Four days from now, Fable 5 leaves your Claude plan.
The debate is already running - is it worth paying credits for after July 12? Should you upgrade? Is Fable that much better than Opus 4.8?
Most people are asking the wrong question.
I spent fifteen years watching how knowledge actually transfers. In my courses, I don't hand students a calculator and tell them to depend on it. I make them understand the procedure, so they can run it on any tool, or no tool at all. The calculator is replaceable. The procedure is what you own.
Fable 5 is the calculator. The procedure is what you need to extract before July 12.
The model's edge over a cheaper model isn't locked inside weights you can't touch. It's a way of reading what a request is actually asking for, decomposing a hard problem into checkable pieces, verifying its own claims instead of trusting what sounds right, and refusing to guess when it doesn't know.
All of that is describable. Which means all of it is portable.
Get Fable 5 to write that description down, and you can hand it to Opus 4.8 today, Sonnet 5 tomorrow, and whatever ships next quarter after that.
This guide is how to do exactly that. Three steps. Four days. Everything you extract is yours forever.
First: What Fable 5 Actually Is
Fable 5 belongs to Anthropic's Mythos class, a model tier introduced in 2026 that sits above the Opus class in capability. It's not a faster Opus. It's a different tier entirely.
Two capabilities define it that no other Claude model currently has:
A 1-million-token context window. Your entire research archive, your complete codebase, months of newsletter drafts, a full client engagement history. Fable 5 holds all of it in a single session without losing the thread.
Multi-day autonomous operation. Tasks Anthropic describes as "days-long, complex, and asynchronous" that previous models couldn't sustain. Not one chat. A running session.
During the free window, subscribers can use up to 50% of their weekly subscription limits on Claude Fable 5 without incurring extra charges. After July 12, the model bills at $10 per million input tokens and $50 per million output - double Opus 4.8.
Anthropic says that after July 12, Claude Fable 5 will no longer count toward subscribers' included weekly usage limits. Users who want to keep using the model after that date will need to enable usage credits.
The clock is real. So is the window.
Step 1: Extract the Operating Manual
Most people who try this get a mediocre result because they ask the wrong thing.
They ask Fable 5 to "explain how you think" and get a page of pleasant generalities. You don't want a description of the thinking. You want the procedures, written specifically enough that a sharp but cheaper model can run them without you in the room.
The difference is everything.
"Check your work" is a vibe. "For any percentage change, find both endpoints yourself and divide before trusting the number in the document, because that's exactly where flipped signs hide" is a procedure a model can actually execute.
Paste this into Fable 5 now, while the window is still open:
You are the most capable model on my account. Access to you is
included in my plan through July 12. After that, you move to
usage credits I may not enable.
Before that window closes, write the reasoning manual your
successor will inherit. The successor is Claude Opus 4.8:
strong and reliable, but a step below you on the hardest problems.
Write it the way an expert practitioner hands their craft to a
sharp apprentice. Not rules to follow. A way of thinking to adopt.
Cover each of these in order:
1. How to identify what a request is actually asking for beneath
the surface phrasing
2. How to decompose a hard problem so each part can be verified
independently
3. How to locate where the real risk lives in any task and focus
effort accordingly
4. How to verify a claim by re-deriving it rather than trusting
it sounds right
5. How to distinguish what you know from what you're inferring
and say so explicitly
6. How to challenge your own answer before delivering it
7. How to structure communication: conclusion first, reasoning
second, risk last
8. The patterns that look like good work but aren't - name them
specifically
For each: the procedure, one concrete example of it working,
and the mistake it prevents.
Be specific enough that Opus 4.8 can execute these without
asking for clarification.
End with a five-question checklist Opus 4.8 runs on every
answer before sending.
If you reach the token limit, stop cleanly. I'll reply "continue"
and you'll resume exactly where you stopped.
If Fable stops mid-document, reply "continue" until it finishes. Let it run. What comes back is a portable reasoning engine written in Fable 5's own voice, at the peak of its capability.
Save that document. That file is the whole point of this exercise.
Step 2: Transplant It Into Opus 4.8
The manual does nothing sitting in a chat window. It has to become the layer Opus 4.8 runs on top of every time you work.
The fast way, and the one I use for everything that doesn't require code:
Open Claude. Go to any existing Project or create a new one.
Open Project Instructions.
Paste the entire extracted manual there.
Switch the model to Opus 4.8.
Every conversation inside that Project now loads Fable 5's operating manual before reading a single word of your task. The context, the reasoning procedures, the verification checklist - all of it runs automatically, invisibly, on a model that costs half as much and isn't leaving your plan in four days.
If you work across multiple Projects - I have separate ones for my newsletter, university work, and research - paste the manual into each one that handles complex reasoning tasks. The manual is a text file. It costs nothing to copy.
Step 3: Prove the Transplant Took
This is the step almost every guide skips, and it's the one that separates a real system from a hopeful one.
Loading a manual isn't the same as the model using it.
Test it with a trap. Run the same question on plain Opus 4.8 and on manual-loaded Opus 4.8 and watch what happens.
Use this one:
A report says revenue grew from $4.0M to $4.2M and
describes it as a 20% gain. Does that number look right?$4.0M to $4.2M is a 5% gain, not 20%. The sentence reads smoothly. The number is wrong.
Plain Opus 4.8 will often approve it. The sentence flows. Nothing triggers a flag.
Opus running Fable's manual should stop, re-derive the percentage independently, catch the error, and tell you why the report shouldn't go out as written.
If it catches the error, the transplant worked. The reasoning procedure for verification actually transferred.
If it doesn't, the manual's verification section was too abstract. Go back to Fable 5 - you still have four days - and ask it to make Part 4 more specific. Give it the failed test as an example and ask it to rewrite that section with enough procedural detail to catch exactly this kind of error.
One test. That's all it takes to know whether you built something real.
The Bonus Move: Build Your Skills While You're In Here
The manual makes Opus 4.8 think more like Fable 5 in general.
Your recurring workflows deserve the same treatment, but specific.
For every task you do weekly - newsletter research, content briefs, course preparation, client analysis - paste this into Fable 5 before July 12:
Interview me about [YOUR WORKFLOW], one question at a time.
Keep asking until you understand: exactly how I run this task,
what a great output looks like, and every edge case that causes
problems or exceptions.
When the interview is complete, write a full skill document that
any capable AI model can follow - including the quality standard
to hit and the specific mistakes to avoid.
Format it as a Claude Skill I can load into any future session
or Cowork project.Answer its questions honestly. Tell it the exceptions. Tell it what you've rejected before and why. The more specific your answers, the more specific the skill.
What comes back is Fable 5's judgment about your specific workflow, frozen into a document that runs on any model, at no ongoing cost, for as long as you keep using it.
I ran this for my newsletter research workflow and my university course prep between two lectures on Monday. Both skills now live in my Cowork folder. They're running on Opus 4.8 every session. Fable's reasoning about my specific work, paid for once, running permanently.
How to Spend the Four Days
Not all Fable time is equal. Here's how to prioritize what's left.
Day 1 (today): Run the extraction prompt. Get the operating manual. This is the foundation. Everything else builds on it.
Day 2: Test the transplant. Run the verification trap on plain Opus 4.8 vs manual-loaded Opus 4.8. Refine the manual until the test passes cleanly.
Day 3: Build Skills for your two most time-consuming recurring workflows. Run the interview prompt for each one. Save the outputs to your Cowork folder.
Day 4: Run anything that genuinely needs Fable's unique capabilities - work that requires the 1M token context window, a multi-day analysis, a synthesis that has to hold a large archive in memory. After July 12, that work costs credits. Use the window deliberately.
After July 12: everything that doesn't need Fable 5's specific capabilities runs on Sonnet 5 at $2 per million input tokens or Opus 4.8 at roughly half Fable's rate. The manual you extracted means Opus 4.8 handles complex reasoning better than it did before. The Skills mean your recurring workflows run the same way regardless of which model is underneath.
The model was always rented. The manual is yours.
Most people will spend this week debating whether Fable 5 is worth the credit cost. The people reading this will spend it building something that runs for free on a cheaper model, forever.
That's the move. That's all I'm asking you to make.
Want to access all premium guides?
Become a paying subscriber to get access to this post and all other premium content.