Drive-P2D - Key Object (Object-1): leaderboard
The Drive-P2D task of identifying the object that most influences the driving decision (single-choice). 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. Accuracy in percent (exact match), zero-shot over all scenarios and averaged over two runs; higher is better.
Metric: Accuracy (%; identify the object in the scene that most influences the driving decision (single choice); zero-shot, all scenarios, mean of two runs). Source: arxiv.org. Saturation forecast: Estimated already saturated. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 2.5 VL 72B Instruct | 68.06 | #845 |
| 2 | GPT-4.1 | 67.48 | #492 |
| 3 | Qwen 2.5 VL 7B Instruct | 65.58 | #1842 |
| 4 | Gemma 3 27B (IT) | 64.46 | #1306 |
| 5 | Gemini 2.5 Pro | 62.9 | #262 |
| 6 | Phi-4 Multimodal Instruct | 60.23 | #3914 |
| 7 | Llama 3.2 11B Instruct | 58.67 | #5039 |
| 8 | Llama 3.2 90B Vision Instruct | 51.39 | #1757 |
| 9 | Gemma 3 4B (IT) | 49.86 | #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-key-object-object-1 · How It Works · Data refreshed daily, snapshot 2026-11-12.