Physical AI Bench Understanding: leaderboard
Physical AI Bench understanding track for embodied reasoning over commonsense physics, space, time, and object interactions.
Metric: Overall Score (%). Source: huggingface.co. Status: saturation imminent. 25 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5 | 69.8 |
| 2 | Qwen 3 VL 235B A22B Instruct | 64.7 |
| 3 | Qwen 3 VL 235B A22B (Thinking) | 63.7 |
| 4 | Qwen 3 VL 32B Instruct | 62 |
| 5 | Qwen 3 VL 32B (Thinking) | 61 |
| 6 | Qwen 3 VL 30B A3B Instruct | 59.5 |
| 7 | GLM-4.5V | 59.2 |
| 8 | Qwen 3 VL 30B A3B (Thinking) | 57.3 |
| 9 | Qwen 3 VL 8B (Thinking) | 57.3 |
| 10 | Qwen 3 VL 8B Instruct | 56.8 |
| 11 | GPT-4o | 56.2 |
| 12 | Qwen 2.5 VL 32B Instruct | 55.3 |
| 13 | Qwen 2.5 VL 7B Instruct | 51 |
| 14 | Qwen 3 VL 2B Instruct | 48.7 |
| 15 | Claude 3.5 Sonnet | 46 |
Interactive version: theaggregate.ai/benchmark?slug=physical-ai-bench-understanding · How It Works · Data refreshed daily, snapshot 2026-09-05.