\n<\/p>\n
Back in April, Anthropic unveiled its Claude Mythos AI model that it says is too powerful to be released publicly. Instead, the company shared access with software vendors, including Apple, in an effort to use the model to improve online security.<\/p>\n
Now, as promised, Anthropic has released what it calls a Mythos-class model that customers can actually use. The new model is called Claude Fable.<\/p>\n
<\/p>\n Anthropic calls the Claude Fable 5 its first Mythos-class model publicly available to customers. <\/p>\n “The capabilities of the Fable 5 exceed those of any model we’ve ever made available,” the company said. “It ranks high in almost all tested benchmarks of AI capability, showing impressive performance in software engineering, knowledge work, vision, scientific research, and many other areas. When the task is long and complex, Fable 5 leads above our other models.”<\/p>\n Anthropic adds that its new Claude Fable 5 model comes with some protections, given its power:<\/p>\n Releasing a capable model comes with risks. Without defenses, Fable 5’s abilities in areas like cybersecurity can be misused to cause serious damage. We have therefore introduced a model with protections which means that questions on certain topics will find an answer in our most powerful model, the Claude Opus 4.8. To release the model safely and quickly, we tuned these protections carefully\u2014sometimes they will catch harmless requests, although they start, on average, in less than 5% of sessions. With more capable models coming in the coming months, we are working to improve our defenses and reduce false positives as quickly as possible.<\/p>\n<\/blockquote>\n Beyond Claude Fable 5, Anthropic also details Claude Mythos 5, which is basically Fable 5 without many protections. Mythos 5 is only available to “a small group of cyberdefenders and infrastructure providers,” according to Anthropic. <\/p>\nClaude Fable 5 and Claude Mythos 5<\/h2>\n
\n