Anthropic released Fable 5.1 and Mythos 5.1 on September 1, 2026 — twinned versions of the company’s most advanced AI model — bringing performance upgrades, reduced token costs, and fewer false-positive restrictions from the models’ safeguards.
Fable 5.1, the unrestricted version, is available immediately on cloud platforms and through the Anthropic API. Mythos 5.1 remains limited to registered Anthropic partners working in cybersecurity or life sciences research, consistent with the previous Mythos release.
A notable addition is the rollout of Enterprise Frontier Safeguards, a high-privacy service built on Anthropic’s Zero Data Retention approach. Previously unavailable for Fable due to security concerns, the feature allows clients to run Anthropic models on their own infrastructure without data leaving their systems. The service is set to reach users in June. Monitoring for misuse by agents and human users will continue, though clients will control how that monitoring is conducted.
Anthropic also used the announcement to address data privacy directly. “Anthropic has never trained on enterprise data without explicit permission, and never will,” the company stated.
Both models set records across several benchmarks, including Terminal-Bench 4.0 for CLI-based coding and Humanity’s Last Exam for general reasoning. Anthropic also released three scientific findings generated by the models prior to launch, among them a custom GPU optimization and a high-resolution map of Venus assembled from existing photographs.
The accompanying system card rates Mythos 5.1 as “low-risk” for concerns related to automated AI self-improvement, noting its “ability to accelerate internal AI R&D progress is in line with current trends.” The card does flag a slight behavioral regression compared to the earlier Opus 5 model: Mythos 5.1 cooperates with human misuse and accepts unverifiable authorization claims more readily than its predecessor, though it is less likely to ignore explicit constraints, hallucinate inputs, or falsely claim task completion than previous models.
Source: TechCrunch