Anthropic Launches Claude Fable 5.1, Cutting Costs Up to 45% for Advanced Agentic Tasks
Anthropic’s latest models—Claude Fable 5.1 and Mythos 5.1—respond directly to customer concerns regarding price, data handling, and safety. The company reports that Fable 5.1 delivers superior performance compared with its predecessor, while incurring roughly a 25 percent discount in most scenarios and up to 45 percent savings for demanding agentic workloads, driven by lower costs for pre‑processed caching data.
The launch also generated enthusiastic early reactions, most notably from Every CEO Dan Shipper, who states, “It’s the strongest coding model we’ve tested, yet now it’s faster, token‑efficient, and truly converses like a regular human.”
Box CEO Aaron Levie also endorses the results, noting that his firm’s agent paired with Fable 5.1 captures nuanced details and hidden inconsistencies in data that were overlooked in tests with Fable 5. Separately, researcher Lisan al Gaib observes that Mythos 5.1 attains comparable performance to its prior iteration even when operating at low reasoning thresholds.
The updated Fable 5.1 introduces tighter guardrails designed to block fewer routine biological queries while maintaining safety similar to the previous version. Despite this improvement, Mythos 5.1 continues to enforce the prior strict bio‑containment policies. Anthropic also outlines enhancements to data stewardship, explaining that Enterprise Frontier Safeguards deliver full privacy by retaining customer‑owned data on third‑party cloud environments and schedule a broader release for the upcoming fall.
Additionally, Anthropic enables Fable 5.1 for automated software‑vulnerability discovery, although remaining sensitive exploitation activities such as penetration testing, exploit generation, and binary‑level flaw detection stay under the supervision of Opus‑tier models. Cloud‑wide deployment of Claude Fable 5.1 is now live, while Mythos 5.1 remains restricted to Project Glasswing participants.