Update, June 14, 2026: Bad news for the curious. Three days after launch, the US government ordered Anthropic to disable Fable 5 again — the model is now gone for everyone. Hope you were quick enough. What happened is in the postscript at the end of this article.
What Is Fable 5?
Anthropic just released Claude Fable 5. It's the first publicly available model from the Mythos class — a full tier above the previous Opus models.
Translation: the most capable AI model you can actually try right now. No government clearance. No special waiting list.
What Can It Do?
Fable 5 scores roughly 10% higher than Claude Opus 4.8 — Anthropic's previous flagship — on benchmarks. It shines at:
- Complex reasoning — multi-step problems where you need to hold a lot in mind at once
- Coding tasks — in tests, it handled a two-month migration in a single day
- Scientific questions and long documents
Will you notice the difference in daily use? Depends. If you're using Claude for simple tasks — rewriting text, quick research — probably not. If you really push it, yes.
What's Free? What Costs Money?
Anyone currently on a paid Claude plan (Pro, Max, Team, or Enterprise) can try Fable 5 for free through June 22. From June 23, you'll need usage credits on top of your subscription.
Free plan users: no access to Fable 5.
Developers using the API pay $10 per million input tokens and $50 per million output tokens. A token is roughly a syllable. That's double the price of Opus 4.8 — until now the flagship for "regular" users. For a quick chat, irrelevant. For large projects, it adds up.
Is This for Me or Just Developers?
Honest answer: for most beginners, not really.
Sure, with a paid claude.ai account you can pick Fable 5 from the dropdown and chat away. But that's not what the model is built for. Its strength is the heavy lifting — combing through entire programs and projects, analyzing huge documents. And that only makes sense via the API, with the matching pile of tokens. On the smallest tier, you won't get far.
So for you as a beginner, Fable 5 is mainly a nice conversation topic at the next get-together — "yeah, I know it, I tried it once." Perfectly fine.
Who actually benefits? Companies willing to invest seriously — via API, through Amazon Bedrock, Google Vertex AI, or Microsoft Foundry. And that's entirely justified: a top model for serious, large-scale work has every right to exist. It's just not something you need on the side in your browser.
What About Mythos 5?
There's also the bigger sibling: Claude Mythos 5. That's the full-power Mythos-class model, without Fable 5's restrictions.
Sounds exciting. Not for regular users though.
Mythos 5 runs through Project Glasswing, a collaboration with the US government. Access is currently limited to cybersecurity organizations and critical infrastructure operators. Interesting to read about — not to try.
The Safety Trick
Fable 5 has built-in rerouting. If you ask about certain sensitive topics — cybersecurity, biology, chemistry — the older Opus 4.8 answers instead of Fable 5. The model tells you when this happens.
Sounds odd. Makes sense though: Anthropic wanted to release its most powerful model publicly without taking on unnecessary risk.
According to Anthropic, this affects fewer than 5% of conversations.
Bottom Line
Fable 5 is Anthropic's most capable publicly available model to date — and at double the token price of Opus 4.8, also its most expensive. Paid plan users can try it free through June 22.
Anthropic is making Mythos-level power accessible. With guardrails, with a price tag — but accessible.
For serious use, that means API and a budget. For everyone else: click around while it's free. Then you'll have something to say at the next AI get-together.
Postscript: Big Launch, Then Shut Down
Fable 5 was publicly available for barely three days before it was over. On June 12, 2026, the US government ordered Anthropic, via an export-control directive, to disable both Fable 5 and its sibling Mythos 5 — citing national security. The order banned access by "any foreign national," whether inside or outside the US. Since that can't be cleanly separated in real time, Anthropic pulled the plug entirely — worldwide, for everyone.
The trigger: a jailbreak technique that allegedly got the model to identify software vulnerabilities in someone else's code. Anthropic disagrees publicly — the technique is narrow, already known, and present in other freely available models too. Apply that standard across the industry, and practically no new model would ever ship.
And here's where it gets interesting. Only days earlier, that same US government had dismissed Anthropic's proposal to be able to slow AI development if needed as an anti-competitive maneuver. Washington's message back then: America must lead, not hesitate. A few days later, that very same Washington yanks a commercial model used by hundreds of millions off the market within 72 hours. Who's hesitating and who's leading here — decide for yourself.
No return date has been announced. Anthropic's other models — Opus 4.8 and the rest — keep running unchanged.