Executive Summary
Anthropic has released Claude Fable 5, a new "Mythos-class" model, now available on the AI Gateway platform. The model is specifically designed to excel at long-running, ambiguous, and multi-step tasks that previously required significant human intervention, such as complex code review and bug analysis. Fable 5 includes built-in safety classifiers to prevent misuse but does not support Zero Data Retention, as Anthropic will retain data for 30 days to monitor for harmful use patterns.
Key Takeaways
* Product Name: Claude Fable 5, a "Mythos-class" model from Anthropic.
* Primary Function: Executes complex, end-to-end work over extended periods (multi-day runs) with reduced need for human check-ins.
* Enhanced Capabilities:
* Sustains productive output on long-running tasks.
* Dependably dispatches parallel sub-agents.
* Improved performance in code review, bug-finding, and repository investigation.
* Higher first-shot correctness on complex problems.
* Availability: Immediately available on AI Gateway. Developers can access it by setting the model parameter to `anthropic/claude-fable-5`.
* Safety & Data Policy:
* Includes blocking classifiers to refuse tasks related to offensive cybersecurity, biology, and other high-risk areas.
* Does not support Zero Data Retention; prompts and completions are retained for 30 days for misuse monitoring but are not used for training.
Strategic Importance
This launch positions Anthropic's models to tackle more autonomous, agent-like workflows, moving beyond conversational AI to solve complex, multi-step professional problems for developers and enterprise users.