Claude Fable 5 is the beginning of the end of ‘one model for every use case’: Here’s how
Follow Us
I remember when the big AI story was whether these models could pass a bar exam. Now one of them is designing drug candidates without a human in the room. We moved fast. Yet there is an underlying change in Anthropic’s release of Fable 5 and Mythos 5 that strikes me as more significant than any particular metric: the identical technology is sold as two distinct products based on your identity. Fable 5 is available to all. Mythos 5 is available to individuals who have been approved by the US government. Identical weights, but separate policies. Even in naming these products, Anthropic made clear what kind of distinction they are making – Fable and Mythos are derived from the same Latin word roots.
SurveyAlso read: No PhD required: Mythos 5 ran an entire drug design workflow by itself
This is unprecedented. Not in having enterprise tiers, which every SaaS startup provides. What’s unprecedented here is the rationing of the capability itself on the grounds of identity rather than any other criteria. It doesn’t matter whether we give a slightly slower or less expensive version to non-approved entities; what matters is the restriction of certain capabilities. An approved cybersecurity expert will be able to think through the offensive exploit chain using Anthropic’s technology. All others will have access to Opus 4.8.
Of course, the immediate argument against this is that it is the responsible choice. Anthropic is right in pointing out that having Mythos class cybersecurity capability in the wrong hands is indeed dangerous. But that doesn’t change the fact that this is the creation of a two-tiered world with respect to artificial intelligence, where the government and institutions of trust have one thing, while the rest of us have another.
This divide will only continue to grow. The next version of Mythos will be even better, while the number of institutions that have been granted permission to use the API will grow slowly. At the same time, classifiers will continue to make mistakes while researchers in India struggle with the fallback system when the chosen biotech startup in San Francisco has their hands on the Mythos 5 API key.
Also read: Claude Fable 5 is less risky Mythos model: Safest AI right now?
Anthropic’s launch post is full of genuinely impressive demos. Fable 5 beat Pokémon FireRed using only raw screenshots – no maps, no helper tools, just vision. Earlier Claude models needed a full harness to attempt the same thing. It’s the kind of detail that makes for great copy, and Anthropic knows it.
But that’s exactly the point. The demos are designed to dazzle. The Pokémon playthrough, the Factorio factory, the fluid simulation set to music – they’re all real, and they’re all remarkable. They’re also a distraction from the quieter, more consequential thing happening underneath: a credentialed hierarchy being dressed up as a safety policy.
The “one model for everyone” era wasn’t perfect. But it was legible. What’s replacing it isn’t.
That’s the real story of this launch. Not the benchmarks. Not the Pokémon playthrough. The doors that just got installed and who holds the keys.
Also read: Microwave capacity guide: What size do you need?
