Stop Wasting Your Fable 5 Tokens

Fable 5 is back, but with half the weekly token allowance and stronger guardrails that offload your tasks to lower models. In this walkthrough I show you how to make every Fable 5 token count: a CLAUDE.md orchestration setup that routes each task to the right model, plus the official plugin that lets Claude Code delegate coding tasks straight to Codex. The full CLAUDE.md is at the bottom, ready to plug and play.

What you need

• Claude Code or Cowork, since the setup lives in a CLAUDE.md file
• Optionally, a Codex subscription for the delegation part. Even the $100 Codex plan is more than enough for a lot of use cases
• Optionally, a Gemini API key if you want to add it to the routing pool

1. Fable 5 is back: what changed (0:00)

Now that the king is back, there are two key things to understand. First, our token allowance for Fable 5 is half of what it used to be: we can only spend fifty percent of our weekly plans on Fable 5. And remember that Fable 5 already consumes twice as many tokens as Opus 4.8. That combination means usage is extremely limited.

2. Stronger guardrails: your tasks get offloaded (0:34)

Second, Anthropic increased the guardrails on the model. Fable 5 now has the tendency to offload any task you give it to lower models such as Opus 4.8, Sonnet 5, and others. The goal is to keep you from hitting your rate limits, but it also means that without direction, the model you wanted is not always the model doing the work. So we need to do two things: avoid burning the allowance, and make sure Fable 5 actually handles the tasks where it is irreplaceable.

3. The fix: routing rules in CLAUDE.md (1:17)

Fable 5 is amazing at taste and architecture. So the fix is to create a list of task types, revising a plan, creating a UI, and so on, and explain to Fable 5 how to route each request: which models it has available and how to select them.

That lives in the CLAUDE.md. For people working with Cowork or Claude Code, CLAUDE.md is the file that personalizes how your Claude behaves. Tell it which tasks belong to Fable 5 and which belong to a model like Opus 4.8, and the routing happens automatically from then on.

4. Fable 5 as orchestrator, plus the effort setting (2:19)

With that file in place, Fable 5 works as the orchestrator. It stays the main model, and we set the effort to at most high. Effort controls how long the model thinks, and because Fable 5 tokens are so precious, we do not want it burning them on long thinking.

Give the orchestrator access to other models, Sonnet, Opus, even Gemini via an API key, and push all of the token-hungry tasks, computer use, reading images, and the like, onto those models. They do the heavy lifting; Fable 5 does the reasoning, some coding, and the decisions. In practice, fifty to sixty percent of your tasks will not be done by Fable 5. They will be orchestrated by it.

5. Orchestrating Codex from Claude Code (3:38)

The second big lever is orchestrating Codex directly from Fable 5. Codex, and here I mean the GPT 5.5 version, is extremely powerful for coding. If you have both subscriptions, combine them: Fable 5 is absolutely amazing at UI and fantastic at UX, and Codex is incredible at back end. Together you get extremely nice UI with Codex's capacity behind it, the best of both worlds before GPT 5.6 lands.

6. The codex-plugin-cc plugin (4:02)

The bridge is an official plugin that gives your Claude Code access to Codex:

https://github.com/openai/codex-plugin-cc

It adopts your user-level configuration from ~/.codex/config.toml automatically, and instead of hand-rolled bash wrappers you get built-in commands for reviews, rescues, and background jobs, all listed in the CLAUDE.md below.

7. The plug-and-play CLAUDE.md (4:48)

Here is the full section from the video. Copy it into your own CLAUDE.md and replace the parts that do not fit your setup. Everything in it came from researching what people on X are doing to get the most out of this model.

Picking the right models for workflows and subagents

Rankings, higher = better. Cost reflects what I actually pay, not list price. Intelligence is how hard a problem you can hand the model unsupervised. Taste covers UI/UX, code quality, API design, and copy.

| model | cost | intelligence | taste |
|---|---|---|---|
| gpt-5.5 | 9 | 8 | 5 |
| sonnet-5 | 6 | 5 | 7 |
| opus-4.8 | 4 | 8 | 8 |
| fable-5 | 2 | 9 | 9 |

How to apply:
- These are defaults, not limits. You have standing permission to override them: if a cheaper model's output doesn't meet the bar, rerun or redo the work with a smarter model without asking. Judge the output, not the price tag. Escalating costs less than shipping mediocre work.
- Cost is a tie-breaker only; when axes conflict for anything that ships, intelligence beats taste beats cost.
- Bulk/mechanical work (clear-spec implementation, data analysis, migrations): gpt-5.5 - it's very cheap and token efficient.
- Anything user-facing (UI, copy, API design) needs taste of 7 or higher.
- Reviews of plans/implementations: fable-5 or opus-4.8, optionally gpt-5.5 as an extra independent perspective.
- Never use Haiku.
- Mechanics: gpt-5.5 is handled natively via the openai/codex-plugin-cc plugin inside Claude Code, automatically adopting your user-level configurations from ~/.codex/config.toml. Avoid writing custom bash scripts; instead, utilize the plugin's built-in tools and skills:
  - /codex:review - Run non-destructive, read-only code quality assessments. Supports --base REF for branch analysis.
  - /codex:adversarial-review - Perform a skeptical design review to pressure-test tradeoffs, auth, and reliability. Append custom focus text at the end of the command to steer the focus.
  - /codex:rescue - Subcontract active debugging, multi-file refactoring, or implementation loops to Codex when a second pass is required.
  - /codex:status / /codex:result / /codex:cancel - Use these to check, fetch, or abort asynchronous jobs when using the --background flag on heavy tasks.
- Claude models (sonnet-5, opus-4.8, fable-5) run via the Agent/Workflow model parameter.

Using gpt-5.5 inside workflows and subagents:
- Subagents and automated workflows should call the plugin's native slash commands or its exposed codex-cli-runtime skills to delegate tasks directly, omitting the need for raw terminal wrappers.
- For closed-loop quality assurance, keep the review gate turned on via /codex:setup --enable-review-gate. This ensures a stop hook automatically challenges Claude's outputs using Codex before finalizing, preventing broken code or weak design assumptions from reaching the main session unvetted.

Pitfalls and tips

• Fable 5 consumes twice as many tokens as Opus 4.8 on top of the halved allowance. Budget for both effects, not just one.
• Cap the orchestrator's effort at high so precious tokens are not spent on extended thinking.
• Keep token-hungry work like computer use and reading images away from Fable 5; route it to the cheaper models in your pool.

Where to go next

For what happens when GPT 5.6 arrives and how the pairing evolves, watch GPT 5.6 Drops Tomorrow: Everything We Know. And to master the tool on the other side of this delegation, see Master Codex Desktop in 30 Minutes.