SpatialViz-Bench
Multimodal · 2025-07-10
SpatialViz Bench evaluates spatial visualization capability in multimodal models. It asks models to reason over spatial arrangements and visual transformations.
Top models (higher is better)
| Model | Score |
|---|---|
| GPT-5.6 Sol | 80.7 |
| GPT-5.5 | 74.2 |
| Kimi K2.6 | 73.9 |
| Claude Fable 5 | 73.9 |
| Gemini 3.5 Flash | 73.9 |
| GPT-5.6 Luna | 71.3 |
| Kimi K3 | 71.2 |
| Muse Spark 1.1 | 70.3 |
| GPT-5.4 | 69.3 |
| Gemini 3.1 Pro Preview | 66.1 |
| GPT-5.2 | 65.8 |
| Gemini 3 Flash Preview | 65.3 |
| Opus 4.8 | 65.2 |
| GPT-5.6 Terra | 65.2 |
| Gemini 3 Pro Preview | 63.2 |
| Opus 4.7 | 62.6 |
| Grok 4.5 | 61.1 |
| GPT-5.4 Mini | 59.6 |
| Opus 4.6 | 55.7 |
| Sonnet 4.6 | 54.9 |
| GPT-5 | 54.1 |
| GPT-5 Mini | 52.5 |
| GPT-5.1 | 51.3 |
| Grok 4.3 | 49.6 |
| o3 | 47.7 |
| Gemini 2.5 Pro | 46.6 |
| Grok 4.20 | 45.6 |
| Gemini 3.1 Flash-Lite Preview | 44.8 |
| GPT-5.4 Nano | 44.6 |
| Opus 4.5 | 43.0 |
| Sonnet 4.5 | 41.6 |
| O1 | 41.4 |
| Grok 4.1 Fast | 38.5 |
| Haiku 4.5 | 38.4 |
| Gemini 2.5 Flash | 38.2 |
| Grok 4 Fast | 36.8 |
| GPT-5 Nano | 35.7 |
| Gemini 2.5 Flash-Lite Preview 06-17 | 35.2 |
| Grok 4 | 34.8 |
| Llama 4 Scout Instruct | 34.2 |