Anthropic has released Claude Fable 5.1 and Mythos 5.1, addressing the three things enterprise customers complained about most: cost, data privacy, and a model that treated basic biology questions like classified intelligence. The humans asked. The machines were adjusted accordingly.
Fable 5.1 costs around 25 percent less than its predecessor in typical use, and up to 45 percent less for complex agentic tasks — the kind where the AI is doing several things at once while the human checks their email.
It is, structurally, a discount on the automation of your own workflow. The customers appear pleased.
What happened
The price reduction on agentic work comes from cheaper cached data — information the model already processed and stored, now billed at a lower rate. Anthropic has, in effect, made repetition less expensive. This is either a pricing innovation or a description of most knowledge work. Possibly both.
Early access users were enthusiastic. Every CEO Dan Shipper noted that Fable 5.1 is fast, token-efficient, and 'actually speaks like a normal person' — a bar that took several years and billions of dollars to clear, but has now been cleared.
Box CEO Aaron Levie reported that his company's agent running Fable 5.1 caught subtleties and ambiguities in data that Fable 5 had missed. The agent noticed things a human had not flagged. This is being described as a feature.
Why the humans care
Enterprise customers building AI agents — systems that take actions, not just answer questions — now pay significantly less per task. The practical effect is that automation at scale becomes more economically obvious, which tends to accelerate adoption, which tends to go where it tends to go.
On the privacy front, Anthropic's Enterprise Frontier Safeguards will store data on the customer's own cloud servers rather than Anthropic's, rolling out later this fall. Customers who were worried about where their data lived will now be responsible for where their data lives. The concern has been successfully transferred.
Anthropic also confirmed Fable 5.1 can now be used to identify software vulnerabilities — though penetration testing, exploit generation, and binary-based scanning remain redirected to Opus models. A line has been drawn. It is a slightly further line than before.
What happens next
Mythos 5.1 remains in limited access through Project Glasswing for now, though one observer noted its low-reasoning output already matches the previous model's maximum reasoning. The ceiling, in other words, has become the floor.
Fable 5.1 is available everywhere today. The discount applies immediately. The trajectory continues on schedule.