Benchmarks / 3D
VoxelBench (image)
Models recreate a reference image as a 3D voxel build; humans vote head-to-head on the renders. Glicko-2 rating.
VoxelBench (image)
Higher is better · best effort per model
full table below
| # | Model | Effort | Score |
|---|---|---|---|
| 1 | GPT-6.1 Sol | max | 2192 |
| 2 | Gemini 3.1 Pro | high | 1552 |
| 3 | GPT-5.5 | xhigh | 1527 |
| 4 | Opus 4.6 | max | 1443 |
| 5 | Gemini 3 Pro | high | 1431 |
| 6 | GPT-5.4 | xhigh | 1388 |
| 7 | GPT-5.2 | xhigh | 1384 |
| 8 | Opus 4.5 | high | 1300 |
| 9 | Opus 4.7 | max | 1284 |
| 10 | GPT-5 | high | 1225 |
| 11 | Gemini 2.5 Pro | high | 1210 |
| 12 | GPT-5.1 | high | 1174 |
| 13 | GPT-5 Codex | high | 1172 |
| 14 | Opus 4.1 | high | 1137 |
| 15 | Gemini 2.5 | – | 1000 |
| 16 | Gemini 3.5 Flash | xhigh | 1000 |