VPCT
Multimodal · 2025-05-24
VPCT contains 100 elementary visual-physics problems in which a ball travels down ramps toward buckets. Models must predict which bucket will catch the ball, and scores are reported as accuracy.
Top models (higher is better)
| Model | Score |
|---|---|
| Gemini 3 Pro Preview | 91.0 |
| GPT-5.2 | 84.0 |
| Gemini 3 Flash Preview | 72.6 |
| GPT-5 | 66.0 |
| GPT-5.1 | 58.7 |
| O4 Mini | 57.5 |
| o3 | 52.0 |
| Gemini 2.5 Pro Preview 03-25 | 48.0 |
| Gemini 2.5 Pro Preview 06-05 | 46.4 |
| Gemini 2.5 Flash Preview (09-2025) | 46.2 |
| GPT-4.5 | 45.0 |
| Gemini Robotics-ER 1.5 | 40.8 |
| Gemini 2.5 Pro Preview 05-06 | 40.5 |
| GPT-5 Mini | 40.2 |
| Opus 4.5 | 40.0 |
| GPT-4o (2024-11-20) | 40.0 |
| Sonnet 4.5 | 39.8 |
| Claude 3.7 Sonnet | 39.0 |
| Opus 4 | 38.0 |
| Gemini 2.5 Flash Preview 04-17 | 38.0 |
| GPT-5 Nano | 37.2 |
| O1 | 37.0 |
| Opus 4.1 | 35.0 |
| Sonnet 4 | 34.0 |
| GPT-4o Mini | 34.0 |
| Claude 3.5 Sonnet (June 2024) | 33.0 |
| Claude 3.5 Sonnet (Oct 2024) | 33.0 |