Get the Claude Code skills and workflows I use to automate my content and my business in the AI Academy community on Skool.

Why this setup

On Anthropic's FrontierCode 2.0 test, Fable 5.2 at High effort scored 45.3%. Opus 5.5 at Max effort scored 45.5%. Anthropic says Fable 5.2 can also play the role of orchestrator, "where it delegates simpler tasks to smaller models".

Price per million tokens:

  • Fable 5.2: $8 input, $40 output.
  • Opus 5.5: $4 input, $20 output.

So each Opus worker costs half as much per token as Fable.

Setup guide

Fable 5.2 needs Claude Code v2.1.321 or later. On Max, Fable is included in the plan. On Pro and standard Team seats, Fable can use up to 50% of the weekly limits until Nov 16. On some plans and seats, Fable usage bills to usage credits. Then the /model picker shows "Requires usage credits" on the Fable row.

  1. Update Claude Code.

    claude update
    
  2. Make Fable 5.2 the main model. In a session, run:

    /model fable
    

    Or start the session on it:

    claude --model fable
    
  3. Set the effort level. Fable 5.2 starts on medium. For the boss, set high:

    /effort high
    
  4. Make an Opus worker subagent. A subagent is a Markdown file with YAML frontmatter. Put it in ~/.claude/agents/ for all your projects, or in .claude/agents/ for one project. Only name and description are required. Set model and effort so the worker does not copy the Fable session.

    ---
    name: worker
    description: Builds one task from the plan and reports back what changed
    model: opus
    effort: medium
    ---
    
    You are a worker. Do the one task you are given, test it, and report
    what you changed and what is still open.
    

    You can also ask Claude to write this file for you. In the current Claude Code, /agents only prints a reminder to do one of these two things. Claude Code sees a new or changed file within a few seconds. If the agents folder did not exist when the session started, restart Claude Code once.

  5. Give Fable the job and tell it to use the worker. For example:

    Plan this feature, then use the worker subagent for each task. Check each result before you hand out the next one.
    

    To force that subagent, @-mention it: @"worker (agent)". For a full conductor prompt, see Ruben Hassid's post in the links.

  6. Keep one model per chat. If you want Opus to build in its own chat, get the plan from Fable first, then start a new Opus chat and paste the plan in.

Watch out

  • The test is Fable alone against Opus alone. FrontierCode does not measure the Fable plus Opus team. Nobody has published a score for this setup.
  • Subagents copy the main model. If a subagent has no model field, it can run on Fable too. Set model: opus, or model: claude-opus-5-5 to be exact.
  • Switching models costs you. The docs say: "Each model has its own cache." After /model, the next request reads the whole chat with no cache hits.
  • Opus cannot read Fable's thinking. When you move from Fable 5.2 to another model, that model loses Fable's reasoning for those turns.
  • Fable uses your limits fast. Watch your usage, and keep the simple work on the Opus workers.

Sources: Anthropic: Claude Fable and Mythos 5.2, Claude Code subagents, Claude Code model configuration, Claude Code prompt caching, What's new in Fable 5.2, Claude help center: Fable access and Ruben Hassid, How to AI