Anthropic Opens the Vault on Its Most Powerful AI Yet - With Guardrails

Anthropic has quietly released Claude Fable 5, the first broadly available model from its high-capability Mythos family. The company describes it as the most powerful AI it's ever put into customers' hands, claiming exceptional performance on software engineering, knowledge work, and vision tasks - especially as projects grow longer and more complex.

What makes this launch unusual is the backstory. Anthropic had previously classified Mythos-class models as too risky for public release, citing their advanced cybersecurity capabilities. Now, after developing new safeguards, the company says it can let Fable 5 loose - with a catch. In sensitive domains like cybersecurity and biology, the model may automatically defer to Claude Opus 4.8, a less capable but more "honest" fallback model. Testing showed 95% of sessions ran entirely on Fable without needing that fallback.

Anthropic also released Claude Mythos 5, which is essentially the same underlying model but with some safeguards removed. Access is currently limited to organizations already in Anthropic's private Project Glasswing initiative. The company plans to expand access gradually through a more systematic trusted-user program.

Pricing reflects the model's power: $10 per million input tokens and $50 per million output tokens - double the rate for Opus 4.8, though half the cost of the earlier Mythos Preview. Anthropic declined to clarify the numbering scheme or how these models relate to the Preview version.

Source: The Verge