Claude Fable 5 & Mythos 5: Key highlights from Anthropic’s latest launch

Anthropic on Wednesday launched Claude Fable 5, a Mythos-class model approved for general use, and Claude Mythos 5, a restricted-access version with fewer safeguards for trusted cybersecurity and life sciences users. Here are some key highlights about these models.

What are Fable 5 and Mythos 5?

Fable 5 is Anthropic’s most capable publicly available model, which leads key benchmarks across software engineering, knowledge work, vision (ability to understand images), scientific research, memory, and long-context tasks. Its advantage grows on longer and more complex workloads.


Mythos 5 is the same underlying model as Fable 5 but with certain safeguards removed.

Anthropic also highlighted major gains in analytical reasoning, drug design, molecular biology hypothesis generation, and genomics research. The company claims Mythos 5 can accelerate parts of drug discovery, generate novel scientific hypotheses, and conduct extended autonomous research.

Safety testing found Mythos 5’s alignment performance (measures how well its behaviour matches instructions) comparable to Opus 4.8, with low levels of deceptive or misuse-related behaviour.

Anthropic also introduced stronger safeguards against jailbreaks and misuse, claiming significantly improved resistance compared with previous models.

To mitigate misuse risks, particularly in cybersecurity and biological research, Fable 5 uses new safety classifiers. Requests related to cybersecurity, biology, chemistry, or model distillation (where a large model transfers its knowledge to a smaller one) are automatically routed to Claude Opus 4.8. Rather than issuing outright refusals, the Opus model responds to sensitive queries using more refined safety filters to distinguish legitimate research from misuse. These safeguards affect fewer than 5% of sessions on average.

For Mythos-class models, the company stated it will retain business customer data for 30 days for safety monitoring purposes, while stating it will not use the data for model training.

What about cost and access?

Both Fable 5 and Mythos 5 are priced at $10 per million input tokens and $50 per million output tokens.

Fable 5 is available immediately through the Claude API and on eligible Claude subscription plans during an initial rollout period.

Meanwhile, Mythos 5 is currently available only to select cyber defenders and infrastructure providers through Project Glasswing, with broader trusted-access programs planned for cybersecurity and life sciences researchers.

The launches come just weeks after India gained access to Claude Mythos Preview, Anthropic’s advanced cybersecurity-focussed AI model, as part of an expansion of its Project Glasswing initiative.

Why now?

The launch of these models marks a major step in Anthropic’s long-term effort to make Mythos-class AI capabilities more widely accessible.

“Our goal is to safely open up access to vetted partners for Mythos 5’s dual-use capabilities, to stay up to date on our plans for trusted access,” the company wrote in its blog post. Dual-use refers to technology that has both civil or commercial and military or malicious use, like drones or satellites.

The launch comes shortly after the company confidentially filed for an initial public offering (IPO) with the US Securities and Exchange Commission.

Anthropic has seen strong business momentum, reporting a revenue run rate of $47 billion in May, up sharply from about $10 billion a year earlier. The company also recently secured fresh funding at a $965 billion valuation, surpassing OpenAI’s $852 billion valuation reported in March.