What's this about?
An Anthropic engineer on Fable 5.1: it is a very good model, he has spent a lot of time with it, and a longer writeup is coming. The practical tip: try it on lower effort for tasks that need less verification or have fewer edge cases — and switching effort no longer breaks the prompt cache.
Why does "switching effort no longer breaks prompt cache" matter?
Because previously, changing the effort level invalidated the cached prefix, so every switch cost you a full reprocess of the entire context. That made effort a decision you made once at the start and lived with. Now you can run cheap for the mechanical parts of a long session and escalate only for the hard step, without paying to rebuild context each time. It converts effort from a session-level setting into a per-task one, which is where it belongs.
Why does this matter to you?
791 likes
- It is the single most actionable thing said about the release, and it comes from someone who worked on it.
- It pairs directly with the quota complaints — effort level is the main lever you control over burn rate.
- "Fewer edge cases" is the right heuristic: reasoning depth buys you correctness in the tail, and the tail is what you should be paying for.