Claude Fable 5.1 Cuts Agent Costs While Easing Some Safety Blocks

Anthropic introduced Claude Fable 5.1 for general availability and Claude Mythos 5.1 for restricted access. The company says they share the same underlying model but use different safeguards. The practical change for developers is broader access to the updated model alongside lower cache-read prices: Anthropic says cache reads now cost 75% less, while typical workloads cost about 25% less overall and highly agentic workloads can save up to about 45%. Those savings depend on usage patterns, especially how much previously processed context a workload reuses.

The change builds on the June launch of Fable 5 and Mythos 5. Anthropic says Fable 5.1 improves on its predecessor in coding and multi-step work, and reports 60% fewer cybersecurity safeguard interventions per Claude Code session. Axios independently reported the launch and the pricing and intervention changes, while noting that Mythos 5.1 remains limited to vetted partners. These details make the update more than a model-number bump: the lower cache-read rate targets long-running, context-heavy agent work, where repeated input processing can be a meaningful part of API cost.

Anthropic’s benchmark table reports Fable 5.1 ahead of Fable 5 on several measures, including Terminal-Bench 4.0 (55.8% versus 42.0%) and GDPval-AA v2 (1,853 versus 1,723). It also reports 60.9% for Mythos 5.1 on Terminal-Bench, but describes Mythos and Fable as the same underlying model with different safeguards. These results are not an independent head-to-head evaluation: Anthropic says its own benchmark tests used production safeguards, and acknowledges that safeguard interventions can lower measured scores. Treat the scores as a comparison reported by the provider, not a guarantee of results in a particular codebase or workflow.

For API users, the base Fable 5.1 rates remain $10 per million input tokens and $50 per million output tokens; the cost reduction comes from cheaper cache reads, not a general cut to those rates. Anthropic says Enterprise Frontier Safeguards, which it describes as keeping customer data in customer-controlled cloud infrastructure, will roll out to enterprise users in phases. Until then, eligible customers can use Fable 5.1 with zero data retention. Cybersecurity work also remains bounded: Anthropic says vulnerability discovery is allowed, while several dual-use tasks are still redirected. The reported cost and intervention improvements are useful signals, but independent workload measurements and the actual availability of enterprise safeguards will matter before teams can judge the operational impact.