Anthropic has unveiled Fable 5, a new artificial intelligence model that brings many of the advanced capabilities of its previously restricted Mythos system to a broader audience. The launch marks a significant step in making the company’s most sophisticated AI technologies more accessible while maintaining strict safety controls.
Mythos first emerged earlier this year through Project Glasswing, a program designed to give select organizations access to advanced AI capabilities for cybersecurity testing and research. The initiative allowed partners to evaluate how emerging AI systems could be used in both defensive and offensive cyber scenarios, helping strengthen digital security frameworks.
Now, Anthropic is extending part of that technology to the public with Fable 5, which the company describes as its most capable publicly available model to date.
According to Anthropic, Fable 5 demonstrates substantial improvements in handling complex, long-duration tasks. Internal evaluations showed the model outperforming the company’s previous flagship system while delivering strong results across software development, document analysis, and reasoning-intensive workflows.
One of the most notable advancements comes in the area of visual understanding. Earlier generations of Anthropic’s AI models struggled with interpreting dynamic visual environments and required additional tools to process complex graphical information. Fable 5 significantly improves on those limitations by analyzing visual content more effectively and maintaining a stronger understanding of spatial relationships and context.
The company showcased the model’s progress through gaming-related benchmarks, where Fable 5 was able to navigate and complete tasks using visual input alone, requiring far less external assistance than previous systems. Beyond gaming, these vision capabilities enable practical applications such as extracting detailed information from scientific charts, interpreting technical diagrams, and recreating software interfaces from screenshots.
To address safety concerns, Anthropic has implemented a layered protection system within Fable 5. Certain sensitive prompts are automatically redirected to a less powerful model that incorporates stricter safeguards. The company acknowledges that these controls may occasionally flag harmless requests, but says they were intentionally designed to prioritize safety during the model’s initial rollout.
Anthropic stated that it plans to refine these protections over time, reducing unnecessary restrictions while maintaining security standards as future generations of AI models become more powerful.
Alongside Fable 5, the company also introduced Mythos 5, a version built on the same core technology but with fewer restrictions. Access to Mythos 5 will initially remain limited through controlled programs before expanding to approved users under a trusted-access framework.
Fable 5 is currently available to subscribers during an introductory access period. After that phase concludes, usage will move to a token-based pricing structure, with costs determined by the volume of input and output processed by the model.
Anthropic says its long-term goal is to eventually integrate Fable 5 into standard subscription offerings once computing capacity and infrastructure resources are sufficient to support broader availability. The company emphasized that expanding access while preserving safety remains a key priority as it continues developing more advanced AI systems.
