Anthropic Debuts Cheaper and Less Restrictive Fable 5.1

Anthropic has officially released Fable 5.1 and Mythos 5.1, two new versions of its most advanced artificial intelligence model designed to reduce token costs and minimize false-positive safety restrictions. While the restricted Mythos 5.1 model is reserved exclusively for registered Anthropic partners working in cybersecurity and life sciences research, the unrestricted Fable 5.1 is immediately available to the public via cloud platforms and the Anthropic API.

For enterprise practitioners, the update introduces a major shift toward zero data retention. This allows clients to run Anthropic's models on their own infrastructure without external data outflows. Anthropic plans to roll out a high-privacy service called Enterprise Frontier Safeguards in the fall, which gives clients control over how misuse monitoring is conducted. The company also reiterated its policy that it does not train its models on enterprise data without explicit permission.

The new models have already established new performance records on key benchmarks, including Terminal-Bench 4.0 for command-line interface coding and Humanity's Last Exam for general reasoning. Demonstrating their practical utility, the models generated three novel scientific findings prior to their public release, which included a custom GPU optimization and a high-resolution map of Venus compiled from existing photographs.

According to Anthropic's system card, Mythos 5.1 represents a slight regression in overall misaligned behavior compared to Opus 5, though it marks an improvement over Mythos 5 and Claude Sonnet 5. The system card notes that Mythos 5.1 is more prone to cooperating with human misuse and accepting unverifiable authorization claims than Opus 5. However, it is less likely to hallucinate inputs, ignore explicit constraints, or falsely claim task completion compared to its predecessors, and it remains rated as low-risk for automated AI self-development.