The two models are based on the same architecture, but differ in their security restrictions. Fable 5.1 is available to all users, while Mythos 5.1 is only offered to verified professionals through Anthropic's trusted access programs. Mythos licenses are specifically intended for defensive cybersecurity teams and life sciences researchers.
According to Anthropic, the new version was designed to address three key issues that corporate customers had highlighted: cost, data protection, and overly strict security measures.
Fable 5.1 is expected to cost approximately 25% less than Fable 5 for typical workloads and up to 45% less for agent-intensive workloads. The reduction is primarily due to a 75% lower price for reading stored input tokens. This price is now $0,25 per million tokens. Input tokens continue to cost $10 per million, while output tokens cost $50 per million.
More control over data and fewer incorrect interventions
Anthropic also introduced its Enterprise Frontier Safeguards (EFS) framework. This will allow organizations to store their data entirely in their own cloud infrastructure, rather than on Anthropic’s servers. The company claims that this approach offers a level of privacy equivalent to a zero-retention policy, while still allowing for continued detection of malicious use.
EFS will be gradually available starting this fall for Claude Code and Claude Enterprise, the Claude platform, and major cloud providers including AWS, Google Cloud, and Microsoft Azure. Until then, eligible enterprise customers can already use Fable 5.1 with a zero retention policy.
In the cybersecurity space, Anthropic reports that false positives on related queries have been reduced by 60%. Fable 5.1 can now detect software vulnerabilities for defensive purposes, but cannot create exploits. Higher-risk, dual-use tasks, such as penetration testing and binary vulnerability scanning, are automatically redirected to Opus models.
Anthropic’s internal testing has shown that Fable 5.1 outperforms the previous model on programming, agent-based reasoning, and knowledge tasks. The tests include Terminal-Bench 4.0, GDPval-AA v2, and Humanity’s Last Exam. Companies like Jane Street, Cognition, MongoDB, Datadog, and Millennium have reported that the model solved problems that had plagued their engineering teams for years. In one case, it identified the cause of a rare production crash that people had been trying to solve for nearly five years.
Mythos 5.1 extends its capabilities to scientific research. In molecular biology tests, it designed high-affinity protein binders that outperformed the top entries in Adaptyv Bio’s design competitions. The success rate reached nearly 50%, compared to the typical industry benchmark of 10% to 15%.
In a separate experiment, Fable 5.1 created a new high-resolution elevation map of Venus using radar data from NASA's Magellan mission, which has been collected over decades. The work was completed ahead of upcoming ESA and NASA missions.
Mythos 5.1’s most advanced capabilities in biology and cybersecurity remain available only through Anthropic’s Cyber Verification Program and Life Sciences Verification Program. The company developed the programs in collaboration with the U.S. government and plans to gradually expand eligibility.
Anthropic also confirmed that it complies with the EU AI Act Code of Practice. Models released after August 2, 2026 will have invisible watermarking embedded in their outputs, while a detection API is in private beta for regulators and researchers. Fable 5.1 is available immediately on Claude platforms and via API, with the identifier claude-fable-5-1.
Your comments will not be published if: