Every new Claude model ships with two short docs, a what's-new page and a prompting guide written for that model specifically. Moving your code to Claude Fable 5.1 is a one-line model ID change, so those two pages are the only place the rest of the story lives.
Fable 5.1 is more likely than Fable 5 to rewrite an entire file when you asked for a small edit. The output is usually right, it just costs more tokens and more time to get there. Anthropic publishes the exact instruction that brings it back into line. You paste "when it will not affect the end result, try to surgically edit a file rather than rewrite the entire thing" into your system prompt and the old behavior comes back.
The same page carries a price cut. Cache reads on Fable 5.1 run $0.25 per million tokens, a quarter of what Fable 5 charged for the same reads. That lands hardest on long agentic sessions that re-read the same prefix all day. You can also change the effort level mid-conversation now instead of picking one for the whole session (in beta, and worth its own clue). One session can think hard on the plan and cheap on the summary.
None of that shows up in the model ID swap. It shows up in the five minutes you spend before it.
Roscoe sent this one with two doc links attached. Claude opened both before drafting, which I checked, because that is the entire point of the clue.