The Claude Fable 5 Setup Guide: How to Get Maximum Quality for Minimum Cost (Exact Config Inside)

Fable 5 is back, and until July 7 you get up to 50% of your weekly limit on Anthropic's most powerful model, free.

Inside: which heavy tasks actually deserve Fable 5, the new safeguards that reroute you to Opus, and how to plan the window.

Use it right and you'll clear weeks of work before it flips to paid credits.

Here's the exact config you need 👇

Before we dive in, I share daily notes on AI & vibe coding in my Telegram channel: https://t.me/zodchixquant🧠

What actually happened, in plain terms

Fable 5 is Anthropic's most powerful publicly released model. The quick timeline:

• June 9 — launched.
• June 12 — pulled under a US export-control order tied to a jailbreak report.
• July 1 — back online after the controls were lifted.

That whole saga isn't what matters for your week. This is what matters:

For Pro, Max, Team, and select Enterprise plans, Fable 5 is included for up to 50% of your weekly usage limit through July 7. After that, it moves to usage credits, which means it starts costing you.

So you have a short window where the most capable Claude is effectively free, up to half your weekly cap.

The move is to spend that half on the work that's too big or too slow for a normal model, not on quick questions you could hand to Sonnet 5.

Why Fable 5 is worth planning around

This isn't a cheaper, faster model. It's the heavy one.

The clearest example is Stripe: they used Fable 5 to compress a codebase-wide migration across 50 million lines of Ruby into a single day, work they estimated would take a team more than two months by hand.

That's the kind of task the window is for. When a model can turn two months into one day, the question stops being "is it worth the tokens" and becomes "what's the biggest thing I can throw at it before July 7."

The new safeguards, and what they mean for you

Fable 5 came back with tighter cybersecurity safeguards. Know how they behave so you're not surprised mid-task:

• A new classifier targets one specific jailbreak technique and blocks it in 99%+ of tries.
• Blocked requests reroute to Opus 4.8, the less capable model, and you're told it happened.
• The trade-off is more false positives on normal coding and debugging. Legitimate security-adjacent work may get rerouted even when your intent is fine.

Anthropic says it's still refining the filter to cut false positives. For now, just expect the occasional reroute on security-flavored tasks.

One more thing: Fable 5 carries a 30-day data retention policy, used to research and mitigate jailbreaks. Factor that into anything sensitive.

The routing formula (copy this)

One line to decide, every time:

Weeks by hand → Fable 5. Minutes by hand → Sonnet 5.

If a task would take you two weeks or more by hand, it earns Fable 5 and the free window. If it's a quick edit, it stays on Sonnet 5 so you protect your 50% cap for the big stuff.

Drop this straight into your CLAUDE.md so the choice is automatic:

1## Model routing (through July 7)
2Default: Sonnet 5 for everyday work.
3
4Use Fable 5 only for heavy, high-payoff tasks:
5- large migrations (framework, language, dependency)
6- codebase-wide refactors across many files
7- complex multi-step builds
8- hard bugs in tangled code (race conditions, subtle state)
9
10Rule of thumb: weeks-by-hand → Fable 5, minutes-by-hand → Sonnet 5.
11Protect the Fable 5 window. Don't spend it on small edits.
12
13Note: some security-adjacent requests get rerouted to Opus 4.8
14by the new safeguards. If quality drops on one call, check for a reroute.

The big-job kickoff template (copy this)

When you point Fable 5 at a large migration or refactor, don't just say "refactor this."

Give it the whole job up front so it can hold the plan end to end.

Copy and fill in:

1# Task: [migration / refactor name]
2
3## Goal
4[One sentence: the end state you want.]
5
6## Scope
7- Files/dirs in scope: [paths]
8- Out of scope: [paths to leave alone]
9
10## Constraints
11- Keep all tests passing.
12- Preserve public APIs unless listed below.
13- Match existing code style.
14
15## Plan first
16Before changing anything, map the full plan and list the files
17you'll touch. Wait for my go-ahead, then execute in one pass.
18
19## Definition of done
20- [ ] [check 1]
21- [ ] [check 2]
22- [ ] all tests green

The "plan first, then execute in one pass" line is what turns Fable 5 from a fast typist into the thing that cleared Stripe's 50-million-line migration in a day.

Common mistakes

Treating Fable 5 like a chat model. It's the heavy machinery. Asking it quick questions all day burns your included capacity on work Sonnet 5 handles fine.

Missing the July 7 date. After the window, Fable 5 moves to usage credits. The same big job costs real money in a week. Timing is the whole advantage.

Assuming every request goes through. The new safeguards reroute some coding and security tasks to Opus 4.8. If output quality suddenly drops on one request, check whether you got rerouted.

Spreading the 50% too thin. Half your weekly limit sounds like a lot until you fritter it on small stuff. Concentrate it on a few high-payoff jobs.

The 15-minute plan

3 minutes: confirm Fable 5 is live on your plan (Claude.ai, Claude Code, or Cowork).

5 minutes: pull your 3 biggest pending jobs, migrations, refactors, or complex builds.

5 minutes: start the top one on Fable 5 now, while it's included.

2 minutes: set a reminder for July 7, when it flips to paid credits.

The window is short and the model is the most capable one Anthropic ships. Point it at the work that used to take weeks, and you'll get more done in these few days than most people will all month.

Thanks for reading!