Benchmarks · creative
VoxelBench (image)
Models recreate a reference image as a 3D voxel build; humans vote head-to-head on the renders. Glicko-2 rating.
13 models · 13 results · data as of 2026-11-06 · voxelbench.ai ↗
Leaderboard
GPT-6.1 Sol max
All results
13 rows| # | Model | Effort | VoxelBench (image) |
|---|---|---|---|
| 1 | GPT-6.1 Sol | max | 2229 ±131 |
| 2 | Gemini 3.1 Pro | high | 1578 ±108 |
| 3 | GPT-5.5 | xhigh | 1527 ±272 |
| 4 | Opus 4.6 | max | 1436 ±82 |
| 5 | Gemini 3 Pro | high | 1436 ±86 |
| 6 | GPT-5.4 | xhigh | 1411 ±90 |
| 7 | GPT-5.2 | xhigh | 1349 ±82 |
| 8 | Opus 4.5 | high | 1314 ±84 |
| 9 | Opus 4.7 | max | 1252 ±82 |
| 10 | Gemini 2.5 Pro | high | 1229 ±94 |
| 11 | GPT-5 | high | 1180 ±90 |
| 12 | Gemini 2.5 | 1000 ±686 | |
| 13 | Gemini 3.5 Flash | xhigh | 1000 ±686 |