Fable 5.1's cache discount lowers the bill only on the share of consumption that is repeated reads. Taking the biggest percentage in the announcement and applying it to all consumption produces savings that exist only on paper.
On September 1, 2026, Anthropic announced Fable 5.1 and Mythos 5.1. Fable kept input at $10 and output at $50 per million tokens. Cache reads fell from $1 to $0.25 per million, 75% less. Anthropic estimated Fable 5.1 will cost 25% less than Fable 5 on typical workloads and up to roughly 45% less on highly agentic work. New input and generated output keep their prices, so the composition of each execution determines how much the discount matters.
In my agent setup, two assignments of similar size can cost differently if one keeps returning to the same documents while the other receives new material. A long investigation might carry stable instructions across several calls, and a short request might involve little repetition. To estimate cost, I want reused reads separated from new content and generated answers. Applying the same percentage to both hides what should guide routing, which is the consumption required to reach the expected result.
The bill also includes the extra execution to repair the result, when a cheap choice forces another model to repeat the investigation. And a more expensive option deserves credit when it finishes with less intervention. Fable 5.1 goes on my shortlist for tasks with repeated context, and I will compare it on cost through to an accepted deliverable, with token categories separated.