Anthropic releases Claude Fable 5.1 with cheaper agentic work

Anthropic has released Claude Fable 5.1, a new general-availability model for coding, knowledge work, and long-running tasks. The company says typical token-billed workloads should cost about 25% less than with Fable 5, with savings reaching roughly 45% for highly agentic work.

Fable 5.1 targets work that runs for hours

Fable 5.1 is positioned around multi-step execution rather than isolated answers. Anthropic reports gains in agentic coding, scientific research, computer use, and business workflows. Its own published evaluations show 55.8% on Terminal-Bench 4.0, 77.9% on the partial OSWorld 2.0 measure, and a GDPval-AA v2 score of 1,853. These are vendor-reported results, and the comparisons depend on harnesses, effort settings, and safeguard interventions.

The practical change is the combination of capability and persistence. Anthropic describes workflows that inspect large codebases, diagnose difficult failures, run experiments, and continue through long task sequences. That makes the model relevant to teams evaluating coding agents, research assistants, and automated internal operations—not only chat interfaces.

Lower cache-read pricing changes the economics

Anthropic is cutting cache-read pricing, which applies when the model reuses inputs already processed and stored. The company estimates that Fable 5.1 will cost 25% less than Fable 5 for typical token-billed workloads and up to approximately 45% less for highly agentic usage. TechSpot independently reports the launch and the reduced cached-token pricing, while noting that Fable 5.1 is available through Anthropic’s API, cloud platforms, and Claude desktop app.

The percentage is an estimate, not a universal discount on every request. Actual savings will depend on cache hits, prompt size, effort level, output volume, and the platform used to access the model. Teams should compare cost per completed task rather than only the headline input and output rates.

Fable and Mythos share a model but not access

Claude Mythos 5.1 uses the same underlying model with more permissive safeguards for vetted cybersecurity and life-sciences work. It is not generally available: Anthropic says access remains limited to trusted-access programs, including a Cyber Verification Program and a Life Sciences Verification Program developed with the US government.

Fable 5.1 is allowed to identify software vulnerabilities as defensive work, but Anthropic says penetration testing, exploit generation, and binary-based vulnerability scanning can still be redirected to other models. Mythos access therefore should not be treated as a general-purpose bypass or as a promise that every security workflow will run unchanged.

What developers should verify before switching

Check the serving platform’s exact model ID, context and output limits, cache terms, data-retention options, and safeguard behavior. Anthropic also says new API accounts will no longer be able to manually edit prior context while preserving the model’s earlier thinking in multi-turn conversations; existing accounts are not currently affected, and rollout is gradual.

For teams already using Fable 5, a representative evaluation should include long code changes, tool calls, failure recovery, and tasks that previously triggered cybersecurity or biology safeguards. The useful benchmark is the completed workflow at its actual effort setting and price.

Anthropic’s next milestone is broader enterprise rollout of Enterprise Frontier Safeguards later this fall. Until then, availability is broad for Fable 5.1, while Mythos 5.1 remains a controlled-access model.