Anthropic's Fable 5 model is a powerful but expensive tool, costing twice as much as Opus 4.8. The model's high intelligence makes it prone to "overdoing it," leading to unnecessary token burn on routine tasks. However, there is a proven way to cut costs by more than 50% — the "10-80-10" method, actively used by the company's own engineers.

The Essence of the Approach: Only Where It's Truly Needed

The idea is simple: Fable 5 should only be used for tasks requiring its unique level of intelligence. All other work is delegated to cheaper models. This method is already being called a "cheat code" for using Fable 5 without financial loss.

The first 10% of the work is dedicated to planning. At this stage, Fable 5 acts as an architect: it defines the structure, approach, success criteria, and constraints. Here, the model doesn't generate code but creates a clear plan — if it's correct, everything else is built on a solid foundation.

The main 80% is execution. Routine tasks, iterations, and minor fixes are handed off to more affordable models: Opus 4.8 for standard tasks, Haiku for light ones, and in some cases, even GPT-5.5 will suffice. It is at this stage that the most tokens are burned, so paying double the price for Fable here is simply uneconomical.

The final 10% is verification. Fable 5 returns to check the result against the original architecture: does it match the plan, and are there any gaps? Since this is a verification of a completed solution, not generation from scratch, token consumption is minimal here.

New Syntax and Automation

Instead of the usual "request — response — check — re-request" cycle, where the user acts as the loop, it is suggested to set a goal with the /goal command and configure an automatic loop via /loop. After that, the model itself launches subtasks, significantly simplifying the process.

Infographic on the token saving system for Fable 5
The 10-80-10 system for reducing costs on Fable 5: planning, execution, and verification.

Additional Saving Tips

It's best to start not with the maximum effort level, but with the medium one: in medium mode, Fable 5 surpasses Opus in high mode. It is also recommended to delete old skills created for previous models — short instructions are cheaper. It's useful to give the model not only the task but also its reason: this reduces the number of edits and, consequently, token consumption. The /usage command helps control spending.

Special attention should be paid to costly mistakes. The first is accidentally launching Fable: Claude Code and the app now use this model by default, so the selector should be checked every session. The second is the absence of a spending limit. Fable will soon switch to pay-per-token, so it's worth setting strict limits in the settings in advance.

Expert opinion from Cryptalist: The 10-80-10 method is not just a trick for saving money, but a fundamental principle for the efficient use of AI resources. In a context where the cost of advanced models is rising, this approach becomes critically important not only for individual developers but also for entire companies seeking to optimize their AI budgets. I recommend integrating this strategy into your workflow today — tomorrow it could become an industry standard.