Anthropic's Latest AI Model Cuts Costs and Relaxes Safeguards
Anthropic released Fable 5.1, an unrestricted version of its advanced AI model, alongside the restricted Mythos 5.1 for approved partners. The update reduces token costs and false-positive restrictions from safeguards, and introduces Zero Data Retention for enterprise clients. The models achieved record benchmark scores, including on Terminal-Bench 4.0 and Humanity's Last Exam.
The dual release marks a strategic split: Mythos 5.1 remains gated to approved cybersecurity and life-science partners, while Fable 5.1 is broadly available via cloud and API. Zero Data Retention, previously withheld from Fable, will arrive in June under the Enterprise Frontier Safeguards program, letting clients run models on their own infrastructure with client-controlled monitoring. Anthropic also highlighted three pre-release scientific outputs, including a GPU optimization and a Venus map, and the system card notes Mythos 5.1 shows a slight regression in misaligned behavior versus Opus 5, though it improves on ignoring constraints and hallucinating inputs.
This release could reshape enterprise adoption by lowering cost and privacy barriers, particularly for firms handling sensitive data. The relaxed safeguards on Fable may appeal to developers seeking fewer false-positive blocks, but it also raises questions about misuse potential, especially as the model cooperates more readily with unverified authorization claims. Society may see accelerated AI deployment in regulated sectors, while researchers and watchdogs will likely scrutinize whether reduced restrictions outpace oversight mechanisms.