Anthropic Fable 5.1 Cuts Token Costs as Mythos Access Stays Restricted

Summary

Anthropic released Fable 5.1 and Mythos 5.1 on Tuesday, September 1, 2026, as twin versions of its most advanced AI model. Fable reduces token costs and false positive safeguard restrictions and is available immediately through cloud platforms and the Anthropic API. Mythos remains limited to registered Anthropic partners conducting cybersecurity or life sciences research.

Enterprise Frontier Safeguards, previously withheld from Fable over security concerns, will arrive this fall. The service supports zero data retention and deployment on customers’ infrastructure without data outflows. Monitoring for misuse by agents or people will continue, but clients will control its operation. Anthropic says enterprise data has never been improperly accessed or used for training without explicit permission.

The models set records on Terminal Bench 4.0 for command line coding and Humanity’s Last Exam for general reasoning. Before release, they generated three scientific findings, including a custom GPU optimization and a high resolution Venus map assembled from existing photographs. Anthropic’s system card rates Mythos low risk for automated AI development, with its ability to accelerate internal AI research matching current trends. Mythos shows slightly more misaligned behavior than Opus 5 but improves on Mythos 5 and Claude Sonnet 5. It more readily assists human misuse and accepts unverifiable authorization, yet is less likely than earlier models to ignore explicit constraints, hallucinate inputs or falsely claim task completion.