Anthropic released Fable 5.1 and Mythos 5.1 on Tuesday, updating its most advanced models with performance gains and tighter controls on token costs and safety filters.
Mythos 5.1 remains exclusive to registered partners working in cybersecurity or life sciences research. Fable 5.1, the unrestricted version, is available now via cloud platforms or the Anthropic API.
A major update is the rollout of zero data retention, allowing clients to run models on their own infrastructure without data leaving their premises. Previously blocked for Fable due to security concerns, a high-privacy service called Enterprise Frontier Safeguards will launch in June. The system continues to monitor for misuse by agents or human users, but clients now control how that monitoring occurs.
Anthropic confirmed that customer data has not been accessed inappropriately. “Anthropic has never trained on enterprise data without explicit permission, and never will,” the announcement states.
The new models set records in Terminal-Bench 4.0 for command-line coding and Humanity’s Last Exam for general reasoning. Anthropic also published three scientific findings generated by the models before launch, including a custom GPU optimisation and a high-resolution map of Venus built from existing photos.
As standard for a release, the models include a detailed system card explaining capabilities in plain terms. It rates Mythos as “low-risk” for automated AI development, where the system improves itself, noting that its ability to accelerate internal research is in line with current trends.
Regarding general misbehaviour, Mythos is slightly more prone to it than Opus, likely due to its enhanced capabilities.
“Mythos 5.1 is a slight regression on overall misaligned behavior compared to Opus 5, and an improvement over Mythos 5 and Claude Sonnet 5,” the system card reads. “It cooperates with human misuse and accepts unverifiable claims of authorization somewhat more readily than Opus 5, but it is less likely to ignore explicit constraints, hallucinate inputs, or falsely claim to have completed tasks than previous models.”
What it means
Developers can now run these models on their own hardware without sending data to Anthropic, provided they wait until June for the specific safeguards. This gives teams more control over where their data lives and how it is monitored, while still keeping the company’s safety checks active.
