VPCT (Visual Physics Comprehension): leaderboard
100 trivially easy (for humans) physics puzzles - balls rolling down ramps into buckets. Yet LLMs struggle badly, exposing a striking gap between impressive language capabilities and basic visual-spatial reasoning.
Metric: Accuracy (%). Source: epoch.ai. Status: saturation imminent. 38 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Best Expert | 100 |
| 2 | Median Human | 100 |
| 3 | Gemini 3 Pro (Preview) | 91 |
| 4 | GPT-5.2 (xHigh) | 84 |
| 5 | Gemini 3 Flash (Preview) | 72.6 |
| 6 | GPT-5.2 (High) | 67 |
| 7 | GPT-5 (High) | 66 |
| 8 | GPT-5 (Medium) | 63.2 |
| 9 | GPT-5.1 (High) | 58.7 |
| 10 | O4 Mini (Medium) | 57.5 |
| 11 | GPT-5.1 (Medium) | 53.3 |
| 12 | O3 (2025-04-16) (Medium) | 52 |
| 13 | Gemini 2.5 Pro (Preview 03-25) | 48 |
| 14 | Gemini 2.5 Pro (Preview 06-05) | 46.4 |
| 15 | Gemini 2.5 Flash (Preview 09-2025) | 46.2 |
Interactive version: theaggregate.ai/benchmark?slug=vpct-visual-physics-comprehension · How It Works · Data refreshed daily, snapshot 2026-09-05.