Anthropic says Claude Fable 5.1 is generally available with cache reads priced at $0.25 per million tokens, a 75% reduction. Axios also reported that Anthropic cut the price of reusing already-processed context by 75%, and framed it as a cost reduction for long-running agents. By Anthropic’s math, that takes cached reads from an implied $1.00 per million tokens to $0.25.
Why caching matters for agents. A long-running agent often sends the same material again at every step: instructions, tool definitions, a codebase, earlier conversation. Caching lets the model reuse that already-processed context at a lower price. The cheaper that reuse is, the less it costs to keep an agent working on a big task.
A simple illustration (hypothetical numbers): if an agent reads 10 million cached tokens during a task, that costs about $10.00 at $1.00 per million and $2.50 at $0.25 per million, a saving of $7.50 on that portion. Every million cache-read tokens costs $0.75 less, but only for tokens served from the cache.
That’s the catch. The cut applies only to cached reads. An application that mostly sends new context, or produces a lot of output, won’t see the same drop in its total bill. The sources don’t give prices or usage figures for uncached inputs, outputs, retries, failed tool calls, or extra turns, and they don’t say what share of typical coding, research, or automation workloads is served from cache.
Performance claims are unquantified. Anthropic says Fable 5.1 is stronger at long-running agentic work and that the lower cache price can mean bigger savings for those workloads. The material includes no benchmark results, task definitions, evaluation methods, or customer data comparing it with earlier releases. Anthropic’s documentation covers feature changes and compatibility details for Fable 5.1 and Claude Mythos 5.1, but the supplied description doesn’t list them, so check them before treating either as a drop-in migration.
Bottom line: The cache-read price is 75% lower, according to Anthropic’s own rate. What matters for your budget is what share of your application’s spending comes from cache reads. Measure that share before and after migrating, and you’ll know whether the cut changes your agent’s operating cost or only a small slice of it.
