Anthropic opens its most powerful AI ever (with safeguards) to the public

The Fable 5 model is the first model in the Mythos class, Anthropic’s most advanced line of AI technology.

Anthropic, creator of artificial intelligence (AI) models Claudemade the most powerful version of its technology available to the general public this Tuesday, although maintaining rrestrictions on its use in sensitive areas.

Nicknamed Fable 5the model is first in class MythosAnthropic’s most advanced line of AI technology, introduced in April but so far limited due to cybersecurity concerns — from becoming widely available.

According to Anthropic, Fable 5 is particularly competent at writing and debugging computer codein answering complex research questions and in image analysis.

In parallel, Anthropic is providing a unrestricted versionwith the name of Claude Mythos 5to companies and organizations that already have access to this family of models, including cybersecurity partners enrolled in your program.

It is restricted group was expanded at the beginning of June to around 200 organizations in more than 15 countries, and is expected to continue to grow.

A Anthropic tem limited access to Mythos for cybersecurity reasonstaking into account what the company describes as the model’s ability to quickly identify vulnerabilities in critical infrastructures, such as banks, electrical networks and software.

When Project Glasswing was released, some critics accused Anthropic of exaggerate the threat to generate media attention. But companies that tested the Mythos came in the meantime validate your capabilitiesand the US Government, which had been involved in a legal dispute with Anthropic, ended up testing the model due to .

The White House has since established a mechanism to test the most powerful models from leading AI companies before they are released.

According to Anthropic, restrictions on Fable 5 will work as follows: most requests related to sensitive topics, such as cybersecurity, will be forwarded to a lower level modelOpus 4.8, which was made available to the public at the end of May.

The company hired external experts to dedicate over 1000 hours trying to find ways around these restrictions — a process known as red-teaming.

Anthropic also launched a reward program for vulnerabilities, which pays to anyone who finds security flaws. According to the company, no one has been able to discover a way to completely unlock the model.

The issue of security, which Anthropic turned into one of its main commercial arguments, led the San Francisco startup to a iron arm unprecedented with the Trump administration due to its refusal to lift restrictions on mass surveillance and lethal autonomous weapons.

Following this dispute, the Pentagon terminated the contracts with the company, whose AI tools were the only ones with security clearance for use in defense.

The launch of Fable 5 arrives with a high price: 10 dollars per million input tokens and 50 dollars per million tokens outputtwice the cost of Opus 4.8. Tokens are units of measurement for the use of an AI model.

An intensive programming session can consume a million tokens in a few hours, or even less.

Despite the exponential revenue growthAnthropic remains far from profitable and pays a high price for computing capacity.

He recently started leasing a data center to Elon Musk for $1.25 billion a month. These launches come at a time of strong financial excitement around AI.

Anthropic and rival OpenAI recently announced that they have submitted plans for a initial public offering of sharesa few days before the record debut on the stock exchange of SpaceX, Elon Musk’s space giantwhich also owns xAI, its AI company.