Anthropic has launched two frontier artificial intelligence models, Claude Fable 5.1 and Claude Mythos 5.1, targeting software engineering, enterprise knowledge work, scientific research, and cybersecurity defense.
The models share the same core architecture but differ in the level of safeguards and access controls applied to potentially dual-use capabilities.
Claude Fable 5.1 is generally available through Anthropic’s platforms and API, while Claude Mythos 5.1 is restricted to vetted users in Anthropic’s trusted-access programs, including cybersecurity defenders and life sciences researchers.
Anthropic Launches Claude Fable 5.1 and Mythos 5.1
The rollout is designed to address enterprise concerns around AI operating costs, data privacy, and false-positive safety interventions that can limit legitimate defensive security work.
Anthropic said Claude Fable 5.1 is approximately 25% less expensive than Fable 5 for typical workloads and can be up to 45% cheaper in heavily agentic deployments.
The savings are primarily driven by a 75% reduction in cached-input read pricing, which now costs $0.25 per million tokens. Standard input and output token prices remain $10 and $50 per million tokens, respectively.
For enterprises with strict data-residency requirements, Anthropic is introducing its Enterprise Frontier Safeguards (EFS) framework.
The framework is intended to enable organizations to retain model-related data in their own cloud environments rather than on Anthropic’s infrastructure.
Anthropic said EFS is designed to provide privacy protections comparable to those in zero-data-retention environments while continuing to detect adversarial misuse patterns.
The company plans to deploy the framework in stages across Claude Code, Claude Enterprise, the Claude Platform, and cloud environments operated by AWS, Google Cloud, and Microsoft Azure. Eligible enterprise customers can already use Claude Fable 5.1 under a zero-retention policy.
Cybersecurity is a major focus of the release. Anthropic reported a 60% reduction in false-positive interventions involving cybersecurity-related prompts.
Under the updated policy, Fable 5.1 can help defenders identify software vulnerabilities and reason about defensive remediation. However, the model is still restricted from generating exploit code.
Higher-risk activities, including penetration-testing workflows and binary vulnerability scanning, are reportedly routed to Anthropic’s more tightly controlled Opus-class systems.
This approach reflects a growing industry trend toward tiered model access, where lower-risk defensive tasks are broadly available while advanced dual-use capabilities are subject to identity verification, usage monitoring, and contextual safeguards.
Anthropic’s internal evaluations indicate that Fable 5.1 improves on its predecessor across coding, agentic reasoning, and enterprise knowledge-work benchmarks, including Terminal-Bench 4.0, GDPval-AA v2, and Humanity’s Last Exam.
Early enterprise users, including Jane Street, Cognition, MongoDB, Datadog, and Millennium, reportedly used the model to resolve difficult engineering issues, including a rare production crash that had remained unresolved for nearly five years.
Mythos 5.1 expands these capabilities into advanced scientific research. Anthropic said the restricted model produced high-affinity protein-binder designs in molecular biology testing and achieved nearly a 50% hit rate, compared with typical industry benchmarks of 10% to 15%.
The company also said Fable 5.1 generated a high-resolution elevation map of Venus from historical NASA Magellan radar data, ahead of future ESA and NASA missions to Venus.
Access to Mythos 5.1’s advanced cybersecurity and biology functions is controlled through Anthropic’s Cyber Verification Program and Life Sciences Verification Program, developed in coordination with the U.S. government.
Anthropic also confirmed compliance with the EU AI Act Code of Practice, including invisible watermarking for outputs from models released after August 2, 2026.
Give your security team the visibility and context to investigate suspicious activity faster and contain threats before business impact grows. Strengthen Your Investigations with ANY.RUN
