Anthropic Releases a Restricted Version of Its Powerful Claude Mythos AI to Subscribers
Anthropic released two new models on Tuesday evening, Claude Fable 5 and Claude Mythos 5. Claude Fable 5 will be available to paying subscribers until June 22, and it is not yet known what the subscription price will be after that. It is a restricted version of Claude Mythos, the AI model that only selected companies were given access to because of its dangerous cyber capabilities. By contrast, Claude Mythos 5 was released only to partners who already had access to Claude Mythos Preview, and is essentially an upgraded version of it.
Claude Fable 5 was launched with safeguards that will automatically route higher-risk queries to lower-performing models. Thus, instead of responding directly to certain requests in the fields of cyber, biology, chemistry and model distillation, Fable 5 will pass them to Claude Opus 4.8, a model with less powerful capabilities. It should be noted that Fable 5 costs twice as much as Opus 4.8.
"For us, it really is about what we call the 'race to the top' , the ability to deliver this technology in a valuable way, and at the same time provide the right safety mechanisms so that you can produce vastly asymmetric benefit relative to harm," Diane Penn, Anthropic's head of product management for research, told CNBC.
With the launch of Claude Fable 5, however, Anthropic is achieving the "ultimate goal" it has stated, to deploy Mythos-level models at scale. The company is also taking advantage of growing momentum and investor interest in its technology ahead of its potentially huge public offering, which is expected as early as this year.
In April, Anthropic released Claude Mythos Preview to about 50 companies only, as part of Project Glasswing. Those companies used the model to identify more than 10,000 high-severity or critical security vulnerabilities within just a few weeks. Last week, Anthropic announced an expansion of the project to another 150 organizations in more than 15 countries, including in electricity, water, health and other critical sectors. The codebases of many of these organizations form the foundation for operations at many organizations around the world, including governments.
Anthropic said the new model shows "outstanding performance" on software engineering tasks. According to a post it published, in several benchmark tests it scored more than 10% higher compared with Claude Opus 4.8.
