Claude Fable 5: Mythos model for all Apfelpatient

For months, Anthropic's most powerful model was considered too powerful for the public – but now it's finally here. With the Claude Fable 5, the company is introducing a Mythos-class model that any customer can use starting today, while the unrestricted Mythos 5 variant remains reserved for a select group.

In April, Anthropic unveiled its Claude Mythos Preview model, explaining that it was simply too dangerous for public release. Instead, access was granted only to select partners – including Apple, which used Mythos Preview as part of its Project Glasswing cybersecurity program to detect vulnerabilities, including those reported in macOS. With Claude Fable 5, the company is now fulfilling its earlier promise and making this performance class widely available – albeit with a security architecture designed to contain precisely those capabilities that initially kept Mythos under wraps.

Two models, one foundation

Anthropic is releasing two variants based on the same technical platform. According to the company, Claude Fable 5 is the first publicly available model in the Mythos class. Its capabilities reportedly surpass anything Anthropic has released to date: In virtually all tested performance benchmarks, Fable 5 is said to be at the current top level, with particularly strong results in software development, knowledge work, image processing, and scientific research. According to Anthropic, the longer and more complex a task, the more pronounced the advantage over its other models becomes.

Alongside this is Claude Mythos 5 – essentially the same model, but without some of the security mechanisms. This version remains reserved for a small circle of cyber defense providers and infrastructure vendors and will initially be rolled out via Project Glasswing in cooperation with the US government, as the successor to Mythos Preview. Anthropic describes Mythos 5 as the model with the strongest cybersecurity capabilities worldwide and announces that access will later be opened via an expanded trust program.

Safety filter instead of emergency braking

Anthropic believes that releasing such a powerful model carries significant risks – especially in the area of cybersecurity, where its capabilities could be misused to cause serious damage. Therefore, Fable 5 launches with a protection concept that does not answer queries on certain topics itself, but instead forwards them to the next most powerful model, Claude Opus 4.8. This affects queries related to cybersecurity, biology and chemistry, as well as distillation; users are notified of the redirection.

Anthropic has deliberately calibrated these filters cautiously. The company acknowledges that occasionally harmless requests are intercepted – but on average, the mechanism is triggered in less than five percent of all sessions. In over 95 percent of cases, Fable 5 therefore operates without any redirection, making its performance virtually identical to that of Mythos 5. With upcoming, even more powerful models, Anthropic intends to further improve the protection mechanisms and reduce the number of false triggers.

Price halved compared to its predecessor

Anthropic has made a significant step downwards in pricing. Both the Fable 5 and Mythos 5 cost ten US dollars per million input tokens and 50 US dollars per million output tokens – less than half the price of the Mythos Preview. This brings the Mythos performance class into a price range that will be attractive to a wider customer base for the first time, even though the models remain considerably more expensive than previous standard models. In direct comparisons, according to Anthropic's own measurements, the two new additions perform favorably compared to the Claude Opus 4.8 and GPT-5.5.

According to Anthropic, both models can operate independently for longer periods than any previous generation of Claude. This improves their use in software development, knowledge work, and image processing, as well as their ability to retain information over extended periods – areas where the company recently introduced new features for autonomously operating Claude models.

Staggered access due to expected demand

Claude Fable 5 is available to customers immediately, but Anthropic anticipates very high and difficult-to-predict demand. Therefore, access via subscription plans will be phased:

Period	Availability of Fable 5
From today until June 22nd	Included in the Pro, Max, Team and seat-based Enterprise plans at no extra charge
From June 23rd	Removal from these tariffs; usage will then only be possible via prepaid credits.
Thereafter	Planned return as a regular component of subscription plans as soon as sufficient capacity is available.

If capacity becomes sufficient sooner, Anthropic intends to extend the free period. The company says the return to standard rates will happen as quickly as possible.

What the Mythos class means for the industry

The public release marks a turning point for a model whose existence for months was defined primarily by its potential for danger. The very cybersecurity strength that made Mythos Preview a tool for Glasswing partners is also the reason for the now-implemented safeguards. The fact that Anthropic continues to restrict the unfiltered version, Mythos 5, to defenders and infrastructure providers demonstrates how fine the line is between protection and misuse in this performance class – a tension that was recently revealed when unauthorized individuals temporarily gained access to Mythos. The competition has also recognized the urgency: OpenAI followed suit with its own security program in response to Glasswing. With Fable 5, the question now shifts from "Who is even allowed to use such a model?" to "How do you keep it secure in everyday use?". (Image: Shutterstock / Stockinq)