One line at the end of your prompt hands the busywork to smaller models, so the smart one works where it matters. Included with Max up to half your weekly limit. On Pro, Fable runs on usage credits.
Fable is the strongest model Claude has ever shipped, and how you pay for it depends on your plan. On Max (and some Team and Enterprise seats), it's included, up to 50% of your weekly limit, drawing from the same pool as everything else. On Pro, it runs on usage credits, pay as you go, from the first message. Either way the catch is the same: by default Fable does everything itself, the hard thinking and the filing, and the filing adds up.
Here's the picture. Fable is the most expensive expert on your team. You would not pay your top expert to rename files and sort folders. You'd hand that to the assistant. The two lines below tell Fable to do exactly that: keep the hard thinking, pass the busywork to smaller models. Same result, and your credits last the week.
For regular work in Claude: research, writing, planning, reports. It gives the model permission to offload the simple parts when that makes sense.
For simple tasks, use your best judgment to hand the work to a smaller model.
For coding and building work. This one is concrete: Claude Code has built-in subagents with their own separate context, and each one can run on a smaller, cheaper model like Haiku. This line tells Fable to route them that way.
For all coding tasks, use your best judgment to assign subagents a lower power model appropriate for that task.
Think of Fable as your strategist. It makes the hard calls, and the cheaper helpers handle the search, the sorting, and the repeat steps. Claude can run those helpers on smaller models while the main model directs the work. Without instructions, it often uses the big model for everything. One sentence of permission fixes that.
Delegation shines when your request mixes hard thinking with obvious grunt work. It can hurt when one brain needs to hold the whole problem.