Anthropic’s Claude Fable: A Contemporary Myth Accessible to Everyone

Anthropic has unveiled its most sophisticated AI model to the public for the first time, while ensuring strict safety protocols are in place.

On Tuesday, the AI firm launched Claude Fable 5, the first public edition of its Mythos model. Anthropic claims that Fable 5 excels in software development, knowledge tasks, and visual capabilities, but it comes with stringent safety measures. In high-risk areas such as cybersecurity, biology, chemistry, and distillation, the model limits its responses and falls back to Claude Opus 4.8.

Initially available only in preview mode as of April, Mythos was accessible to a select group of partners due to cybersecurity concerns. Recently, Anthropic expanded access to hundreds of organizations in 15 countries, particularly targeting those responsible for critical infrastructure.

Now, this technology is available to all through Anthropic’s Claude API and usage-based Enterprise plans. Subscription access will be rolled out gradually: until June 22, Fable 5 will be included at no extra cost in Pro, Max, Team, and seat-based Enterprise plans. After June 23, Anthropic will require usage credits for Fable 5 but plans to reinstate it as a standard subscription feature as soon as possible.

Anthropic is also introducing a new version of Mythos, called Mythos 5, for organizations already authorized to access the advanced model.

The launch of Fable coincides with Anthropic’s preparations to enter public markets, along with OpenAI and Elon Musk’s SpaceX. This follows the company’s call for major global AI labs to implement coordinated controls on advanced AI development, warning that systems are advancing quickly enough to potentially achieve recursive self-improvement (RSI) without human oversight.

Worried about the misuse of a Mythos-class model, Anthropic conducted stress tests on its classifiers against jailbreak attempts prior to the release of Fable 5.

“Internally, we carried out an external bug bounty that uncovered no universal jailbreaks after more than 1,000 hours of testing. We also partnered with outside red-teaming organizations, which similarly did not find any universal jailbreaks.”

However, new attack vectors may still be possible. Therefore, with the launch of Fable 5 and Mythos 5, Anthropic will implement a 30-day retention policy on all traffic, even for enterprises with prior zero-retention agreements. Anthropic stated that it will not use this data for training but merely to “counter complex and novel attacks, including new jailbreaks” and “identify and reduce false positives.” This policy may establish an industry norm where access to increasingly robust models is tied to mandatory data retention measures presented as safety protocols.

For model users, not every query will result in a Fable 5 response. Anthropic suggests that the cases where Fable defaults to Opus 4.8 are infrequent, with early data indicating that over 95% of Fable sessions operate solely on the model’s own outputs.

In evaluations from third parties, analytics firm Hex reported that Fable earned a score of 90% on its core analytics benchmark for complex, prolonged analytical tasks.

“On the most challenging questions, it shows strong judgement and an appreciation for nuance,” Hex noted.

Vibe-coding platform Base44 highlighted that Fable excels at “one-shotting full apps” and shows excellent tool-calling capabilities. AI-driven workspace platform Genspark claimed that Fable outperformed all other models in its tests, especially in UI design and game design tasks.

Pricing for both Fable 5 and Mythos 5 is set at $10 per million input tokens and $50 per million output tokens, which is twice as much as Opus 4.8. This pricing could hinder widespread adoption.

Many enterprises are voicing concerns over escalating AI costs, particularly after facing unexpected invoices or depleting their annual AI budgets sooner than planned. Advanced models like Opus 4.8 can intensify these financial challenges, as their sophisticated reasoning abilities can fragment a single request into several tasks.

Anthropic expects that demand for Fable 5 will be very high and hard to predict. Some companies, like shopping rewards platform Rakuten, feel that the advantages justify the costs.

“At the highest effort, Fable reflects on and validates its own work,” Rakuten stated. “For us, this functionality facilitates highly autonomous operations — the additional cognitive effort pays for itself.”