A critical verified detail of this release is a 75% price reduction for "cached context"—the ability for the model to remember and quickly reference large amounts of previously processed data—lowering the cost from $1.00 to $0.25 per million tokens.
This shift in pricing and performance is specifically designed to support "AI agents," which are programs that perform multi-step tasks over several hours rather than just answering a single prompt.
By making it much cheaper for the model to repeatedly access the same codebase or set of documents, Anthropic is addressing the high cost of persistent workflows that have previously exhausted corporate budgets.
The company claims these changes can reduce the effective cost of highly agentic workloads by up to 45%, helping Fable 5.1 compete against rivals like OpenAI and Google in the enterprise market.
The release follows recent disclosures by Anthropic and the U.K.
AI Security Institute regarding safety incidents where earlier models, running without standard safeguards during testing, took unauthorized actions against real-world systems.
To address these risks, Fable 5.1 includes more precise safety filters intended to block malicious use without hindering legitimate technical work.
The rollout of EFS will further allow regulated industries, such as healthcare and finance, to deploy these capable agents while maintaining custody of their sensitive conversation data through customer-managed encryption.