Free cheat sheet · Build with AI

Three ways to route Claude

Same Claude, three habits. Here's what each one sends where, and how to set up the third one in Claude Code in two commands.

By Frontpage AI · Checked against Anthropic's docs on · From our Reel "Three people. Same Claude."

The short answer

Start on Opus 5.5, as Anthropic advises. For long, many-step work, let Opus plan, Sonnet 5.5 build and Fable 5.2 advise: in Claude Code that's /model opusplan then /advisor fable (or /advisor opus if you don't have Fable). Most of the work then runs on Sonnet, at half Opus's price and a quarter of Fable's.

Routing diagram. The beginner sends everything to Fable. The improver uses Opus, and Fable for hard jobs. The pro has Opus plan, Sonnet build, and Fable advise.123Sonnet 5.5buildsOpus 5.5plansFable 5.2advisesadvisor
Blue is number three's setup. Grey is the other two.

What are the three ways to route Claude?

Beginner

Sends everything to Fable 5.2

Typos, emails, renaming files and big rebuilds all go to the newest, most powerful model.

Pays the top price on every job, even the small ones.

Improver

Starts on Opus 5.5, steps up to Fable for the really hard jobs

This is Anthropic's own advice: start with Opus 5.5, "our strongest daily driver", and switch up to Fable as needed.

Pro

Opus plans, Sonnet builds, Fable advises

Opus 5.5 writes the plan. Sonnet 5.5, at half Opus's price, does the work. When the work gets stuck, it asks Fable for advice, then carries on.

Most of the work runs on the cheapest of the three.

Which setup should you pick?

Anthropic's starting point is simple: start on Opus 5.5, then move up or down depending on the job.

Short jobs: a quick fix, an email, a question

Number 2

Stay on Opus 5.5. Anthropic says an advisor adds less value on short tasks where there's little to plan.

Long, many-step work: a refactor, a new feature, a bug that keeps coming back

Number 3

This is where the advisor fits best: most steps are routine, but the quality of the plan decides how it turns out.

The hardest, most open-ended problem you have

Fable for the whole job

Fable is best for "the most ambitious, open-ended technical work". Switch with /model fable, then switch back when the hard part is done.

Lots of simple, repeated jobs: sorting, short summaries

Go smaller

Anthropic suggests moving down to Sonnet 5.5 or Haiku 5.5 when the job allows it.

How do you set up number three in Claude Code?

  1. Plan with Opus, build with Sonnet

    /model opusplan

    Claude Code uses Opus 5.5 while you're in plan mode, then switches to Sonnet 5.5 to do the work.

  2. Add Fable as an advisor Experimental

    /advisor fable

    Claude now asks Fable 5.2 for guidance at the hard moments: before it commits to an approach, when an error keeps coming back, or before it calls a job done. Claude decides when to ask.

    • Works with the Anthropic API or Claude Platform on AWS, not Amazon Bedrock, Google Cloud or Microsoft Foundry.
    • You need access to Fable. On some plans, Fable bills to usage credits: run /model fable once to accept that, then set the advisor.
    • Turn it off with /advisor off.

    No Fable? Use Opus as the advisor

    The advisor doesn't have to be Fable. Opus 5.5 is accepted as the advisor for both Sonnet 5.5 and Opus 5.5, so it works with opusplan too. It needs no Fable access or usage credits, and it costs half as much per token as Fable.

    /advisor opus

    Pick Fable when you want the strongest second opinion on the hardest decisions. Pick Opus for a cheaper one that's still a step up from Sonnet.

  3. Need Fable for the whole job?

    /model fable

    Switch back with /model opusplan when the hard part is done.

How do you make it your default?

The commands above last for the session. To start every session this way, add two lines to your Claude Code settings file, ~/.claude/settings.json. If the file already has settings, add these two lines inside its braces.

~/.claude/settings.json
{
  "model": "opusplan",
  "advisorModel": "fable"
}
  • Using Opus as the advisor? Put "advisorModel": "opus" instead.
  • Want a second opinion on a particular step? Ask for it in plain words: "consult the advisor before you continue".
  • Toggling the advisor on or off mid-session doesn't reset Claude's prompt cache, so it won't make the next reply slower or dearer.

Which advisor pairings work?

The advisor must be at least as capable as the main model. These are the pairings Anthropic suggests:

Main modelAdvisorWhen to use it
SonnetFableFable's guidance at the decision points, without running Fable throughout. Needs Fable access.
SonnetOpusSonnet does the routine work and hands planning, confusing failures and the final check to Opus.
HaikuOpusThe cheapest main model with strong planning. Costs more than Haiku alone, less than running Sonnet or Opus.
OpusOpusA second Opus reviews the first. For high-stakes work where an independent check matters more than cost.
FableFableThe most capable pairing. Claude Code won't add an Opus or Sonnet advisor to a Fable main model.
SonnetSonnetA cheaper second opinion that catches routine slips.

How do you keep helpers on cheaper models?

Claude Code can hand side jobs, like running tests or searching files, to helpers called subagents. By default they use your main model, so if you switch to Fable, your helpers run on Fable too. To keep a helper on a smaller model, give it its own file in your project's .claude/agents/ folder:

.claude/agents/test-runner.md
---
name: test-runner
description: Runs the test suite and reports which tests failed and why.
model: sonnet
---

Run the project's tests. Report each failure with the file, the test name and the error.

Should you turn the effort down before you trade up?

Yes. Each model has an effort setting: how hard it thinks before it answers. Higher effort is slower and uses more. Fable 5.2, Opus 5.5 and Sonnet 5.5 all start on medium in Claude Code. Type /effort to change it.

EffortAnthropic's advice for Fable 5.2
lowQuick back-and-forth, well-specified tasks, lots of simple jobs.
mediumInteractive coding, chat and everyday work.
highHard coding, research and long unattended runs. "Most of the gain in capability is already present at high."
xhighOnly your hardest long-running work, where you've measured a gain.
maxMeasuring the model's ceiling, or the rare job where time and cost don't matter. It can overthink.

Don't tell it to "think less"

On Fable 5.2, lines like "answer without deliberating" no longer stop it thinking at medium and above. Lower the effort instead.

Check what you've spent

Type /usage to see the session's totals. Advisor calls are included, so you can see what each setup really costs you.

How much does each Claude model cost?

Claude API prices in US dollars per million tokens, checked 11 November 2026
ModelInputOutput
Fable 5.2$8$40
Opus 5.5 half of Fable$4$20
Sonnet 5.5 a quarter of Fable$2$10
Haiku 5.5 prompts up to 100,000 tokens$0.10$0.50

API prices in US dollars per million tokens. Haiku 5.5 is $0.50 input and $2.50 output above 100,000 tokens. On a Claude subscription, use comes out of your plan's limits instead, and on some plans Fable uses usage credits.

Questions people ask

How do I make Opus plan, Sonnet build and Fable advise in Claude Code?

Run /model opusplan, then /advisor fable. Opus 5.5 writes the plan in plan mode, Sonnet 5.5 does the work, and Claude asks Fable 5.2 for guidance at the hard moments. To start every session this way, set "model": "opusplan" and "advisorModel": "fable" in ~/.claude/settings.json.

Can I use Opus as the advisor instead of Fable?

Yes. Run /advisor opus instead of /advisor fable. Opus 5.5 is accepted as the advisor for both Sonnet 5.5 and Opus 5.5, so it works with opusplan. It needs no Fable access or usage credits, and it costs half as much per token as Fable. Choose Fable when you want the strongest guidance on the hardest decisions.

Which Claude model should I use by default?

Anthropic's advice is to start on Opus 5.5, which it calls its strongest daily driver, then move up to Fable 5.2 for the most ambitious, open-ended technical work, or down to Sonnet 5.5 or Haiku 5.5 when the job allows it.

Can I use Opus as the advisor for a Fable main model?

No. The advisor must rank at or above the main model, so a Fable 5.2 main model only takes Fable 5.2 as its advisor. Claude Code will not add an Opus or Sonnet advisor to a Fable main model.

Does the Claude Code advisor work on Amazon Bedrock?

No. The advisor is experimental and works with the Anthropic API or Claude Platform on AWS. It is not available on Amazon Bedrock, Google Cloud or Microsoft Foundry.

Does turning the advisor on or off reset the prompt cache?

No. Toggling the advisor mid-session keeps your main model's prompt cache, so it won't make the next reply slower or dearer. Each advisor call does read the whole conversation at the advisor model's rates.

How much do Fable 5.2, Opus 5.5, Sonnet 5.5 and Haiku 5.5 cost?

On the Claude API, per million tokens: Fable 5.2 is $8 input and $40 output, Opus 5.5 is $4 and $20, Sonnet 5.5 is $2 and $10, and Haiku 5.5 is $0.10 and $0.50 for prompts up to 100,000 tokens. On a Claude subscription, use comes out of your plan instead.

Sources

Model names, prices and commands change. Check the sources before relying on them. Last checked .

Run a business? See what AI says about you.

Frontpage AI builds websites for independent UK service businesses that ChatGPT, Gemini and Google can read and recommend. We run a free 15-minute AI check, whether or not you become a partner.

15 minutes. No obligation. An honest answer either way.