What's this about?
Fable 5.1 is priced the same per token as Fable 5, with 75% cheaper API cache reads. Anthropic estimates that works out to roughly 25% less for typical workloads, and up to about 45% for heavily agentic work.
Why do cache reads dominate agentic pricing?
Because an agent re-sends the same context on every step. Your system prompt, the codebase it has read, the tool definitions, the conversation so far — all of it goes back to the model each turn. Caching means the provider stores that processed prefix and charges a reduced rate to read it rather than reprocessing it, but at agentic volumes that reduced rate is still most of the bill. Cutting it by 75% therefore helps long-running agents far more than it helps one-off chat, which is exactly the differential Anthropic quotes.
Why is it worth noticing?
- It is a price cut aimed precisely at the workload the model is being sold for.
- It does not affect subscription metering, which is why the price story and the quota complaints coexist without contradicting each other.
- dotey's follow-up is worth knowing: with cache reads this cheap, Anthropic advises against frequent `/compact` — but if the cache has already expired (one hour on Claude Code subscriptions), compacting becomes expensive again.