The rumors started in April. Anthropic Claude Mythos Preview. An AI so sharp it could find every hole in the internet’s armor. Scary? Yeah. That’s why they held it back. Slow-rolling the release is safer when your tool can break the net.
But now it’s here. Sort of.
Anthropic officially released the first accessible member of the Mythos family: Claude Fable 5. On Tuesday. They’re calling it their most advanced model that you can actually use without needing a government badge. It handles software engineering better. Knowledge work? Cleaner. Images? It understands them.
It even beat Pokemon FireRed.
Not metaphorically. Literally. Using vision-only controls, the AI played and beat a 2004 Game BoyAdvance title. Previous models failed. This one won.
But power costs money. The full Mythos lineup eats compute like crazy. So Anthropic isn’t giving this away for nothing. In a few weeks, they’re going to charge subscribers extra to touch Fable 5.
The Safety Locks
Is it safe?
Anthropic claims yes. They built Fable 5 as the “safe” sibling of Claude Mythos 5. The real monster—Mythos—stays behind the glass of Project Glasswing, reserved for trusted cybersecurity pros. Fable? It has guardrails Mythos doesn’t.
“Fable 5 complied with zero harmful singleturn requests relating to planning a cyberattac, exploit development, or defense evasion.”
They tested it against thirty different jailbreak techniques. It didn’t break.
But there’s a hole. Or rather, a deliberate blank space. AI-enabled bioterror. Horrifying? Undoubtedly. Real? Getting closer. Because of that, Fable 5 won’t answer most questions about biology or chemistry. If you ask, it redirects you to Opus 4.8 at least for now. Anthropic says it’s talking with the US government about when Mythos 5 might come out. That conversation probably sparked the White House push for pre-release government review of AI models. Which, by the way, isn’t a rule yet.
One more change: data.
Business users now agree to 30-day retention. Not for training. For defense. They’re keeping your data to stop future attacks.
How to Get Your Hands On It
It’s available now. But barely.
Anthropic is rolling this out conservatively. Staged. The Pro, Max, Team, and enterprise plans can access it right now until June 22nd. That’s it. A short window.
Then on June 23rd it disappears from the standard menus.
You can still use it, sure. But you’ll burn usage credits. Pay-as-you-go style. Enterprise users might have some in the tank. The rest of you? Expect extra fees on your bill. Developers using the API can start now. It costs $10 per million input and $50 million output tokens.
Why the weird release schedule? Capacity. Fable 5 uses double the resources of the Opus line. Demand will be high. Keeping Claude online means limiting who can access Fable at first.
Anthropic wants to eventually make this standard in paid plans. When? No date. If at all? Unclear.
