Anthropic Releases Claude Fable 5, Its First Mythos-Class Model Made Public
Key Takeaways
- ▸Claude Fable 5 shows exceptional performance in software engineering, knowledge work, and vision tasks, with advantages growing on longer and more complex queries
- ▸First public release of the Mythos-class models, previously deemed too risky due to their advanced cybersecurity and biology capabilities
- ▸New safeguards block responses in high-risk domains, with 95% of Fable sessions running entirely on model without falling back to Opus 4.8
Summary
Anthropic has announced Claude Fable 5, the most powerful model it has made widely available to the public. The release marks the first public launch of the Mythos-class of AI models, a family Anthropic previously withheld from public release due to safety concerns around its exceptional capabilities in cybersecurity and biology tasks. The company says new safeguards make the public release possible, with the system blocking responses in high-risk domains and falling back to Claude Opus 4.8 when encountering sensitive queries—a fallback that occurs in only 5% of sessions.
Alongside Fable 5, Anthropic is releasing Claude Mythos 5, which uses the same underlying model but with safeguards lifted in certain areas. Access to Mythos 5 is currently restricted to organizations in Anthropic's private Project Glasswing initiative, with plans to expand through a more systematic trusted-access program in the future. Pricing for both models is significantly higher than Claude Opus 4.8: $10 per million input tokens and $50 per million output tokens—double the rate of Opus 4.8 but half what Mythos Preview access costs.
- Claude Mythos 5 offered to Project Glasswing organizations with safeguards removed in certain areas
- Pricing at $10/$50 per million tokens (input/output) represents double Claude Opus 4.8 rates
Editorial Opinion
Claude Fable 5's release represents a pragmatic shift from Anthropic's previous approach of outright refusal to release advanced models. By deploying sophisticated safeguards rather than indefinite lockdown, Anthropic demonstrates that capability and caution need not be mutually exclusive. However, the tiered access model—with unrestricted Mythos 5 confined to a select group—suggests Anthropic remains hesitant about fully unfettered deployment, treating the safeguards as guardrails rather than a complete solution to AI safety concerns.
