Anthropic Launches Claude Fable 5, a Revolution in Public AI

⚡

Key Takeaways

1Anthropic unveiled Claude Mythos 5, its first public model in the Mythos class, on June 9, 2026.

2The model includes safeguards to redirect sensitive queries to Claude Opus 4.8, limiting malicious uses.

3Claude Mythos 5, without these safeguards, is reserved for partners of the Project Glasswing, a cybersecurity program.

💡Why it matters — This advancement marks a turning point in public access to advanced AIs, while raising questions about the control of these technologies.

Anthropic Unveils Claude Fable 5, a Major Advancement

On June 9, 2026, Anthropic announced the launch of Claude Fable 5, marking a significant milestone in the availability of its most advanced artificial intelligence models. This model, belonging to the Mythos class, is the first of its kind to be accessible to the general public. Positioned above the Opus range, Claude Fable 5 is presented as the most powerful model developed by Anthropic to date. To frame its use, safeguards have been put in place, redirecting sensitive queries to Claude Opus 4.8.

A Mythos Class Model Now Accessible to All

Historically, Mythos class models were considered too dangerous for open dissemination, as revealed by the launch of Claude Mythos Preview in April. Claude Fable 5 stands out for its exceptional performance in various fields such as software engineering, knowledge work, vision, and scientific research. To limit malicious uses, Anthropic has integrated classifiers capable of detecting queries related to sensitive areas such as cybersecurity, biology, chemistry, and distillation. When such queries are identified, they are automatically redirected to Claude Opus 4.8, the company's most advanced public model. Anthropic claims that over 95% of sessions do not require a switch to another model. In the Claude app, it is possible to disable this redirection, which pauses the conversation rather than transferring it to another model.

Mythos 5, Reserved for a Select Circle

In parallel, Anthropic has introduced Claude Mythos 5, which is based on the same model as Fable 5 but without the safeguards. Its access is limited to partners of the Project Glasswing, a cybersecurity program conducted in collaboration with the U.S. government. Anthropic also plans to soon open a dedicated access program for biological research, allowing the use of Fable 5 without restrictions on biology and chemistry.

A Launch That Extends Anthropic's Cyber Strategy

This dual launch fits into a trajectory initiated in April when the company refused to make its first Mythos model public, deeming it too dangerous and capable, according to them, of detecting thousands of vulnerabilities. Since then, Anthropic has ramped up initiatives in the field of cybersecurity, notably with Claude Security, its vulnerability detection tool. The Glasswing initiative, expanded to about 150 entities across a dozen countries, has, however, faced criticism. Arthur Mensch, co-founder and CEO of Mistral AI, expressed reservations, labeling this approach as "fear marketing." Some observers are also concerned that such programs foreshadow a control by large companies and the U.S. government over access to the most powerful AIs.

Immediate Availability, But Costs to Anticipate

Claude Fable 5 is available today on the Claude API and on Enterprise offerings billed by usage. For subscriptions, deployment is gradual: the model is included at no extra cost in the Pro, Max, Team, and Enterprise plans per seat until June 22. Starting from June 23, it will be removed from these plans, and its use will require "usage credits," before a planned return to standard access when capacity allows.

Fable 5 Deployment Schedule

June 9: Immediate availability on the API and usage-based Enterprise offerings
Until June 22: Included at no extra cost in Pro, Max, Team, and Enterprise plans per seat
From June 23: Removal from plans, access via "usage credits"

In terms of pricing, Fable 5 and Mythos 5 are charged $10 per million input tokens and $50 for output, which is double that of Opus 4.8 but half of the Mythos launched in April. In the app, the model also consumes twice as fast the usage quota compared to Opus, a point to watch for subscribers.

Pricing and Consumption of Fable 5

$10 per million tokens for input, $50 for output
Double that of Opus 4.8 and half of Mythos Preview
In the app, Fable consumes twice the usage quota of Opus

Finally, Anthropic accompanies this launch with a new data retention policy. For its Mythos class models, user activity will be retained for 30 days, including for clients who previously benefited from a non-retention policy. The company assures that it does not exploit this data to train its models and reserves it for defense against complex attacks as well as for reducing false positives.