Claude Fable 5: a Mythos-class1 model - Anthropic's most capable model, now available to everyone

Anthropic just made their most powerful model available to the public for the first time.
What it is: Claude Fable 5 is a Mythos-class model, the same tier of AI that was locked behind Project Glasswing since April and only accessible to cybersecurity partners like Apple, Google, Microsoft, and JPMorgan. Fable 5 is the public version with safety classifiers layered on top.

What makes it different: Most model launches are either locked down with hard refusals or released wide open. Fable 5 does something new. When you hit a sensitive area like cybersecurity, biology, chemistry, or model distillation, your query quietly routes to Opus 4.8 instead. You still get a response, just from a slightly less capable model. No walls. This triggers in less than 5% of sessions.

What it can actually do:

• Stripe used it to migrate 50 million lines of Ruby in a single day. That's a two-month team project compressed into hours

• Scores highest among frontier models on Cognition's FrontierCode eval, even at medium effort

• Beat Pokémon FireRed using only raw screenshots with zero helper tools

• Top scores on Hebbia's finance benchmark for document reasoning, chart interpretation, and problem solving

• The unrestricted Mythos 5 version accelerated drug design by 10x and autonomously conducted genomics research that outperformed a model published in Science

Who it's for: Developers building on the API (model string: claude-fable-5, $10/M input, $50/M output). Pro, Max, and Team subscribers get free access through June 22 before it moves to usage credits. Cybersecurity and biomedical teams can apply for the trusted access program to get full Mythos 5 without classifiers.

The key thing here isn't just the benchmarks. It's that the gap between Fable 5 and everything else gets wider the harder and longer the task is. That's a fundamentally different scaling behavior than what we've seen from previous models.