Drive-P2D: leaderboard
Drive-P2D (Tang et al., Zhejiang University, 2026) poses human-annotated selection questions on front-camera driving images from nuScenes, KITTI and BDD100K; models reason inside think tags and only the final answer is scored. The score is the unweighted mean in percent of six task accuracies (key object and object state; weather and lighting and special scene factors; optimal action and action safety) over 6,650 questions, zero-shot and averaged over two runs; higher is better.
Metric: Average score (%; unweighted mean of the six task accuracies, zero-shot, all scenarios; 6,650 questions on front-camera images from nuScenes, KITTI and BDD100K, mean of two runs). Source: arxiv.org. Saturation forecast: Estimated already saturated. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4.1 | 64.84 | #492 |
| 2 | Gemini 2.5 Pro | 57.36 | #262 |
| 3 | Qwen 2.5 VL 72B Instruct | 57.33 | #845 |
| 4 | Llama 3.2 90B Vision Instruct | 56.35 | #1757 |
| 5 | Phi-4 Multimodal Instruct | 51.89 | #3914 |
| 6 | Qwen 2.5 VL 7B Instruct | 50 | #1842 |
| 7 | Llama 3.2 11B Instruct | 49.55 | #5039 |
| 8 | Gemma 3 27B (IT) | 46.78 | #1306 |
| 9 | Gemma 3 4B (IT) | 34.1 | #3848 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=drive-p2d · How It Works · Data refreshed daily, snapshot 2026-11-12.