What Is Claude Fable 5? Features, Use Cases, Pricing, and Safety Explained

If you follow AI news even loosely, you've probably seen the name Claude Fable 5 pop up over the past few weeks. It launched with a lot of fanfare, got pulled offline days later because of a government order, and is now being restored. That's a lot to untangle for a model that's barely a month old.

This post breaks down what Claude Fable 5 actually is, what it can do, what it costs, and why Anthropic built it with a very different safety approach than earlier Claude models.

What Is Claude Fable 5?

Claude Fable 5 is Anthropic's most capable model available for general use. It belongs to what Anthropic calls the "Mythos class," a tier of models built for long, complex, multi-step work rather than quick one-off answers. The name itself is a nod to that idea. Fable comes from the Latin word for "that which is told," which echoes the meaning behind Mythos.

Here's where it gets interesting. Fable 5 isn't a separate model built from scratch. It's the same underlying system as Claude Mythos 5, a restricted model that Anthropic has only made available to a small number of vetted cyberdefense and government partners through a program called Project Glasswing. Fable 5 is essentially that same model, but with extra safety classifiers layered on top so it can be released to everyone else.

Key Features of Claude Fable 5

• State of the art benchmark performance: Anthropic says Fable 5 is state of the art on nearly all tested capability benchmarks, with especially strong results in software engineering, knowledge work, vision, and scientific research. According to Anthropic, the longer and more complex a task gets, the bigger its lead over previous Claude models.

• Long, autonomous work sessions: Fable 5 is built to work for extended stretches with minimal supervision. Running inside an agent harness like Claude Code, it can plan across multiple stages, delegate parts of a task to sub-agents, and check its own work as it goes.

• Strong coding and migration ability: During early testing, Stripe reported that Fable 5 compressed months of engineering work into days. In one case, it completed a codebase-wide migration across a 50 million line Ruby codebase in a single day, a job that would normally take a full team more than two months by hand.

• Vision and document understanding: The model can read diagrams, charts, and tables embedded in files and PDFs, which is useful for document-heavy fields like finance, legal, and architecture. It also uses vision to check its own coding output against the original design.

• Improved memory across long tasks: Fable 5 can hold context across millions of tokens and refine its output using notes it writes for itself along the way, which helps it stay on track during long research or coding sessions.

Use Cases of Anthropic Claude Fable 5

Fable 5 is aimed at people who need more than a quick chat response. A few areas where it stands out:

• Software engineering. Large refactors, complex implementations, and multi-day coding projects that would normally need a full engineering team.

• Knowledge work. Deep research, analysis, and polished deliverables that are ready for review rather than heavy editing.

• Scientific research. Anthropic has highlighted early use in drug design, where Mythos 5 reportedly helped internal researchers speed up parts of the process by around ten times.

• Cybersecurity (through Mythos 5). For vetted partners, the underlying model has been used to help defenders find and fix vulnerabilities in real software.

Because of how capable the underlying model is, Anthropic has been careful about who gets access to which version, which brings us to pricing and safety.

Pricing of Claude Fable 5

Claude Fable 5 and Claude Mythos 5 share the same pricing: $10 per million input tokens and $50 per million output tokens through the API. That's roughly double the cost of Claude Opus 4.8, but Anthropic has pointed out it's still less than half the price of the earlier Claude Mythos Preview.

On subscription plans, access has rolled out in stages rather than being a flat, permanent inclusion, so it's worth checking Anthropic's current plan details before assuming what's included at a given tier.

Safety: Why Fable 5 Comes With Guardrails

This is really the core of the Fable 5 story. Because the underlying Mythos-class model is so capable, especially in cybersecurity and biology, Anthropic built Fable 5 with safety classifiers that can automatically decline or redirect certain requests. If a query touches on high-risk areas like advanced cybersecurity or bioweapon-adjacent topics, the system routes that request to Claude Opus 4.8 instead of answering with Fable 5's full capability. Mythos 5, the more restricted version reserved for vetted partners, does not have these same classifiers lifted.

Anthropic has said it stress-tested these safeguards extensively before release, including an internal bug bounty and outside red-teaming, and reported that no universal jailbreak was found during that process.

That said, the rollout hasn't been smooth. Just days after launch, the U.S. Commerce Department issued an export control directive ordering Anthropic to suspend access to both Fable 5 and Mythos 5 for any foreign national, citing national security concerns tied to a reported jailbreaking technique. Anthropic complied and took both models offline worldwide, while disputing how serious the underlying vulnerability actually was. More recently, Commerce withdrew those controls after further review, and Anthropic has begun restoring access with an additional set of cybersecurity-focused classifiers in place. Given how quickly this situation has moved, it's a good idea to check Anthropic's own announcements for the current status before relying on Fable 5 for anything time-sensitive.

Why This Matters Beyond the Headlines

The Fable 5 story is a pretty clear real-world example of something a lot of organizations are grappling with right now: how do you deploy powerful AI responsibly, and what happens when regulators, safety teams, and business goals pull in different directions? Questions like risk classification, auditability, and lifecycle governance aren't abstract anymore. They're playing out in public with real deadlines and real consequences.

If your work touches AI adoption, procurement, or compliance in any way, it's worth building a working understanding of how these governance frameworks actually function rather than piecing it together from news coverage. That's exactly the gap our course, Responsible AI: AI Ethics, Governance & Compliance, is meant to fill. It walks through frameworks like the EU AI Act, NIST's AI Risk Management Framework, and ISO 42001, along with practical governance tools like auditing, risk management, and ethical oversight across the AI lifecycle. It's a short, self-paced course, so it's easy to fit in alongside other work.

The Bottom Line

Claude Fable 5 represents a genuine jump in what a generally available AI model can do, paired with an unusually direct example of how quickly AI governance decisions can change on short notice. Whether you're evaluating it for your own workflows or just trying to keep up with the broader conversation around frontier AI safety, it's a model worth watching closely over the coming months.