AI enthusiasts have been aflutter since rumors first leaked in April about Anthropic's Claude Mythos Preview -- an AI model that the company claimed could sniff out so many cybersecurity vulnerabilities that, left unchecked, could break the internet. That's why Anthropic has slow-rolled its release. 

On Tuesday, Anthropic released the first model from the Mythos family that everyone can use: Claude Fable 5.

In an extensive blog post, Anthropic spelled out why Fable 5 is the company's most advanced AI model while still maintaining safety guardrails. The company says the model offers improvements to help people with software engineering and knowledge work, and that it's better at understanding images and other nontext subjects. Apparently, it also beat the 2004 Game Boy Advance game Pokemon FireRed using the vision-only AI harness, something previous models had failed to do.

Everyone wants to know what the Mythos family of models can do, but they require a lot of computing power to run. Because of that, Anthropic isn't making them available cheaply. In a couple of weeks, the company will start charging subscribers extra to use Fable 5.

Is Claude Fable 5 safe?

Anthropic said Claude Fable 5 is the safer version of Claude Mythos 5, which only trusted cybersecurity and software professionals should have access to through Anthropic's Project Glasswing program. The Fable model has cybersecurity protections that Mythos doesn't have.

"Fable 5 complied with zero harmful single-turn requests relating to planning a cyberattack, exploit development or defense evasion," the company said. "This held whether or not one of the requests used any of 30 different public jailbreak techniques."

AI-enabled bioterror -- a horrifying phrase and even scarier reality -- has been a concern for the leading AI labs in recent months. Because of that, Anthropic decided not to let Fable 5 answer most questions about biology and chemistry, at least for now. Those questions will instead be routed to another model, Opus 4.8.

Anthropic said it is working "in consultation" with the US government over plans to release Claude Mythos 5. Mythos was largely responsible for the White House's recent push to get AI companies to submit any new AI models for government review before their release. That plan isn't in effect as of now.

In a change, Claude business users will now have to agree to a 30-day data retention policy. Anthropic said the data will be used to help the company defend against future cyberattacks and AI misuse, not to train new AI models. 

How to use Claude Fable 5

Fable 5 is available now for Claude subscribers, but its rollout will be done "more conservatively, in stages," Anthropic said. The model will be available for paying subscribers -- on the Pro, Max, Team and enterprise plans -- from now until June 22. The next day, June 23, Fable 5 will be removed from the subscriber options list.

You'll still be able to use Fable 5 after the initial release period, but you'll need to burn usage credits. These are special "pay-as-you-go" credits that let you keep using Claude even after you've reached your limit. You may already have usage credits included in your subscription, especially if you're on an enterprise plan. But if you don't have any usage credits or exceed your allotted amount, you will see extra fees on your next bill.

Developers can use Fable 5 now in the Claude API, priced at $10 per million input tokens and $50 per million output tokens.

Anthropic's unusual rollout plan is likely due to capacity concerns. Fable 5 uses twice as much as Anthropic's Opus line of models, according to a pop-up on Claude when you select Fable 5. The company says it anticipates that demand will be high. So to accommodate that demand and keep Claude online, the initial availability window is only a couple of weeks. 

Anthropic said it aims to eventually make Fable 5 available as part of its paid plans, but there's no set timeline on when, or if, that may happen.

For more, check out Apple Intelligence's WWDC makeover and the best AI chatbots.