Anthropic released Claude Fable 5.1 on September 1 for coding, knowledge work and longer agent tasks. The company also announced Claude Mythos 5.1, but it is not a second model with separate underlying weights. Anthropic says the two names refer to the same model under different safeguards and access arrangements.
Fable 5.1 is generally available through Anthropic's API and supported cloud platforms. Mythos 5.1 is limited to trusted-access programs for cybersecurity and life-sciences research. A developer signing up for normal Fable access should not expect to receive Mythos capabilities or permissions.
Anthropic estimates that typical token-billed Fable workloads will cost about 25% less than Fable 5, chiefly because cached-input reads are cheaper. For highly agentic tasks it suggests savings as large as roughly 45%. Those are the company's workload estimates, not a uniform reduction in every token price or an independent completed-task cost comparison.
The launch also introduces a phased plan for Enterprise Frontier Safeguards, where customers control the infrastructure holding sensitive data while misuse monitoring remains available. That system was planned for later in the fall, not generally deployed on September 1.