Anthropic unveils its inaugural Mythos-class model, Claude Fable.
Anthropic has unveiled Claude Fable 5, claiming it to be the most powerful AI model they’ve made widely accessible to date. The company highlights that Fable 5 excels in software engineering, knowledge work, and vision, showcasing a growing advantage over other models as tasks increase in length and complexity.
This release is the first from Anthropic’s Mythos class of AI models, which the company previously deemed too risky for public use due to their impressive capabilities in cybersecurity. The launch of Fable 5 was made feasible by implementing new safeguards designed to prevent responses in certain high-risk areas, with the system reverting to Claude Opus 4.8—an earlier model noted for its “honesty” when it debuted last month.
Cybersecurity and biology are two key areas where these safeguards may restrict responses, as they are considered sensitive for advanced AI systems. In testing, Anthropic reported that 95 percent of Fable sessions utilized only Fable responses, without needing to revert to Opus 4.8.
Additionally, Anthropic is introducing Claude Mythos 5, though details on this model remain sparse. According to the company, Mythos 5 shares the same underlying architecture as Fable 5, but with some safeguards relaxed. Currently, access seems to be limited to a growing number of organizations involved in the Claude Mythos Preview through Anthropic’s Project Glasswing initiative. These users will have the option to upgrade to Mythos 5, with plans to broaden access over time via a more structured trusted-access program.
Anthropic has not clarified how either model connects to Claude Mythos Preview or the reasoning behind the “5” designation, given that there appear to be no prior releases in the Mythos or Fable series.
Pricing for both models is notably higher than the previous flagship, with costs set at $10 per million input tokens and $50 per million output tokens—double the rates for Claude Opus 4.8, yet still half of what users pay for Mythos Preview.
