AI Development

Claude Fable 5.2 in Claude Code: high effort, one CLAUDE.md sentence and outcome-based prompts

Fable 5.2 starts at medium effort in Claude Code and ranks below Opus 5.5. With /effort high it beats it at a lower cost per task. Plus one CLAUDE.md line.

Luca Di Domenico

Luca Di Domenico

· 5 min read

In short

Claude Fable 5.2 costs 20% less than Fable 5.1 ($8/$40 per million tokens) and generates up to 30% faster, but in Claude Code it starts at medium effort, where it ranks below Opus 5.5 on the Artificial Analysis leaderboard. With /effort high it reaches 62 points versus 59 for Opus 5.5 at max, at $9.08 per task instead of $10.04. The CLAUDE.md sentence Anthropic recommends makes it merge repeated code in about half of the tries instead of under one in ten. It works best on big jobs: refactors, migrations, hard bugs, long sessions and non-coding work.

To get the most out of Claude Fable 5.2 in Claude Code you need three moves: raise effort to high, add the code-ownership sentence Anthropic recommends to your CLAUDE.md, and ask for the outcome instead of the steps. Fable 5.2 came out on November 9, 2026. It costs 20% less than Fable 5.1 and, according to Anthropic, generates up to 30% faster. But in Claude Code it starts at medium effort, and at that level it performs below Opus 5.5.

The starting numbers:

Why do you need to configure Fable 5.2 by hand in Claude Code?

In Claude Code, Fable 5.2 starts with reasoning effort set to medium. Fable 5.1 started at high. The Claude Code documentation says so plainly: Fable 5.2, like Opus 5.5, Sonnet 5.5 and Haiku 5.5, defaults to medium.

At medium, Fable 5.2 ranks below Opus 5.5. On the Artificial Analysis leaderboard, checked on November 11, 2026:

  • Fable 5.2 at medium: 56 points on the Intelligence Index;
  • Opus 5.5 at its maximum (max): 59 points, $10.04 per task;
  • Fable 5.2 at high: 62 points, $9.08 per task.

So Fable 5.2 at high beats Opus 5.5 at max and costs less per task. In Claude Code you set it like this:

/model fable
/effort high

The level you type after /effort stays saved for later sessions too. If /model fable doesn’t give you Fable 5.2, your Claude Code version is too old: update it with claude update and check with claude --version that you’re on 2.1.321 or later.

Which sentence should you add to CLAUDE.md for Fable 5.2?

Anthropic recommends telling Fable 5.2 that the code is its own. The sentence, from the official Fable 5.2 prompting guide, is this:

This codebase is yours: what it's for and how it's built are your call, and you'll keep building on it.

It looks like a small thing, but it matters. Without this sentence, Fable 5.2 bolts new pieces on top of existing code, even when that repeats code that’s already there. And it flags the problems it finds, but sometimes doesn’t fix them.

With the sentence, the behavior changes. In Anthropic’s tests, Fable 5.2 merged repeated code in about half of the tries, versus under one in ten without it. The result is more maintainable code, which AI agents also handle better in later sessions.

The same guide lists two costs. With the sentence, Fable 5.2 uses about 1.5 times the output tokens and makes changes nobody asked for, mostly new tests and documentation. So review the diff, not just the final summary. If the code belongs to a client and you want it changed as little as possible, leave the sentence out.

CLAUDE.md is also where rules written for older models pile up: I explained why it pays to clean it out at regular intervals.

How should you ask Fable 5.2 for things?

Tell Fable 5.2 the result you want, not the steps to get there. Describe the end state ("checkout has to work without an account too") and let the model plan the route.

You don’t need to remind it to check its work either: Fable 5.2 already does that on its own. Instructions like that are a habit left over from earlier models. I collected more prompt habits worth changing in these 4 changes for Opus 5.5.

Which jobs is Fable 5.2 worth using for?

Fable 5.2 does its best work on big jobs. For small, quick requests, the extra time and tokens don’t pay off. The cases where I use it:

  1. Rebuilding a major part of your app.
  2. Moving the app to a different technology. In its announcement, Anthropic mentions migrating a production service from Python to Rust.
  3. Finding the cause of a bug you can’t figure out.
  4. Long sessions you would normally split into several.
  5. Work that isn’t programming.

On that last point there’s a concrete figure. On Scale Labs’ Remote Labor Index, a test built from real freelance projects sourced from Upwork, Fable 5.2 completes 41.3% of projects at a level a client would accept. Opus 5.5 stops at 21.25%, Fable 5.1 at 17.92%. According to Scale Labs, as reported by The Decoder, the biggest gains are in audio and music editing, architecture and interior design, and 3D modeling.

Where should you start?

If you’re still on Fable 5.1, switching takes five minutes: update Claude Code, pick fable, set effort to high, and add the sentence to the CLAUDE.md of the projects you own. Then hand it the big job you’ve been putting off for weeks: that’s where you’ll see the difference.

Sources

  • Introducing Claude Fable 5.2 and Claude Mythos 5.2 — Announcement of November 9, 2026: output up to 30% faster than Fable 5.1, medium effort as the Claude Code default, a Python-to-Rust service migration among the use cases, Remote Labor Index table.
  • What's new in Claude Fable 5.2 — Documentation checked on November 11, 2026: Fable 5.2 costs $8 per million input tokens and $40 per million output tokens, Fable 5.1 cost $10 and $50 (-20%).
  • Claude Code, Model configuration — Checked on November 11, 2026: Fable 5.2 defaults to medium effort (other unlisted models default to high), requires Claude Code 2.1.321 or later, is selected with /model fable; /effort followed by a level saves it as the default.
  • Prompting Claude Fable 5.2 — Checked on November 11, 2026, "Say who owns the code" section: the sentence "This codebase is yours..."; with it the model merges repeated code in about half of the tries, under a tenth without; about 1.5 times the output tokens and unrequested changes.
  • Artificial Analysis, LLM Leaderboard — Checked on November 11, 2026: Fable 5.2 at medium 56 points, at high 62 points and $9.08 per task; Opus 5.5 at max 59 points and $10.04 per task.
  • Remote Labor Index (RLI), Scale Labs — Checked on November 11, 2026: real freelance projects sourced from Upwork; automation rate 41.30% for Fable 5.2, 21.25% for Opus 5.5, 17.92% for Fable 5.1.
  • Anthropic's Fable 5.2 completes 41 percent of real freelance jobs, up from 2.5 percent a year ago — Article of November 10, 2026 quoting Scale Labs’ post on X: Fable 5.2’s biggest gains are in audio and music editing, architecture and interior design, and 3D models.

Want to know what this means for your product?

If you're building software with AI and want a second opinion on how you're working, tell me about your project.

Tell me about your project
Claude CodeClaude FableAI AgentsAI Development

Follow the blog

One email for every new article

I only write when an article is out: title, summary and link. You can unsubscribe with one click from any email.

RSS feed

An RSS feed is the up-to-date list of articles: add it to a reader app like Feedly or Inoreader and new articles show up there, no email needed.

lucadidomenico.studio/en/blog/rss.xml