Anthropic launches Claude Fable 5.1 with lower token costs

Anthropic has launched Claude Fable 5.1 for coding, knowledge work, and long-running problem-solving tasks. The same underlying model is also offered as Claude Mythos 5.1, but that track remains limited to trusted-access programs with safeguards aimed at cybersecurity and life-science work.

The release changes three practical variables for teams evaluating Claude: capability, token economics, and data handling. Fable 5.1 is generally available, while Mythos 5.1 is not a general-access product.

Claude Fable 5.1 targets long-running engineering work

Anthropic positions Fable 5.1 as a model for substantial codebase research, feature development, and agentic workflows. Its published evaluation set covers agentic scientific research, terminal coding, multidisciplinary reasoning, computer use, business workflows, and general knowledge work.

The company reports that Fable 5.1 can discover software vulnerabilities without developing exploits for them. That distinction matters for security teams: the release is presented as a capability and safeguard change, not as permission to automate offensive exploitation.

Cache-read pricing is the main cost change

Anthropic says Fable 5.1 will cost about 25% less than Fable 5 for typical token-billed workloads because cache-read pricing is lower. Highly agentic workloads may see savings of up to approximately 45%, according to the announcement.

Those figures are workload-dependent estimates, not a universal subscription discount. Teams should compare their own cache-hit rates, effort settings, and tool-call patterns before updating a budget. Fable 5.1 defaults to High effort in Claude Code and Medium effort in Claude Cowork and Claude.ai.

GitHub Copilot exposes the access and retention trade-off

GitHub lists Claude Fable 5.1 for Copilot Pro+, Max, Business, and Enterprise users across its model picker and coding surfaces, with a gradual rollout. Business and Enterprise administrators must enable the policy; it is off by default.

GitHub also confirms a material privacy qualifier: Fable 5.1 requires data retention by default so Anthropic’s safety classifiers can operate. The retained prompts and outputs are not used to train Anthropic’s models. Eligible enterprise customers can receive time-bound zero-data-retention access while Anthropic rolls out Enterprise Frontier Safeguards.

Mythos 5.1 remains a controlled-access track

Mythos 5.1 uses the same underlying model with a different safeguard configuration. Anthropic says access is currently limited to trusted programs, including an upcoming biology access program developed with the US government. It should therefore be treated as a controlled research-access offering, not as a generally available API or consumer feature.

For teams deciding whether to adopt Fable 5.1, the immediate checklist is short: confirm the model is enabled for the account, review provider pricing for the actual endpoint, and get written approval for the required retention posture. The next major milestone is Anthropic’s phased rollout of Enterprise Frontier Safeguards later this fall.

Sources and methodology

This report uses Anthropic’s launch announcement as the primary source and GitHub’s independent Copilot changelog for availability and retention corroboration. Claims about pricing reductions and Mythos access are attributed to Anthropic and should be checked against current commercial terms before procurement decisions.