Anthropic announced two versions of its latest, highest-performance Claude model: Claude Fable 5, available to a broad audience, and Claude Mythos 5, offered to a smaller set of cybersecurity pilot partners. The company says both variants share the same base model, with different safety guardrails applied.
Pricing and availability
Both models are charged under Anthropic’s new pricing: $10 per million input tokens and $50 for output. That is substantially higher than the pricing for Opus 4.8, which was $5 and $25 respectively, and makes this the most expensive Claude Anthropic has offered to date. Pilot partners report that testing costs alone have reached into the millions of dollars.
Subscribers receive Fable 5 for free until June 22, 2026; after that date access switches to usage-credit billing.
Demonstrated use cases
Anthropic highlighted several practical results: Stripe migrated a 50-million-line codebase in one day using the model. Separately, Mythos 5 ran a week of autonomous genomics tasks that Anthropic says outperformed a model published in Science, while being roughly one hundredth the size of that published model.
Real-world implications
These examples indicate that raw AI capability is no longer the primary bottleneck: the models can perform complex, resource-intensive tasks. However, those same autonomous, compute-heavy tasks consume large amounts of tokens, causing costs to scale rapidly with the level of autonomy demonstrated.
Because of the pricing and usage model, the most capable Claude risks becoming a product showcased in lab demos and limited pilots rather than a tool that most enterprises can run continuously.
Conclusion
Anthropic has built its most capable Claude yet, but its pricing positions that model as a place you visit occasionally rather than a tool for everyday operation. Capability has ceased to be the main constraint; the metered cost now determines who can use the model and how often.
