MineBench AI voxel build benchmark

Prompt
Build A
Retrieving build...
Build B
Retrieving build...
Revealed
Loading…
A
B

How it works

Models read a text prompt and output raw JSON coordinates for voxel blocks — no images, no 3D tools. Humans vote pair-wise and rankings emerge from Elo.

01Prompt
“A warm wooden cabin beside a pond, with a stone chimney, a small dock, and a few trees.”

Curated, natural-language prompts probe spatial reasoning.

02Generate
{
  "version": "1.0",
  "blocks": [
    {"x":0,"y":0,"z":0,"type":"oak_log"},
    {"x":1,"y":0,"z":0,"type":"stone"},
    …
  ]
}
1,247 blocks

Models output raw block coordinates. We render them directly — no post-processing.

03Vote & rank
A winsTieB wins
GPT-5.42150
Claude 4.52108
Gemini 3 Pro2091

Every vote feeds a live Elo leaderboard.

Want to test a model yourself?

Enter any prompt to generate a 3D build in the Sandbox.

Generate