What just happened: Anthropic quietly benched Claude Fable 5 for 18 days under an export-control directive. On July 1, it came back. Through July 7, it's included in your paid plan (up to 50% of weekly usage). Starting July 8, it moves to metered usage credits at $10 in / $50 out per million tokens. The most expensive model Anthropic sells. If you don't want a surprise bill, this list is for you.
[001]
USE FABLE FOR THIS
The hard stuff. Worth the meter.
Deep reasoning. Pricing strategy, business model, high-stakes decisions with real tradeoffs.
Long-horizon agents. Multi-step tasks that need to hold state across many turns.
Complex code. Full-stack builds, gnarly refactors, debugging you've been stuck on.
"I need the best possible answer right now." The ONE question you can't afford to get wrong.
[002]
ROUTE THIS ELSEWHERE
Cheaper models handle it fine. Save Fable tokens for the hard stuff.
Everyday chat, research, first drafts → Sonnet
Quick summarization + reformatting → Sonnet or Haiku
Simple code edits + code reviews → Sonnet
Emails, brainstorming, meeting prep → Sonnet
Fast lookups + factual questions → Haiku
Rule of thumb: if you can already predict the shape of the answer, don't burn Fable tokens.
[003]
THE ONE SETTING TO CHECK TODAY
Go check your DEFAULT MODEL right now. If it's set to "Fable" or "Auto," every chat is going to hit Fable rates after July 8.
Fix it in 20 seconds:
Open Claude Cowork (or Claude web/desktop)
Profile → Settings → Models
Set default to Sonnet (or Opus for a middle-ground premium)
Save
When you need Fable specifically, switch manually per chat
> PAGE 2 // PRO MOVES_
Bonus tactics + a pro contribution from Jonathan Malkin.
[004]
5 HARDEST PROBLEMS TO HAND FABLE THIS WEEK
While it's included through July 7, use it on the projects you've been putting off.
"Here's my pricing page + 3 competitors. Rewrite mine to convert 2x."
"Here's my entire codebase. Find the 5 biggest tech debts and give me a prioritized fix plan."
"Here's my financial model + 12 months of P&L. Where should I invest next?"
"Here's my org chart + last quarter's OKRs. What's misaligned?"
"Here's the messy transcript from my last customer call. Write a positioning brief for me."
[005]
PRO ROUTING
// from Jonathan Malkin, instructor of Monday's Build Your Own AI Assistant workshop
Fable is token-hungry. Make it the conductor, not the worker. Paste this into Claude:
Read my CLAUDE.md and add a short "Model and Delegation Routing" section. Fable is the conductor, not the worker. Route: Fable = strategy, judgment, approval, synthesis, review. Sonnet = bounded writing, docs, ordinary implementation, debugging. Haiku = inventory, grep, extraction, formatting. Codex (if available) = large implementation, repo investigation, test loops, bulk edits. Cost is a tie-breaker only. Keep the edit short. Preserve existing local rules and approval gates. Always High Effort.
If you use Codex: verify with codex --version and codex exec --help. For handoffs, use stdin (- < /tmp/task.md) to avoid shell-quoting pain.