Anthropic is bringing its most powerful AI model to the general public for the first time, but it’s doing so with guardrails.
On Tuesday, the AI company announced Claude Fable 5, the first publicly available version of the Mythos model. Anthropic says Fable 5 excels in software engineering, knowledge work, and vision, but has severe safety limitations. In high-risk areas such as cybersecurity, biology, chemistry, and distillation, the model blocks the response and falls back to Claude Opus 4.8.
Launched in preview in April, Mythos was initially limited to a small number of partners due to cybersecurity concerns. Last week, Anthropic expanded its access to hundreds of organizations in 15 countries, once again focusing on organizations that manage critical infrastructure.
Now, a version of that technology is available to everyone through Anthropic’s Claude API and pay-as-you-go enterprise plans. Access with subscriptions will be rolled out in stages. Until June 22nd, Fable 5 is included in Pro, Max, Team, and seat-based Enterprise plans at no additional charge. On June 23, Anthropic will remove Fable 5 from these plans, which will now require usage credits, and will restore it as a standard subscription feature as soon as possible.
Anthropic is rolling out a new version of Mythos, called Mythos 5, to organizations that have already been approved to access advanced models.
Fable’s launch comes as Anthropic prepares to enter the public market alongside OpenAI and Elon Musk’s SpaceX. This also follows an appeal by AI companies calling on the world’s major AI laboratories to establish a coordinated brake pedal in frontier AI development. Anthropik warned that systems are advancing rapidly and could soon achieve recursive self-improvement (RSI), where systems improve themselves autonomously without human intervention.
Wary of the possibility of Mythos class models falling into the wrong hands, Anthropic says it attempted a jailbreak and stress-tested the classifier before releasing Fable 5.
“We implemented an external bug bounty internally, but over 1,000 hours of testing did not find a universal jailbreak. We then worked with an external red team organization and were similarly unable to find a universal jailbreak.”
However, new attacks may still exist. As a result, with the release of Fable 5 and Mythos 5, Anthropic said it will now require 30 days of retention on all traffic, even if businesses previously had a zero-retention agreement. The company said it will not use this data for training, only to “defend against complex and novel attacks, including new jailbreaks” and “identify and reduce false positives.” This policy could set a precedent for an industry where access to increasingly powerful models comes with mandatory data retention policies built in as safeguards.
For those who stick with this model, not every question will have an answer in Fable 5. According to Anthropic, there are rare cases where Fable has to be deferred until Opus 4.8, and early data shows that at least 95% of Fable sessions are running entirely in the model’s own responses.
Analytics firm Hex said in a statement that for the first time, Fable has scored 90% in third-party testing on an analytics benchmark that is core to complex, long-running analytics tasks.
“We demonstrate strong judgment and attention to nuance when it comes to the most difficult questions,” Hex said.
Vibe coding platform Base44 said in a statement that Fable excels at “one-shot complete apps” and has great tool calls. According to AI-powered workspace and agent platform Genspark, Fable outperformed all other models in its evaluation and performed significantly better on tasks such as UI design and game coding.
Both Fable 5 and Mythos 5 are priced at $10 per million input tokens and $50 per million output tokens, double the price of Opus 4.8. This price alone may be a deterrent to widespread use.
Many companies are increasingly critical of AI costs as they see AI bills coming in and annual AI budgets being exhausted prematurely. Advanced models like Opus 4.8 can exacerbate these issues because they have advanced inference skills that can split a single request into multiple tasks.
Anthropic said it expects demand for Fable 5 to be very high and difficult to predict. And certainly some, like shopping rewards platform Rakuten, may think the increased price is worth it.
Rakuten said in a statement, “Fable will make every effort to reflect on and verify its own efforts.” “For us, that’s what allows us to operate highly autonomously. The extra thinking pays off.”
If you buy through links in our articles, we may earn a small commission. This does not affect editorial independence.
