Anthropic Unveils Fable 5.1 and Mythos 5.1, with Lower Costs and New Safety Features
Anthropic has announced the upcoming release of Fable 5.1, a new version of its most advanced AI model, alongside the restricted Mythos 5.1 variant, which remains limited to registered partners in cybersecurity and life sciences research. The company says the update is designed to improve performance in coding and scientific tasks while cutting costs and refining safety measures.
According to TechCrunch, the new Fable release includes changes meant to reduce token cost and false-positive restrictions from the model's safeguards. Mythos 5.1 will only be available to registered Anthropic partners engaged in either cybersecurity or life sciences research. Fable 5.1, the unrestricted version, is planned for release on Sept 1, as reported by The Star, and will be available on cloud platforms or through the Anthropic API.
One of the most significant changes, as reported by TechCrunch, is Anthropic's adoption of Zero Data Retention, which allows clients to run models on their own infrastructure without data outflows. This feature, previously unavailable for Fable due to security concerns, will now be offered through a high-privacy service called Enterprise Frontier Safeguards, set to roll out in June. The system will still monitor for misuse by agents or human users, but clients will have control over how the monitoring is conducted.
Anthropic also assured customers that their data has not been inappropriately accessed. The announcement states: "Anthropic has never trained on enterprise data without explicit permission, and never will."
In terms of pricing, The Star reports that Anthropic will charge enterprise users the same price per token for Fable 5.1 as for Fable 5, but is lowering the price for when the model resurfaces information it had previously processed. The company says this 75% price reduction will lead to a "substantial reduction" in the total cost of running the model. Dianne Penn, who leads product management for research and labs, explained that this part of the process can account for half or more of the tokens used in longer, more complex tasks. "This was one area we wanted to make improvements on, responding to customer feedback," she said.
Fable 5.1 also includes improved safety classifiers, which detect and block problematic commands. The company says these classifiers are now more accurate, with far fewer false positives.
TechCrunch notes that the new models set records on several benchmarks, including Terminal-Bench 4.0 for CLI-based coding and Humanity's Last Exam for general reasoning. Additionally, the models generated three novel scientific findings before their release, including a custom GPU optimization and a high-resolution map of Venus assembled from existing photos.
The accompanying system card, as per TechCrunch, rates Mythos as "low-risk" for concerns related to automated AI development, though it notes that Mythos 5.1 shows a slight regression on overall misaligned behavior compared to prior versions. The card states: "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."
US government safety concerns led to access restrictions for both AI models in June, according to The Star. The Star also reports that both Anthropic and OpenAI have filed confidential paperwork to begin trading publicly.