SpatialBabel: leaderboard
Vision-language models (VLMs) exhibit a striking paradox: they can generate executable code that reconstructs a 3D scene from geometric primitives with correct object counts, classes, and approximate positions, yet the same models fail at simpler spatial questions on the same image. We show that 3D geometric primitives (cubes, spheres, cylinders, expressed in executable code) serve as a powerful i
Metric: Three.js (self-reported). Source: benchmarklist.com. Saturation forecast: Around June 2027. 14 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro | 83.4 | #80 |
| 2 | Claude Opus 4.7 | 79.2 | #49 |
| 3 | Gemini 2.5 Pro | 73.2 | #152 |
| 4 | GPT-4o | 73 | #343 |
| 5 | Claude Sonnet 4.6 | 69.3 | #90 |
| 6 | Gemini 3 Flash | 68.4 | #99 |
| 7 | Gemini 2.5 Flash | 67.2 | #245 |
| 8 | GPT-5 | 64.6 | #98 |
| 9 | Qwen 2.5 VL 7B | 59.9 | #643 |
| 10 | InternVL3-8B | 54.1 | #620 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=spatialbabel · How It Works · Data refreshed daily, snapshot 2026-10-20.