Anthropic has released Claude Fable 5, a frontier-class AI model designed for general use that can autonomously handle complex, multi-hour tasks like code migration and research tool development. The model represents a significant milestone for the company, which first disclosed its Mythos-class capabilities in April and has now made a version safe enough for public access.

What Makes Claude Fable 5 Different From Previous Models?

Claude Fable 5 is built on Anthropic's Mythos architecture, a leap forward in AI capability that enables longer autonomous work sessions than any previous Claude model. The company also released an unrestricted version called Mythos 5 to members of its Project Glasswing program, which focuses on sensitive applications in cybersecurity and biology.

The practical implications are striking. Stripe reportedly used Fable 5 to migrate a 50-million-line Ruby codebase in a single day, a task that would have required two months of manual team effort. For vision tasks, Anthropic describes Fable 5 as "the new state-of-the-art model," noting that it successfully played through the entire Pokemon FireRed game using only visual input, a feat earlier models struggled to complete even with access to external tools.

Stripe

Wharton School professor and AI researcher Ethan Mollick provided additional examples of the model's capabilities. He described giving Fable 5 a 19-page specification document for building a survey analysis tool and watching it work for nine and a half hours to generate what he called an "extremely sophisticated" tool that researchers have needed for years but was never economically viable to create.

How to Access Claude Fable 5 and Understand Its Limitations?

• Pricing Structure: Fable 5 costs $10 per million input tokens and $50 per million output tokens via Anthropic's API, roughly double the cost of the previous-generation Claude Opus 4.8 model and more than three times the cost of Claude Sonnet 4.6.
• Free Trial Period: Users on Anthropic's Pro, Max, Team, and Enterprise subscription plans have free access to Fable 5 from now through June 22, after which they will need to pay for usage credits.
• Safety Guardrails: Anthropic redirects queries about cybersecurity, biology, chemistry, or AI model distillation to the older Claude Opus 4.8 instead, a restriction that triggers in less than 5% of interactions with the model.

However, Mollick noted that these safety restrictions may be overly cautious, suggesting they "trip at the faintest hint of a security problem," which could prevent even well-intentioned users from using Fable 5 to improve their own code security. Additionally, the model is restricted from assisting with cutting-edge AI and machine learning research, a limitation likely motivated by both safety concerns around AI self-improvement and competitive considerations with other AI labs.

Mollick

Anthropic says it will restore Fable 5 access through its subscription plans "as quickly as we can" once it has sufficient computing capacity to support the demand.

Anthropic

What Do These Capabilities Mean for AI Development?

The release of Claude Fable 5 marks a turning point in how frontier AI models are deployed to the public. Unlike previous releases that were either heavily restricted or limited to enterprise customers, Fable 5 is available to anyone willing to pay for API access, while still maintaining guardrails designed to prevent misuse in sensitive domains.

The real-world examples Anthropic and independent researchers have shared suggest that models at this capability level can handle work that previously required teams of engineers or researchers. The ability to work autonomously for extended periods, combined with strong reasoning and vision capabilities, opens new possibilities for software development, data analysis, and research workflows. Yet the company's cautious approach to safety restrictions reflects ongoing industry concerns about how powerful AI models should be deployed and what safeguards are necessary as these systems become more capable.