it that. If you have a personal or professional decision you have been dodging because the answer is probably inconvenient, give it that.
Build the question first
I’m seeing a lot of people absolutely paralysed about what to ask Fable. With such limited usage and such a tight time constraint I get it. Easy to get blocked.
Here’s a concrete tip…use a “dumber” model to help you construct your question.
Do not sit there staring at the Fable box trying to write the perfect prompt.
Use a cheaper, slightly dumber model as the workshop. (God, it feels weird calling Opus 4.8 or GPT-5.5 dumb… but you know what I mean.)
Brain-dump your situation into Opus or GPT-5.5 first. Tell it you have one chance to ask a much smarter model a question. Get it to interview you. Get it to pull the uncomfortable context out of you. Get it to draft the full one-shot prompt.
Then pressure-test it:
what is the prompt missing?
what would Fable need to know if I get no follow-ups?
what assumptions am I hiding?
And because your usage may be capped it’s important to ask in one go. You don’t want to end up answering lots of back and forth with Fable as each turn will eliminate tokens. So get it right first time as much as you can!
Assume you get one prompt per day. Especially on the $20/month plan!
Maybe a handful more if Claude is feeling generous. Maybe five if you are on the $200 plan and burn through your allowance like an idiot. On $200/month I’m getting 4-5 every 5 hours. And maxing them out every time!
So the prompt needs three bits:
context: who you are, the situation, what you have tried
question: one specific, high-stakes thing
safety net: assumptions, counter-arguments, likely missing context, and what it should do if it would normally ask a follow-up
This is different from normal chat prompting. Normal prompting is often a conversation - we have “turns” - going back and forth. With Fable it’s much closer to briefing an expert before they walk into a boardroom and you are not allowed to speak again.
Ask the nasty thing
The best Fable question is probably hiding behind something uncomfortable.
What decision have I been avoiding?
Where am I working hard on the wrong thing?
What would embarrass me if someone brilliant looked at my business for an hour?
I gave Fable my AI brain and asked versions of that. It has my project logs, my decisions, the stuff I have been working on, the stuff I said I would do and then quietly wandered away from. If you want to build that sort of shared context, the AI brain guide is here.
And it called me out. Brutally. It was quite upsetting!
It basically said I keep killing winners. I start a project, prove people want it, get bored when it starts working, then wander off to the next shiny thing…
Rude. And 1000% accurate. Which is annoying.
That is what this model is for. Not "rewrite this LinkedIn post in a more professional tone." Nah, use it for BIG gnarly questions and problems.
The way I like to think about it is that answers are getting cheap. Good questions are scarce. And that is on you.
One final point for those who may be feeling overwhelmed by having to get full usage out of Fable right now. i) It will eventually come to our normal subscriptions which is cool but ii) more importantly: ALL models will eventually be at this level. It may be 6 months from now. It may be 12 months from now. But it will happen.
So if you feel you’ve missed the boat this time don’t worry. This level of capability will become the norm. Which is both terribly exciting and terribly scary!
To the Task,
Kyle