VoxelBench: leaderboard
VoxelBench — building 3D voxel structures from natural-language specs, human-rated; ladder rating.
Metric: Rating. Source: voxelbench.ai. Saturation forecast: Around December 2027. 59 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-6 Astra (Max) | 2602 | #6 (GPT-6 Astra) |
| 2 | GPT-6.1 Sol (Max) | 2573 | #7 (GPT-6.1 Sol) |
| 3 | Claude Opus 5.5 (Max) | 2522 | #3 (Claude Opus 5.5) |
| 4 | GPT-6 Sol (Max) | 2383 | #14 (GPT-6 Sol) |
| 5 | Claude Opus 5 (Max) | 2291 | #12 (Claude Opus 5) |
| 6 | GPT-5.6 Sol (Max) | 2233 | #19 (GPT-5.6 Sol) |
| 7 | Claude Fable 5 (Max) | 2090 | #16 (Claude Fable 5) |
| 8 | Grok 4.6 (xHigh) | 2067 | #26 (Grok 4.6) |
| 9 | DeepSeek V4.1 Flash (1017) | 2033 | #20 |
| 10 | GPT-5.5 Pro | 1989 | #17 |
| 11 | GPT-5.5 (xHigh) | 1932 | #31 (GPT-5.5) |
| 12 | GPT-5.6 Terra | 1929 | #41 |
| 13 | Kimi K3 (Max) | 1924 | #30 (Kimi K3) |
| 14 | Muse Wonder (Max) | 1882 | #15 (Muse Wonder) |
| 15 | DeepSeek V4.1 Flash | 1857 | #36 |
No result here: #9 Claude Fable 5.1, #10 Muse Wonder Full.
Interactive version: theaggregate.ai/benchmark?slug=voxelbench · How It Works · Data refreshed daily, snapshot 2026-10-20.