Anthropic Launches Claude Fable 5.1 and Mythos 5.1

Anthropic has introduced Claude Fable 5.1 and Claude Mythos 5.1, splitting its latest model into a generally available product and a restricted-access version with different safeguards. The release combines lower token costs, customer-controlled enterprise privacy, and tighter controls for cybersecurity and life-sciences work.

Two Versions, One Base Model

Anthropic says Fable 5.1 and Mythos 5.1 are the same underlying model. The difference is the protection layer around it. Claude Fable 5.1 is generally available for coding, knowledge work, and long-running problem-solving tasks, while Mythos 5.1 is available through trusted access programs.

That split lets Anthropic expose stronger capabilities without treating every use case as equally safe. Mythos is intended for vetted cybersecurity and life-sciences work. Fable includes safety classifiers that can decline requests, making it the version designed for broad access through Anthropic’s products and cloud partners. The system card is the key reference for the model and safeguard boundaries.

Lower Cost for Long-Context Work

The commercial change is concentrated in cache reads: tokens a model can reuse after processing context. Anthropic says Fable 5.1 cache reads now cost 75% less, or $0.25 per million tokens. Other token prices remain $10 per million input tokens and $50 per million output tokens, according to the Claude platform.

Anthropic estimates overall costs will be about 25% lower than Fable 5 for typical workloads, with savings of up to approximately 45% for highly agentic tasks. That matters for coding agents and research workflows that repeatedly send project context, tool output, and intermediate instructions. Customers should still measure their own effort levels and tool use rather than treating the headline percentage as a universal discount.

Safeguards Become a Usability Feature

Anthropic is presenting safety precision as part of product usability. Its announcement says the latest cybersecurity safeguards block 60% fewer false positives than before. Fable 5.1 can identify software vulnerabilities for defensive work, but Anthropic says it does not permit the development of exploits.

In biology, the company says its latest safeguards fire 85% less often on benign elementary-biology and medical questions than the safeguards that launched with Fable 5. More advanced life-sciences research remains directed to Opus models, while qualified scientists can seek Mythos access through a program developed with the US government. Anthropic’s biology-safeguards update explains the distinction.

Privacy and Restricted Access

Enterprise Frontier Safeguards, or EFS, addresses the tension between misuse monitoring and customer confidentiality. Under the EFS design, customer data is stored in cloud infrastructure controlled by the customer rather than Anthropic. The company describes the arrangement as equivalent to zero data retention and says rollout will happen in phases beginning later this fall.

Anthropic says it developed EFS with more than 100 customers across industries and with AWS, Google Cloud, and Microsoft Azure. Mythos 5.1 is currently available only to vetted cyberdefenders and life scientists, with access limited to a set of US organizations. Cyberdefense applicants can register through the Claude Vulnerability Research Program.

What Enterprise Buyers Should Test

Fable 5.1 is more than a routine model refresh: it packages capability, cost, and control as one purchasing decision. Buyers should test their own long-context and agentic workloads, measure cache reuse, and confirm which deployment surfaces support EFS. They should also separate general availability from Mythos access; the two versions share a model, but not the same operating permissions.

Anthropic’s text-watermark work, alignment research, Responsible Scaling Policy, and earlier Fable and Mythos launch show how the company is building access and oversight into deployment. Business 2.0 has also examined AI accountability tooling, enterprise AI controls, confidential model evaluation, research agents, and AI governance. The practical question is whether an organization can deploy frontier capability at a predictable cost, with safeguards precise enough for legitimate work and strict enough for dangerous requests.