Run all of this inside the full Agent OS — join the AI Profit Boardroom →

agentos.guide › blog

Fable 5.2 vs Opus 5.5: What Separates Them and Three Prompts to Route Work in Your Agent OS

By Julian Goldie · 18 November 2026 · agentos.guide

A golden planner node directing smaller glowing worker nodes over dark aubergine

Fable 5.2 vs Opus 5.5 is the big Claude question for anyone running an Agent OS, and the answer is to use Opus 5.5 as your everyday engine and promote the hard, long jobs to Fable 5.2.

Book a free SEO strategy session →

This guide covers the differences that matter for agents, then gives you prompts that route work and keep costs down.

If you would like agents working on your website's organic growth, book a free SEO strategy session.

Facts are from Anthropic's Fable 5.2 page, Opus 5.5 page and launch post.

What separates them

Fable 5.2Opus 5.5
Released9 Nov 202622 Sep 2026
Default efforthighmedium
Price per tokenDoubleBaseline
FrontierSWE v269.9%62.3%
FrontierCode 2.049.1%45.5%
Best rolePlanner, advisor, long jobsDaily driver

Both have a 1M-token context window, 128K output, adaptive thinking that cannot be switched off and a June 2026 knowledge cutoff.

Why Fable 5.2 can cost less per job

Anthropic reports that Fable 5.2 at High effort matched Opus 5.5's best FrontierCode score for roughly half the cost per task.

That happens because Fable needs fewer tokens to reach the same answer.

So test Fable at a lower effort before you rule it out on price.

🔥 Want your Agent OS routing between Opus and Fable for you?

Inside the AI Profit Boardroom you get step-by-step video trainings, the installable Agent OS, four coaching calls a week and 3,400+ members testing the newest models on real business work.

→ Join the AI Profit Boardroom

Prompt 1: the escalation router

For every task, start on claude-opus-5-5 at effort "medium".
Escalate to claude-fable-5-2 at effort "high" only if:
- the tests still fail after two attempts, or
- the task touches more than 10 files, or
- I mark it as "big".
Tell me in one line which model you used and why.

This keeps most work on the cheaper model and spends Fable 5.2 where Anthropic says it shines: long, ambiguous, multi-step jobs.

Prompt 2: Fable plans, Opus builds

Use claude-fable-5-2 as the planner. Ask it to break [project]
into numbered steps with acceptance checks. Then run each step on
claude-opus-5-5 at effort "medium". After every three steps, ask
Fable to review the work against the checks.

Anthropic describes this orchestrator role directly: Fable 5.2 delegates simpler tasks to smaller models.

Book a free SEO strategy session →

Prompt 3: finish the whole task

Complete the entire task before you stop. Do not end your turn
with a list of remaining steps or an offer to continue. If you hit
an obstacle, try two different approaches before asking me.

Anthropic's Fable 5.2 notes say the model sometimes ends long turns early at low and medium effort, and its prompting guide recommends saying plainly that the whole task should be finished.

📞 Want Claude working on your search traffic?

Book a free SEO strategy session and we will show you how agents fit into your SEO and AI search plan.

→ Book your free strategy session

Settings that trip people up

Prompt 4: the cost log

After every task, add one line to costs.md with: date, task name,
model used, effort level, input tokens, output tokens, and whether
the result passed review first time. Once a week, summarise which
model gave the most passes for the fewest tokens.

A week of this log tells you more about Fable 5.2 vs Opus 5.5 in your own Agent OS than any benchmark table.

Look for jobs where Fable passes first time and Opus needs a second attempt; those are the ones worth escalating.

Prompt 5: subagent limits

You may use up to 4 subagents at once. Run every subagent on
claude-opus-5-5 at effort "medium" unless the step is marked "hard".
Before starting more than 4, stop and tell me why.

Anthropic's Fable 5.2 notes say the model delegates readily when the harness allows it, and that subagents can be a large share of a session's cost.

A clear cap keeps a planner on Fable 5.2 from spending your budget on a swarm of expensive helpers.

Effort defaults in Claude Code

Anthropic sets Fable 5.2 to medium effort by default in Claude Code and the Claude apps, while the API default is high.

Book a free SEO strategy session →

If your Agent OS calls the API directly, set the effort yourself so the behaviour matches what you see in Claude Code.

The effort guide lists the levels from low to max.

Saving money on scheduled agents

Agents that run overnight, such as a nightly code review or a weekly content audit, can use the batch API at half price on both models.

Fable 5.2 also bills cached input at 2.5% of its normal rate, against 5% on Opus 5.5, so agents that reread the same instructions every turn get cheaper the longer they run.

Put your long, stable instructions at the top of the prompt so they cache well.

When a request gets refused

Fable 5.2 has strong cybersecurity safeguards, and some agent tasks that look harmless can trip them.

Set up refusal fallback so a declined step retries on Opus 5.5 automatically, and keep a list of what was refused so you can reword those prompts.

Speed for interactive work

If Fable 5.2 feels slow while you are watching it work, Claude Code supports its fast mode, which Anthropic says speeds up generation by up to 2.5 times at premium pricing.

Use it for live sessions and leave it off for background jobs.

Prompt 6: the second opinion

Here is the plan claude-opus-5-5 wrote for [task]. Using
claude-fable-5-2 at effort "high", review it as an advisor. List the
three biggest risks, anything missing, and one simpler approach if
there is one. Do not rewrite the plan; just advise.

Anthropic describes Fable 5.2 as an Advisor that Haiku, Sonnet and Opus models can consult on hard problems.

Asking for advice rather than a rewrite keeps Fable's output short, so you get its judgement without paying for a full second draft.

A weekly routine for your Agent OS

  1. Monday: review last week's cost log and move any job that failed twice on Opus 5.5 to the Fable 5.2 list.
  2. Midweek: run one big task end to end on Fable 5.2 with the finish-the-task prompt.
  3. Friday: compare passes per token for both models and adjust the router's escalation rules.

Opus 5.5 is a strong builder on its own, per Anthropic's Opus 5.5 notes, so expect the Fable list to stay short.

Which to choose

Make Opus 5.5 your default engine and Fable 5.2 your specialist.

Members of the AI Profit Boardroom get the Agent OS with Claude routing and walkthroughs for both models.

Book a free SEO strategy session →

FAQ

Should my Agent OS use Fable 5.2 or Opus 5.5?

Use Opus 5.5 by default and escalate hard, long or ambiguous tasks to Fable 5.2.

Why does Fable 5.2 stop before finishing?

Anthropic notes it can end long turns early at lower effort. Tell it to finish the whole task, or raise effort.

Is Fable 5.2 too expensive for agents?

Not always. At High effort it matched Opus 5.5's best coding score for about half the cost per task in Anthropic's test.

Can Fable 5.2 manage other models?

Yes. Anthropic describes Fable 5.2 as an Orchestrator that delegates simpler tasks to smaller models.

Final word on Fable 5.2 vs Opus 5.5

Escalate only the hard jobs and Fable 5.2 vs Opus 5.5 pays off in your Agent OS; for search growth, book a free SEO strategy session.

About Julian

I'm Julian Goldie, an AI entrepreneur, SEO expert and the founder of the AI Profit Boardroom, which has 3,400+ members.

I help business owners scale with AI agents, automation and SEO.

My YouTube channel has 400,000+ subscribers, and I run Goldie Agency, a seven-figure SEO agency.

→ Get my best AI training inside the AI Profit Boardroom

Related guides on agentos.guide

→ Hermes Fable 5: Put Claude Fable 5.1 Behind The Hermes Agent In Your Agent OS, With 5 Copy-Paste Prompts

→ The Best Hermes Agent Model, And The Five Model Lanes I Route Inside My Agent OS

→ Hermes Agent Subagents: The Quick Start, And How They Fit Your Agent OS

→ Hermes Advanced Delegation: The Quick Start, And Whether It Belongs In Your Agent OS

Get the whole Agent OS

The installable system behind every guide on this site — dashboard, agents, memory and pipelines — updated as new tools land, with 3,400+ members and four coaching calls a week.

Join the AI Profit Boardroom →

📺 Video notes + links to the tools 👉

🎥 Learn how I make these videos 👉

🆓 Get a FREE AI Course + Community + 1,000 AI Agents 👉

← all guides on agentos.guide