What Is Claude Fable 5? A Complete Guide to Anthropic’s Most Powerful AI Model—Performance, Pricing, and How to Use It

On June 9, 2026, AI leader Anthropic officially released its new model, Claude Fable 5.

This is a historic release: for the first time, capabilities previously available only to select organizations in the “Mythos class” are being offered to general users—with built-in safety guardrails. Anthropic says the model outperforms all previous models across a wide range of domains, including coding, image recognition, legal work, and science.

This article explains Claude Fable 5’s features, performance, pricing, and how it differs from other models.

Why Claude Fable 5 Is Getting Attention

What Is the “Mythos” Class?

Anthropic has long offered model lines such as Claude Opus and Sonnet. In April 2026, it announced Mythos—a tier above those, representing its most capable models.

Mythos is exceptionally strong at discovering and exploiting cyberattack techniques. Because of that risk, it has not been released to the public and has been limited to defensive cybersecurity use cases for trusted partners.

Claude Fable 5 is the answer to that constraint. It uses the same base model as Mythos but adds safety classifiers that automatically detect and block dangerous queries, making the capability available to general users.

Key Features of Claude Fable 5

1. State-of-the-Art on Nearly Every Benchmark

Anthropic describes Fable 5 as “state of the art on nearly every evaluated AI capability benchmark.” Notable scores include:

• SWE-bench Pro (code repair): 80.3%

  • Claude Opus 4.8: 69.2%
  • GPT-5.5: 58.6%
  • Gemini 3.1 Pro: 54.2%

• FrontierCode Diamond (mergeable quality evaluation): 29.3%

  • Claude Opus 4.8: 13.4% (more than double)
  • GPT-5.5: 5.7%

Beyond coding, it leads competing models across knowledge work, image understanding, tool use, law, biology, and more.

2. Exceptional Vision (Image Recognition)

Fable 5’s vision improvements are especially noteworthy.

A striking demo: clearing Pokémon Red using only screenshots of the game screen. No maps or helper tools—just “seeing” the screen—and it played from start to finish.

Tasks that were difficult even for earlier Claude models with many auxiliary tools are now achievable with visual input alone, showing major advances in image understanding and long-horizon planning.

3. Built for Long, Complex Tasks

Fable 5’s biggest strength is not one-off Q&A but long-running, complex work:

• Context window: 1 million tokens
• Maximum output: 128,000 tokens

You can load large codebases or documents at once and run multi-step workflows continuously. Anthropic positions this model for “the most demanding reasoning and long-horizon autonomous work.”

Stripe reported using it for a full migration of a 50-million-line Ruby codebase—work that previously took two months, completed in one day.

4. Safety Design: Fallback Mechanism

Fable 5 includes a safety design not seen in previous models.

When high-risk queries related to cybersecurity, biology, chemistry, or model distillation are detected, Fable 5 automatically falls back to Claude Opus 4.8 for the response.

In other words, the most capable model is not used for dangerous use cases—a two-tier design. When fallback occurs, Opus 4.8 pricing applies.

Claude Fable 5 Pricing

API pricing for Claude Fable 5:

That is roughly 2× Claude Opus 4.8 and 3×+ Claude Sonnet 4.6.

You do not need Fable 5 for every task. A practical approach is Sonnet or Opus for everyday questions and light work, and Fable 5 for high-stakes, complex jobs where failure is costly.

Free Trial Period

Through June 22, 2026, Pro, Max, Team, and Enterprise subscribers can use Fable 5 at no extra charge. Use this window to try it on your hardest real-world workflows.

Claude Fable 5 vs. Claude Mythos 5

Claude Mythos 5, released at the same time, shares the same base model as Fable 5 but differs in safety limits and audience.

Mythos 5 is offered through the Project Glasswing program to U.S. government cybersecurity defense personnel and similar trusted partners.

Where You Can Use It

Claude Fable 5 is available on:

• Claude Platform (claude.ai)
• Amazon Web Services (Amazon Bedrock)
• Google Cloud
• Microsoft Azure Foundry
• GitHub Copilot (Pro+, Max, Business, Enterprise)

The API model string is claude-fable-5.

Best Use Cases

Fable 5 shines especially in:

• Complex code generation, review, and refactoring
• Reading and analyzing large document sets
• Long-running autonomous agent work
• Specialized reasoning in law, medicine, finance, and similar fields
• UI understanding and spatial reasoning from images
• Continuous improvement using file-based memory

For short copy, simple summaries, and routine Q&A, Opus or Sonnet is usually enough.

Summary

Claude Fable 5 is Anthropic’s most capable publicly released model to date. By delivering Mythos-class capability with safety guardrails, it symbolizes a shift in the AI race—from raw performance alone to how much you can safely delegate.

Through June 22, 2026, paid plan users can try it at no extra cost. Start with your hardest workflows and see how much Fable 5 can solve that previous models could not.

For the backstory—why Mythos was deemed “too dangerous,” and how Anthropic reversed course toward public release—see our related article below.

→ AI ‘Too Dangerous to Release’ Gets Full Rollout in Weeks—The Core Contradiction Behind Anthropic’s Mythos Policy Reversal

Reference: Anthropic official announcement “Introducing Claude Fable 5 and Claude Mythos 5” (June 9, 2026)