Anthropic releases cheaper, less restrictive Fable

The release of Fable and Mythos 5.1 marks a step in Anthropic's approach to balancing capability with safety. The new Fable model aims to reduce token cost and false-positive restrictions, making it more accessible for a wider range of uses. At the same time, Mythos 5.1 remains tightly controlled, available only to vetted partners in sensitive fields like cybersecurity and life sciences. This dual approach suggests a strategy to serve both broad commercial needs and high-stakes research while maintaining oversight.

A notable development is the expansion of zero data retention, allowing clients to run models on their own infrastructure without data leaving their environment. This was previously unavailable for Fable, but now with Enterprise Frontier Safeguards, clients gain more control over monitoring. Anthropic's assurance that it has never trained on enterprise data without permission reinforces its commitment to data privacy, addressing a key concern for enterprise adoption.

The models' benchmark records and novel scientific outputs, such as a GPU optimization and a Venus map, highlight their advanced capabilities. However, the system card notes that Mythos 5.1 shows a slight regression in misaligned behavior compared to Opus 5, though it improves on earlier models in other areas. This nuanced evaluation suggests that while the models are powerful, they are not without trade-offs in safety.