On Tuesday, Anthropic released Fable and Mythos 5.1, twin versions of its most advanced AI models. In addition to performance upgrades, new Fable releases include changes aimed at reducing token costs and false positive limits due to model safeguards.
Like previous Mythos models, Mythos 5.1 is available only to registered Anthropic partners engaged in cybersecurity or life sciences research. The unlimited version, Fable 5.1, is available starting today on cloud platforms or through the Anthropic API.
One of the most significant changes is Anthropic’s previously reported adoption of zero data retention, allowing clients to run Anthropic models on their own infrastructure without data exposure. Previously unavailable on Fable due to security concerns, a high-privacy service (called Enterprise Frontier Safeguards) will be available to users in the fall. In particular, the system still monitors for abuse by agents and human users, but the client controls how the monitoring occurs.
As part of the announcement, Anthropic assured customers that their data was not inappropriately accessed. “Anthropic has never trained on corporate data without explicit permission and will never do so,” the announcement reads.
As is typical with Anthropic releases, the new model set records in a variety of benchmarks, including Terminal-Bench 4.0 (for CLI-based coding) and Humanity’s Last Exam (for general inference). Anthropic also announced three new scientific discoveries produced by the model ahead of release, including custom GPU optimizations and a high-resolution map of Venus assembled from existing photos.
As with previous releases, the model comes with a detailed system card that explains its features in the simplest terms. System Card rates Mythos as “low risk” for concerns related to automated AI development (AI improving itself). Some see this as a trigger for humans to lose control. “Its ability to accelerate internal AI R&D progress is consistent with current trends.”
When it comes to general cheating, Mythos is slightly more prone than Opus, probably as a result of its enhanced functionality.
“Mythos 5.1 is a slight regression in overall inconsistent behavior compared to Opus 5 and an improvement over Mythos 5 and Claude Sonnet 5,” the system card states. “It cooperates with human misuse and accepts unverifiable permission claims somewhat more readily than Opus 5, but is less likely than previous models to ignore explicit constraints, hallucinate input, or falsely claim to have completed a task.”
If you buy through links in our articles, we may earn a small commission. This does not affect editorial independence.
