The field guide · Calculator

In Fable 5, the architecture is the price tag

The same day's work can cost a few dollars or a few hundred — the bill depends on how you run the model, not what you ask it. Move the sliders.

30
20,000
5,000
1
Drive it by hand
All-Fable, no caching
$0/day
$0 / month
Every task runs start-to-finish on Fable 5 at $10/$50 per MTok — decisions, thinking, and the whole bulk of the output. Parallel copies each re-read the same context and pay for it.
The laddered OS
Fable decides · cheap models execute · caching on
$0/day
$0 / month
Fable makes only the two decisions per task (run/hold + plan); a cheaper model does the typing; the stable context is cached. Same work, different bill.
2.9× cheaper
The laddered architecture saves $0/month on this workload — and nothing about the work changed.
How this is calculated (assumptions, not a promise)

Pricing per million tokens, input/output: Fable 5 $10/$50 · Opus 4.8 $5/$25 · Sonnet 5 $3/$15 · Haiku 4.5 $1/$5. Verify against platform.claude.com/docs — prices move.

All-Fable per task = copies × [context×$10 + (plan + thinking)×$50] + work×$50, no caching (the naive baseline doesn't set it up). Plan ≈ 700 output tokens; thinking tokens follow the effort setting.

Laddered per task = Fable decides+plans on cached context (repeated reads billed at ~15%) then the executor model does the work: context×0.15×$10 + (plan+thinking)×$50 + (context×0.15 + plan)×exec_in + work×exec_out. One executor, no redundant fan-out.

These are deliberately simple. Real bills vary with retries, tool calls, and cache hit rates — but the ratio is the point, and the ratio is structural.

Read the ladder in full: the routing, the caching, and the budget cap that stops spending at a hard line are Layers 00 and 06 of the nine-layer field guide — with a runnable starter kit.
← The nine-layer field guide The starter kit →