Claude Fable 5 came back online on 1 July after an 18-day export-control suspension. Subscribers get it through 7 July at half their weekly limits, then access moves to metered credits at $10 per million input tokens and $50 per million output tokens. That short runway changes the question. It is not whether Fable is impressive. It is what the model can do that Opus or GPT cannot. Five jobs stand out.
It can pick its own hardest problems
Point Fable at your own memory and files and it will rank the work that rewards deep thinking. In one walkthrough it surfaced a strategy stress-test, a premium-tier offer, and a content-to-conversion audit across every platform. The value is the triage, not the answer. It reads a large corpus and tells you where to aim.
It gives strategic advice that argues with itself
Feed it a plan document covering goals, positioning and finances, wire it to your live tools, then ask for a three-month focus. Fable can convene a panel of personas, a target customer, a sceptic and an operator, and let them fight over the recommendation before it lands on your desk. Adversarial self-review is hard to get from a smaller model.
It hardens half-built apps before launch
This is the high-leverage one. Point Fable at a vibe-coded project and it hunts the bugs other models miss. In one demo on a fitness app that looked polished, it found more than 12, including a data leak that surfaced on sign-out. That single catch is the difference between a clean launch and a public mess.
It plans features for a cheaper model to build
Use the expensive brain for the thinking, the cheap one for the typing. Fable drafts the architecture, picks the database, lays out risks and open questions, and can sketch a proposed design. Then it hands a clean brief to Opus or GPT to implement. You get frontier-grade planning without paying frontier rates for every line of code.
It untangles code no smaller model can hold
Refactoring across a sprawling repo defeats lesser models because they lose the thread halfway through. Fable holds context across the whole thing. Stripe ran a codebase-wide migration on a 50-million-line Ruby codebase in a day, work it reckoned would have taken a team more than two months by hand. Most repos are far smaller, but the principle scales down.
The through line is simple. Fable earns its keep on work that needs it to hold everything at once, plan, audit, refactor, pressure-test. With six days on the clock, the real waste is pointing it at the emails and blog posts a cheaper model would finish just as well.
