008 — The 3D Response Surface

Model × Verbosity → Quality · drag to rotate

X-axis: Models (Haiku → Sonnet → Opus) Z-axis: Verbosity (Tight → Medium → Full) Y-axis: Composite Score

Averaged across both transcripts. Surface shows the quality "landscape" — note the dramatic cliff where Haiku drops on complex inputs, and how Sonnet Full forms the plateau at the summit.