Anthropic just redeployed Claude Fable 5 — the most capable model they have ever released — and right now it's included on the Pro, Max, or Team plan you already pay for. That window closes on July 7, 2026. After that, using it costs usage credits per token.

Most people are going to spend these last few days the wrong way: asking it questions, running one-off tasks, treating the best model on Earth like a slightly better chatbot. This is the move that actually banks the value — the one from the video.

The idea in one line

Don't spend three days of genius *doing tasks*. Spend three days of genius *building the machine that does those tasks forever* — then run that machine on a cheap model after the window shuts.

You're not renting genius for three days. You're using three days of genius to build something that runs itself.

Why now — the timeline, straight from Anthropic

Fable 5 has had a strange month. It shipped, got pulled offline under U.S. export controls, and came back — but only for a while, and only if you pay for a plan.

WhenWhat happened
June 9Fable 5 launches — Mythos-class, the tier above Opus
June 12U.S. export controls force Anthropic to disable it
July 1Controls lifted; Fable 5 restored to Pro / Max / Team / select Enterprise
Now → July 7Free — included on your plan (up to 50% of your weekly limit)
After July 7Usage credits only — $10 / 1M input, $50 / 1M output

That is the whole reason to move this week. It is genuinely the strongest model available, and for a few more days it costs you nothing beyond the plan you already have. Anthropic has said it may return to subscriptions later "as soon as capacity allows," but there is no promised date — so treat the 7th as the deadline.

How good is it, really

Anthropic reports Fable 5 as its most capable model — leading its own lineup on real-world coding benchmarks like SWE-bench Pro (80.3% vs 69.2% for Opus 4.8, on Anthropic's own runs). More useful in practice: it ships with a 1M-token context window, so it can hold an entire project in its head at once. That combination — top-tier reasoning plus whole-project memory — is exactly what makes the "build the machine" move possible.

One honest thing about *this* window

When Anthropic redeployed Fable 5, they noted that during the restored period, some routine coding and debugging may fall back to Opus 4.8. Translation: don't spend these few days having Fable write every line for you. Spend them having Fable think — plan, architect, reason over your whole project, and design systems a cheaper model can build and run. That's not a downside; it's the highest-leverage way to use it anyway.

The keeper: design an automation, not a task

Here is the exact prompt. Open Claude, set the model picker to Fable 5, then describe your most repetitive task and paste this:

I do this repetitive task over and over: [describe it in plain English].

Don't just do it once. Design the automation that does it forever.
Give me: (1) the exact step-by-step logic, (2) the decision rules and
edge cases, (3) the precise prompt or script a CHEAPER model can run
every day to reproduce your quality, and (4) how I'd trigger it on a
schedule. Build it once, at your level — so a small model can run it
after you're gone.


What you get back isn't an answer to one task — it's a reusable system: the logic, the edge cases, and a prompt or script tuned so a cheaper model reproduces Fable's quality. That system keeps running long after July 7, for free.

Three more plays worth running before the 7th

If you have window left after building your automation, these are the highest-leverage "use it now" prompts.

1. Clean up a messy folder

Go through this folder. Tell me what's junk, what's duplicated,
what's misfiled, and what's actually important. Then propose a clean
structure and, once I approve, reorganize it — moving and renaming
files so a stranger could find anything in 10 seconds. Show me the
plan before you touch anything.


2. Audit your ugliest project

Audit this project. Find what's ACTUALLY broken, not what just looks
messy. List each real issue with the file, the line, and one sentence
on why it fails — worst first. Then flag the single most dangerous
silent bug that wouldn't show up until production.


3. Save a whole project into one memory file

Read this entire project. Write ONE markdown memory file that captures
everything you'd need to continue tomorrow with zero context loss: what
this is, the architecture, key decisions and why, what's done, what's
next, and the gotchas. Write it so a fresh instance of you could pick
up exactly where we are.


## Tools & links

What most people get wrong

They treat a temporary window as a reason to do *more* — more questions, more quick rewrites. The window is short, so the right instinct is the opposite: do one thing that lasts. A cleared to-do list is gone by Monday. A working automation runs every day after. Build the thing that outlives the window.

One step to take today

Open Claude, switch to Fable 5, and run the keeper prompt on the task you repeat most. In twenty minutes you'll have a system a cheap model can run for free — and you'll have used the best model on Earth for exactly what it's best at, before it's gone on the 7th.

Enter your email to read the full how-to

Confirm via the email link to unlock this and every future SharpenDaily how-to. Free.

No spam. Unsubscribe anytime.