Anthropic introduced Claude Fable 5 and Claude Mythos 5. What they can do

Anthropic has introduced new AI models Claude Fable 5 and Claude Mythos 5. 

This was announced by Anthropic.

The new models can operate autonomously for longer than previous Claude versions. 

Technical capabilities 

According to Anthropic, Claude Fable 5 is the company’s most powerful model among those available for general use.

The model showed strong performance in programming, data work, document analysis and scientific research.

However, Fable 5 will not fulfill certain requests in cybersecurity, biology, chemistry, and model distillation.

If the classifier detects a potential threat, the request will be automatically routed to the less powerful Claude Opus 4.8 model. 

The company also introduced Claude Mythos 5 — a model on the same architecture but without some of the cybersecurity restrictions.

Initially it will be available through the Project Glasswing program, implemented in partnership with the U.S. government and organizations working in cyber defense.

Anthropic plans to expand access to Mythos 5 later through a trusted access program.

How Claude Fable 5 and Claude Mythos 5 were tested

Anthropic says Fable 5 and Mythos 5 demonstrate improved results in software development, performing complex analytical tasks, image processing and handling large amounts of context. 

Mythos 5 was also used in drug discovery research, protein design and genomics.

During testing by Stripe, the Fable 5 model performed a migration in a Ruby codebase of 50 million lines in one day — a task that would have taken the engineering team over two months. 

On the Hebbia financial benchmark, the model showed the best performance in document reasoning and interpretation of charts and tables.

The model also independently completed the game Pokemon FireRed with a minimal visual interface.

Security measures

The company is instituting mandatory 30-day retention of interaction data with models of this level.

Anthropic says this data will be used to detect new security threats and attempts to bypass protective mechanisms.

Subscription cost for the new models

Claude Fable 5 is available on the Claude API platform and in Enterprise corporate plans. For users on Pro, Max, Team and standard Enterprise subscriptions, access is free until June 22.

From June 23, using the model on these plans will require purchasing additional credits until the company expands infrastructure capacity to return Fable 5 to the base subscriptions.

The cost of using Claude Fable 5 and Claude Mythos 5 is $10 per million input tokens and $50 per million output tokens.

More about Mythos

Mythos is positioned as one of the most «advanced» models in programming and «agent» AI — that is, one capable of autonomously performing complex tasks.

It can find vulnerabilities in cybersecurity and potentially create or improve cyberattacks.

The Mythos model is also used by the U.S. National Security Agency (NSA), despite the Pentagon earlier including the company on a list of risky suppliers.

Read also: Anthropic plans to open access to Mythos-level models, but will first improve their security