Physical AI Bench - Understanding Overall: leaderboard
The "Understanding Overall" subtask of Physical AI Bench. SHI Labs' benchmark for physical-world AI video models — scores physical understanding and (conditional) video generation for physical plausibility and correctness.
Metric: Overall Score (%). Source: huggingface.co. Saturation forecast: Around January 2027. 25 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 69.8 | #98 |
| 2 | Qwen 3 VL 235B A22B Instruct | 64.7 | #273 |
| 3 | Qwen 3 VL 235B A22B (Thinking) | 63.7 | #238 (Qwen 3 VL 235B A22B) |
| 4 | Qwen 3 VL 32B Instruct | 62 | #283 |
| 5 | Qwen 3 VL 32B (Thinking) | 61 | #294 (Qwen 3 VL 32B) |
| 6 | Qwen 2.5 VL 72B Instruct | 60.8 | #377 |
| 7 | Qwen 3 VL 30B A3B Instruct | 59.5 | #384 |
| 8 | GLM-4.5V | 59.2 | #346 |
| 9 | Qwen 3 VL 8B (Thinking) | 57.3 | |
| 10 | Qwen 3 VL 30B A3B (Thinking) | 57.3 | #351 (Qwen 3 VL 30B A3B) |
| 11 | Qwen 3 VL 8B Instruct | 56.8 | #420 |
| 12 | GPT-4o | 56.2 | #343 |
| 13 | Qwen 2.5 VL 32B Instruct | 55.3 | #453 |
| 14 | Qwen 2.5 VL 7B Instruct | 51 | #654 |
| 15 | Claude 3.5 Sonnet | 46 | #350 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=physical-ai-bench-understanding-overall · How It Works · Data refreshed daily, snapshot 2026-10-20.