claude-fable-5-1 — pricing, context window and limits

The top of the lineup, with the cheapest cache reads in it

Fable 5.1 replaced Fable 5 in the top tier at the same per-token price for input and output. The change that moves a real bill is the cache: a read costs a fraction of Fable 5's rate, the deepest cache discount of any Claude model. On a long agentic session that re-reads a large prefix every turn, most of the input bill is cache reads, so this one rate outweighs the headline price.

A deeper read discount has a side effect worth planning for: a cache miss now costs far more relative to a hit. Anthropic's own guidance for idle gaps that outlast the short TTL but not the long one is that a cheap keep-alive request on the default short TTL usually beats paying the pricier write for the longer TTL. See cache reads got cheaper on Fable 5.1 and Opus 5.5.

The breaking changes from Fable 5

Forced tool choice (any or tool) returns a 400. Thinking blocks are bound to the model that produced them — only Mythos 5.1 can read a Fable 5.1 block, so a fallback to another model runs without that reasoning. And editing an earlier turn invalidates every later thinking block; for accounts created on or after the enforcement date that edit is a 400. See thinking block bound to a different conversation.

Who it is for

Anthropic positions Opus 5.5 as the default and Fable 5.1 as the step up for the hardest long-running agentic and research work, or where Opus 5.5 at higher effort still falls short. It requires Anthropic's data-retention window, so zero-retention organisations cannot call it, and it has no Priority Tier.

Related

See migrating: Fable 5 to Fable 5.1 and Opus 5.5 vs Fable 5.1.