Fable 5 is out of the ban — and it’s only here for about a week. So before it disappears, I ran the one test I actually care about: I gave Fable 5 and Sonnet 5 the exact same hard app to build — native macOS, one shot each, zero edits from me. One of them genuinely surprised me. Then I changed a single thing about the request, and it beat both.
This isn’t a benchmark roundup. We spend a couple of minutes on where Fable 5 and Sonnet 5 actually land — including my Care Benchmark, where “did the thing you asked for survive into the plan” matters more than the leaderboard number — and then we build. The app is a native SwiftUI radial launcher with real performance and animation demands, the kind of thing that used to be off the table in one shot. You’ll see Sonnet 5’s build, Fable 5’s straight build (and the single bug that turned up all day), and the version that changed my mind about which model to reach for. Cost and tokens at the end — including whether /goal is the token bonfire eve…
![]()