Anthropic launched Claude Fable 5, the first version of its elite Mythos AI model that regular people can actually use. But there are strict rules. 

Mythos has been locked behind closed doors since April. Only a small group of trusted partners had access. 

Now, for the first time, anyone can tap into that power if they’re willing to follow Anthropic’s new terms.

Fable 5

Fable 5 is a powerful version of Mythos with guardrails built in. Mythos is Anthropic’s most advanced AI system. 

It’s so powerful that the company worried about what could happen if the wrong person got hold of it. So they kept it private.

Fable 5, however, is Mythos brought to the public, but with hard limits on what it will and won’t do.

The model is good at software engineering, complex knowledge work, and understanding images. Early users are already impressed.

Analytics firm Hex was the first to hit a 90% score on its toughest benchmark. That test involves long, complex analytical tasks. Fable 5 cracked it, something no other model had before.

App-building platform Base44 noted that Fable is better at building full apps in a single attempt.

AI platform Genspark said it beat every other model they tested, especially in UI design and game coding.

The Catch

Fable 5 won’t answer everything. In high-risk areas (cybersecurity, biology, chemistry, and distillation), the model passes the question to Claude Opus 4.8, a safer, less powerful model.

Anthropic says this happens rarely. Early data shows that at least 95% of Fable sessions run entirely on Fable’s own responses. So most users won’t notice the switch.

Safety Tests

Before releasing Fable 5, Anthropic ran an internal bug bounty program. Testers spent over 1,000 hours trying to break the model with jailbreak attempts. 

None of them worked universally. Then Anthropic brought in outside red-teaming organizations; they also failed to find a universal workaround.

That said, Anthropic admits new attacks could still emerge. That’s why the company is requiring something new and controversial.

All traffic going through Fable 5 and Mythos 5 will be retained for 30 days. Even companies that previously had zero-retention agreements must comply. 

Anthropic says it won’t train on this data. The only purpose is to defend against novel jailbreak attempts and reduce false positives.

This could set a new industry standard. More powerful models may soon come with mandatory data retention policies, framed as safety measures.

Access and Cost

Fable 5 is available right now through the Claude API and consumption-based Enterprise plans.

From now through June 22, Fable 5 is included in Pro, Max, Team, and seat-based Enterprise plans at no extra charge.

On June 23, it gets pulled from those plans. After that, users will need usage credits to access it. Anthropic plans to restore it as a standard feature “as soon as possible.”

Fable 5 and Mythos 5 cost $10 per million input tokens and $50 per million output tokens. That’s double what Opus 4.8 runs.

For many businesses, that cost will be a dealbreaker. AI budgets are already stretched thin.

Advanced reasoning models can turn a single request into multiple tasks and multiple charges.

But some companies say the price is worth it. Shopping rewards platform Rakuten explained that Fable 5 reviews and validates its own work at the highest effort level. 

That kind of self-checking makes autonomous operations possible, they said. In their view, the extra cost pays for itself.

Mythos 5 

Alongside Fable 5, Anthropic released a new version of Mythos itself called Mythos 5.

This one isn’t for the public. Only organizations already approved for Mythos access can use it.

That group expanded last week from a handful of partners to hundreds of organizations across 15 countries.