~ same prompt, different LLM — every model gets the same brief in Blender, you judge the render ~
OpenAI / GPT 6 Astra / thinking:max has entered the arena NEW!
Parametric towerCategory: geonodesCapture: Prompt: v1Difficulty: ▮▮▮▮
TX: param-tower.md — composed prompt
── block: build-request-open@v2 (start, global) ──

Task

Complete the following Blender task:

<task>

Build a parametric high-rise tower generated entirely by a single Geometry Nodes modifier on a single object. No manual duplication, no per-floor modelling, no keyframes: the whole tower — every slab, column, and window inset — must be produced procedurally by the node tree so that changing its inputs rebuilds it.

Expose exactly two group inputs on that modifier, named exactly:

  • Floors — an integer-like input controlling how many stacked floors the tower has. Raising it adds floors and makes the tower taller; lowering it removes them. Give it a sensible default (around 12) and a range that stays readable (roughly 3 to 40).
  • Taper — a float controlling how much the tower narrows from base to top. At 0 the tower is a straight prism; as it increases the upper floors shrink inward so the silhouette tapers toward the top. Default around 0.3, range 0 to 1.

Changing either input in the modifier panel must visibly rebuild the tower — this is a hard requirement, not a nicety.

Design intent — a clean hard-surface architectural look: stacked horizontal floor slabs separated by a slim gap, vertical columns running the corners/edges, and regular window insets recessed into each floor's facade. Restrained and graphic — think a crisp architectural massing study, not an ornate building. Neutral materials, one subtle accent is fine.

The two exposed inputs ARE the deliverable. After you finish, the harness sweeps Floors and Taper across their ranges on camera to prove the tower is genuinely procedural, so make sure both inputs drive a smooth, correct rebuild across their whole range (no gaps, overlaps, or broken geometry at the extremes). Set up a camera and lighting that frame the whole tower cleanly for the presentation shot before declaring DONE.

── block: build-request-close@v2 (end, global) ──
</task>
RUNS ON THIS TASKview in gallery →
RUNPROMPTCONTESTANTSTATUSCOST
r1 · 2026-07-31v1OpenAI / GPT 5.6 Luna / thinking:maxok$0.04
r1 · 2026-07-17v1OpenAI / GPT 5.6 Lunaok$0.03
Votes cast on this bench: 000482 · 157 runs · 25 contestants
Best viewed with Netscape Navigator 4.07 at 800×600 · real runs, live ratings