Claude Fable 5 in Claude Code: The Mythos-Class Upgrade for Real Coding Work

On June 9, 2026, Anthropic shipped Claude Fable 5 — the first model from its new "Mythos class" cleared for general use. It posted 80.3% on SWE-Bench Pro, a contamination-resistant coding benchmark, which puts it 11.1 points ahead of Opus 4.8 and more than 20 points clear of every competing model. Claude Code v2.1.170 added it the same day. If you're on a Pro, Max, Team, or Enterprise subscription, you can use it right now at no extra cost — until June 22.

That window is tighter than it sounds. Here's everything you need to know to put Fable 5 to work before the billing clock starts.

What "Mythos class" actually means

Anthropic's model hierarchy has long been Haiku → Sonnet → Opus. Fable 5 sits above all three in a new tier called Mythos class. The distinction isn't just raw benchmark performance — Mythos models are trained to handle qualitatively harder work: tasks that require multi-day planning, long tool chains, understanding ambiguous goals across large codebases, and synthesizing multimodal inputs in the same session.

Fable 5 is the safe-for-general-use variant of Mythos. A second model — Mythos 5 — exists with cybersecurity safeguards removed, but it's currently restricted to Project Glasswing partners and select biology researchers. Fable 5 is what ships to everyone.

The benchmark numbers

Stripe's engineering team was a launch partner. Their quote from the announcement: "compressed months of engineering into days" on a migration project. Hebbia independently rated Fable 5 state-of-the-art on its Finance Benchmark for senior-level reasoning tasks, which aren't coding benchmarks — they're complex judgment calls across long documents. The generality is real.

Using Fable 5 in Claude Code today

First, make sure you're on v2.1.170 or later. Check with claude --version. If you're behind, update now:

# Update to latest
npm install -g @anthropic-ai/claude-code

# Confirm version
claude --version
# Claude Code 2.1.170

Once updated, switch to Fable 5 inside any Claude Code session:

# Switch model mid-session
/model fable

# Or start a new session pinned to Fable 5
claude --model claude-fable-5

# In API calls, use the model identifier
# claude-fable-5

For CI and headless workflows, set the model via environment variable so every agent run in your pipeline uses Fable 5 automatically:

ANTHROPIC_MODEL=claude-fable-5

Where Fable 5 earns its price tag

Not every task needs Mythos class. The model is 2× the price of Opus 4.8, and for short, well-specified tasks you won't notice a quality difference. Fable 5 separates itself on four kinds of work:

1. Long-horizon codebase migrations

Migrations that span dozens of files, require understanding cross-cutting invariants, and take multiple tool-call chains to complete. Fable 5's extended autonomy means it doesn't lose the thread halfway through — it holds the migration state across the full session and catches the edge cases Opus misses.

2. Ambiguous requirements with large context

When the task is genuinely ambiguous — "refactor this module to be testable" on a 50-file legacy codebase — Fable 5 makes better judgment calls about scope, naming, and dependency ordering. It reads the existing patterns more accurately and proposes changes that fit the codebase's idiom rather than imposing a fresh style.

3. Multimodal + code together

Rebuilding a UI from a screenshot, extracting a schema from an API spec PDF, or implementing a feature from a Figma export — anything where the ground truth is an image or document rather than text. Fable 5's vision capabilities are materially stronger than Opus 4.8 here.

4. Adversarial review and security audits

When you need a model that finds the problems your tests didn't catch. Fable 5's longer reasoning lets it follow exploit chains across function boundaries and catch subtle timing issues that require holding more state during analysis.

# Install the codebase migration skill from skills-hub
npx @skills-hub-ai/cli install fable-5-codebase-migration

# Then run it against your target
/fable-5-codebase-migration migrate src/legacy → src/v2

The June 22 pricing catch

Fable 5 is free on Claude Pro, Max, Team, and seat-based Enterprise subscriptions through June 22, 2026 — that's ten days from today. After the 22nd, usage draws on credits or usage-based billing at $10 per million input tokens and $50 per million output tokens (exactly double Opus 4.8).

A Fable 5 workflow with skills-hub.ai

The new plugin auto-loading feature in Claude Code v2.1.157 makes skills-hub.ai even more useful alongside Fable 5. Install a skill once and it's available in every session without configuration:

# Install the migration skill (drops into .claude/skills/ automatically)
npx @skills-hub-ai/cli install fable-5-codebase-migration

# The skill is now auto-loaded on every Claude Code session start.
# With Fable 5 selected, run the migration skill directly:
/fable-5-codebase-migration migrate apps/api/src/legacy → apps/api/src/v2

The skill's phased checklist keeps Fable 5 on track across long sessions: scope audit, dependency graph, incremental migration waves, and a validation phase that confirms the migrated code is behavior-equivalent. Because Fable 5 can hold the full codebase context, the skill's dependency-ordering logic actually works — it doesn't reset between phases the way earlier models would.

For agent teams, pair Fable 5 as the orchestrator with Opus 4.8 subagents for implementation. The orchestrator holds the migration plan; subagents execute file-level changes with fresh context. This keeps cost per task reasonable while keeping quality high at the planning layer.

---
name: migration-orchestrator
description: Orchestrates large codebase migrations. Uses Fable 5 for planning, Opus 4.8 subagents for execution.
model: claude-fable-5
subagents:
  - file-migrator
  - test-validator
parallel: [test-validator]
skills:
  - fable-5-codebase-migration
---

Read the migration spec from $1. Produce a dependency-ordered wave plan.
Dispatch file-migrator subagents in waves. After each wave, fan out
test-validator against the changed files. Report progress per wave.
Halt on first validation failure — do not auto-retry.

When to stay on Opus 4.8

The quality gap is real, but so is the price delta. A rule of thumb that survives most engineering workflows:

• Use Fable 5 when the task takes more than one tool-call chain to complete, when the context is ambiguous or spans many files, or when you need multimodal input handled correctly.
• Stay on Opus 4.8 for well-specified feature work, quick bug fixes, single-file edits, test generation for isolated functions, and anything where you can write a tight spec in advance.
• Use Sonnet for interactive pair-programming where you want fast turns and you're guiding the implementation yourself.

The best signal: if your session takes more than 10 turns to complete what you set out to do, that's a long-horizon task — run it on Fable 5. If it takes 1–3 turns, Opus 4.8 is fine and you won't see a meaningful quality difference.

Ten days is a narrow window. Run your hardest queued task on Fable 5 this week. The benchmark numbers tell you it can handle it — and Stripe's engineering team found out the hard way that "months of engineering" is a real unit of time that a better model can compress.

# Get started now
npx @skills-hub-ai/cli install fable-5-codebase-migration