Anthropic has released Claude Fable 5.2, the strongest model it makes available to the general public. It is out on all platforms at $8 per million input tokens and $40 per million output tokens — down from $10 and $50 — and Anthropic says it generates up to 30% faster than the version it replaces.
Alongside it sits a twin most people will never touch. Mythos 5.2 shares Fable's core but unlocks capabilities in cybersecurity and the life sciences that Anthropic considers too sensitive for general release. Access stays limited to vetted organisations working in those fields.
The two-model strategy is becoming Anthropic's signature: one frontier model for everyone, and a restricted sibling for the small set of users trusted with its most dangerous skills. It is a product decision and a safety policy in the same shape.
The background
Frontier labs have always known more about their models than they release. Internal versions are tested with capabilities that never ship, and safety teams argue over which skills are safe to hand to the public. What is new in the Fable and Mythos pairing is that the split has become the product itself: two named models, one for everyone and one for a vetted few, differing mainly in the dangerous knowledge they are allowed to use.
The capabilities in question — cybersecurity and the life sciences — are the classic dual-use pair. The same model skill that helps a hospital defend its network can help an attacker probe one; the same biology knowledge that accelerates research can lower the barrier to misuse. Every frontier lab manages that tension somehow. Anthropic is managing it with a gate.
Why it matters
The pricing move matters alongside the safety story. Cutting Fable to $8/$40 per million tokens while making it up to 30% faster is an aggressive play for the public market — the tier where developers choose default models and ecosystems form. The restricted twin costs Anthropic nothing in that market, because the customers it serves were never going to be served openly anyway.
As governance, the model is interesting because it is voluntary. No law required Anthropic to keep Mythos behind verification; the company built the gate itself, and in doing so set a visible example of what 'responsible release' can look like in practice — tiered access based on who the user is and what they are trusted to do.
The wider story
Expect the pattern to spread. As models gain stronger skills in sensitive domains, the choice facing every lab is between holding capability back entirely, releasing it to everyone, or doing what Anthropic has done — stratifying access. Regulators watching that experiment will draw conclusions about whether the industry can tier its way out of harder rules.
What happens next
The questions are about the gate itself: who qualifies as a vetted organisation, how access decisions are reviewed, and whether the restricted model's capabilities eventually migrate into the public one as safeguards improve. Fable 5.2 will be judged as a product. Mythos will be judged as a precedent.
Common questions
What is the difference between Claude Fable 5.2 and Mythos 5.2?
Fable 5.2 is Anthropic's strongest model available to the general public, released on all platforms. Mythos 5.2 shares the same core but unlocks additional capabilities in cybersecurity and the life sciences, and access to it is limited to vetted organisations working in those fields.
How much does Claude Fable 5.2 cost?
Fable 5.2 is priced at $8 per million input tokens and $40 per million output tokens, down from $10 and $50 for its predecessor. Anthropic also says the model generates output up to 30% faster than the version it replaces.




