Anthropic Launches Claude Fable 5, a Public Version of Mythos, for a Limited Time
Claude Fable 5 is the name of Anthropic’s new AI model, and it has one notable feature: it is built on the same architecture as the Mythos class of models, which we have often discussed in recent weeks. If you want to take advantage of it, now is the time: it is “free” for a limited period.
Table of Contents
From Mythos to Fable 5: the safer version of a model considered dangerous
Since April 2026, we have been hearing about the Claude Mythos Preview model and Anthropic’s Glasswing project. A cutting-edge AI model above all the others to the point that it would pose global security risks. That is believable, given the number of vulnerabilities it has helped uncover in applications over the weeks.
As a result, Anthropic decided to launch two models:
- Claude Mythos 5, access to which is reserved for a small group of partners (including Europe is not among them),
- Claude Fable 5, a version intended for the general public, in which Anthropic has introduced numerous safeguards.
"Mythos 5 will initially be deployed as part of the Glasswing project, in collaboration with the U.S. government, as an update to Claude Mythos Preview.", Anthropic explains.
To allow businesses and users to benefit from this power without compromising safety, Anthropic has developed robust guardrails for this same class of models. This is how the secure public version was born: Claude Fable 5.

Strict guardrails backed by Opus 4.8
To ensure Fable 5 cannot easily be exploited by cybercriminals, Anthropic has put protection mechanisms in place that can block or automatically redirect sensitive requests.
Now, if a user tries to submit requests related to critical domains, the model applies strict filtering. This is especially true for offensive cybersecurity, biology, and chemistry. Any request touching on these topics is automatically blocked or routed to another model recently made available: Claude Opus 4.8.
"Our early data shows that more than 95% of Fable sessions require no fallback; for those sessions, Fable 5’s performance is practically identical to Mythos 5’s.", Anthropic says. Fable 5 and Mythos 5 are therefore very close.
Excessive consumption compared with Opus
Fable 5 is now available, at no additional cost, for those with a Claude subscription (Pro, Max, Team). However, this is a short-lived benefit, as Anthropic says you will only be able to enjoy it until June 22, 2026.
In reality, unlimited Fable 5 is not happening: this model is too compute-hungry. It is therefore too expensive for Anthropic to run. Once this trial period expires, access will switch to usage-based pricing.
And it will sting! If you use it intensively, the bill will climb: Fable 5 consumes tokens twice as fast as the Opus model. Anthropic does not give real-world examples, but it is essential to use it only for certain tasks.
"The pricing for the two models is $10 per million input tokens and $50 per million output tokens. Developers can use claude-fable-5 via the Claude API.", Anthropic says.
What do you think?


