Anthropic Releases Claude Fable 5.1 to Slash Agentic Costs

Anthropic has officially released Claude Fable 5.1, a new model designed to address customer complaints regarding high costs, strict safety safeguards, and data retention. Alongside it, the company introduced Mythos 5.1, which is currently limited to participants in its Project Glasswing program. The standout feature of Fable 5.1 is its aggressive pricing structure. It is typically 25 percent cheaper than its predecessor, Fable 5, and offers savings of up to 45 percent for complex agentic workflows. These cost reductions are primarily driven by cheaper pricing for cached data that has already been processed and stored.

For developers and enterprise practitioners, the update brings notable performance improvements. Box CEO Aaron Levie reported that Fable 5.1 successfully identified data subtleties and ambiguities that Fable 5 overlooked. Meanwhile, Every CEO Dan Shipper praised it as their strongest coding model, noting that it "crucially actually speaks like a normal person." On the benchmarking front, Mythos 5.1 achieves a significant efficiency milestone: when set to low reasoning, it matches the benchmark scores of the previous Mythos model configured to Max reasoning.

Anthropic has also refined its safety guardrails. Fable 5.1 features "more precise safeguards" that are less prone to blocking benign biology queries, though Mythos 5.1 retains the stricter biology restrictions of the older version. Furthermore, Fable 5.1 is now authorized for identifying software vulnerabilities. However, highly sensitive cybersecurity operations like penetration testing, exploit generation, and binary-based vulnerability scanning will still be routed to Anthropic's Opus models.

To resolve data privacy concerns, Anthropic announced Enterprise Frontier Safeguards, which will roll out later this fall. This feature ensures "complete privacy" by keeping customer data stored on their own cloud servers rather than Anthropic's infrastructure. For practitioners, these updates mean they can deploy highly capable agents at a fraction of the previous cost, write code more naturally, and maintain strict control over proprietary data.