Hey there,

Back in April we covered the release of Anthropic’s most powerful model, Mythos. The model was so good at coding and cybersecurity tasks that Anthropic was forced to pause its rollout. But not anymore. Claude Fable 5 is a new Mythos-class model that was just released to the public. This week we’ll cover what exactly Fable 5 is, how Anthropic made it safe for release and who can use it. 

What is Mythos? 

To understand Fable 5, we first need a quick refresher about Mythos. Mythos was Anthropic's new family of models that it released to select partners back in April. The model never received a public release because of its high cybersecurity capabilities. It was so good at finding and exploiting vulnerabilities in the tools and websites we use daily, that Anthropic determined it couldn’t be released in its current state. Instead they created Project Glasswing that focuses on using models like Mythos to help secure core infrastructure. 

What is Fable 5?

Fable 5 is a Mythos-class model that was made safe for general use. Simply put, it performs better than any other model the company has released publicly. Fable 5 scores better than other models in nearly every tested benchmark, making it truly state-of-the-art. It excels at software engineering, knowledge work, vision, and research. Fable 5’s performance continuously outpaces other models the longer it works. This makes it uniquely positioned to solve large, complex problems.  

How Anthropic made it safe

The problem with Mythos was that it was too good at cybersecurity tasks and was ultimately a risk if made public. To address this issue, Anthropic used a simple idea, restricting the topics it could answer. Of course a model that has huge knowledge gaps isn’t useful so they created a system where restricted topics aren’t ignored, but instead passed on to another model. By passing queries about biology, chemistry, and cybersecurity to Opus 4.8, Fable 5 could be released publicly without needing to worry about misuse. In an effort to release Fable 5 sooner, Anthropic has tuned the safeguards to be extra vigilant, meaning they will sometimes catch harmless requests. 

Alongside Fable 5, Anthropic has released a new version of Mythos into Project Glasswing. Unlike Fable 5, Mythos 5 keeps its capabilities fully intact. This is why it’s getting a limited release to vetted partners rather than going public. 

How to access it

Fable 5 is currently available to all Pro, Max, Team, and Enterprise plans, however, this is only temporary. On June 23, all paid plan users will lose access to this model and it will only be available through usage credits. This means that you’ll have to pay for Fable 5, even if you have the $100/month Max plan. Anthropic explained that they plan on bringing Fable 5 back to the paid plans in the future if capacity allows. 

My Thoughts

The Mythos problem was something that many AI skeptics were worried about, a model so good at something that it would be a danger to the public if widely released. Anthropic’s choice to not only pause the release of Mythos, but to create Project Glasswing around it went a long way in establishing a precedent for how AI labs should approach potentially dangerous models. I believe that limiting certain capabilities for Fable 5 was the correct path forward. 

What I like about Anthropic’s fallback approach is that they are still shipping a state-of-the-art model and just limiting its abilities around a few key areas. It also seems like they created this system to ensure that no harmful prompts slip through, bringing a “better safe than sorry” approach. But honestly, I don’t see many users actually running into the failsafes when using Fable 5. This will most likely only impact users already working in those areas.