# Your Complete Guide to Fable 5.2 (When to Use It, Effort Levels, Prompting)

## Executive summary

This guide details the strategic use of Claude Fable 5.2, positioning it as a highly capable but expensive model. The core recommendation is to adopt a 'Plan with Fable, Build with Opus' workflow. Use Fable 5.2 for complex, open-ended tasks like architecture planning, research, and decision-making, and then transition to the more cost-effective Opus 5.5 for executing well-defined, concrete steps and implementation. This split approach maximizes performance while managing costs.

## Key takeaways

- The Plan/Execute Model: Use Fable 5.2 for the initial planning, searching, and making big decisions (the 'planner'). Once a detailed handoff document (e.g., `plan.md`) is created, switch to Opus 5.5 (the 'workhorse') to handle the implementation, validation, and iteration, as this is where the task is well-defined.
- Cost and Efficiency Management: Fable 5.2 is significantly more expensive than Opus 5.5 (e.g., Max reasoning cost was $26.54 vs. less than half the cost for Opus 5.5). For simpler or well-defined tasks, Opus 5.5 is often 'good enough,' making the split approach economically necessary.
- Optimal Effort Levels: For Fable 5.2, 'High' reasoning is generally recommended, as 'Max' effort is rarely worth the cost increase. For Opus 5.5, 'Medium' effort is often sufficient for simpler tasks, especially when the plan is scoped out by Fable.

## Technical details

- Prompting Strategies for Fable 5.2: To improve results, explicitly prompt Fable 5.2 to take ownership (e.g., 'You are the owner and long-term maintainer of this codebase'). Also, set sub-agents to use a cheaper model (like Sonnet or Opus) rather than Fable 5.2 to prevent unnecessary token expenditure. Focus prompts on the desired outcome, not specific steps.
- Workflow Orchestration: Implement workflows using tools like Archon (an open-source harness builder) or by manually creating a handoff document. The process involves: 1) Fable 5.2 for planning/context loading/web search; 2) Creating a handoff document; 3) Starting a new session and switching the model to Opus 5.5 for implementation.
- Claude Code Tools: Within Claude Code, the `/advisor` tool can be enabled to allow the workhorse model (e.g., Opus 5.5) to periodically advise using the more capable LLM (Fable 5.2). Custom sub-agents can have their model set in the YAML front matter.

## Practical implications

- Build engineers should design agentic workflows that explicitly separate the 'planning' phase (Fable 5.2) from the 'execution' phase (Opus 5.5) to optimize cost and performance.
- When building CI/CD pipelines or complex agents, ensure that sub-agents are configured to use the most cost-effective model necessary for their specific task (e.g., Sonnet or Opus for context loading, not Fable 5.2).
- Before deploying agents, audit and prune unnecessary rules or context files (`claw.md`) to prevent token waste and potential constraint on the LLM's capabilities.

## Topics

LLM Orchestration, Agentic Workflow, Claude Code, Fable 5.2, Opus 5.5, Prompt Engineering, Cost Optimization, Firecrawl, Dynamous, Archon

Source: https://www.youtube.com/watch?v=fLjGgbW0Ki8
