Drive-P2D - Object State (Object-2): leaderboard
Metric: Accuracy (%; determine a designated object's state, such as a traffic light's colour (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 |
|---|---|---|
| 1 | Gemini 2.5 Pro | 74.35 |
| 2 | GPT-4.1 | 73.74 |
| 3 | Qwen 2.5 VL 72B Instruct | 68.24 |
| 4 | Gemma 3 27B (IT) | 63.55 |
| 5 | Llama 3.2 90B Vision Instruct | 63.48 |
| 6 | Phi-4 Multimodal Instruct | 63.1 |
| 7 | Qwen 2.5 VL 7B Instruct | 59.34 |
| 8 | Llama 3.2 11B Instruct | 53.45 |
| 9 | Gemma 3 4B (IT) | 50.07 |
Interactive version: theaggregate.ai/benchmark?slug=drive-p2d-object-state-object-2 · How It Works · Data refreshed daily, snapshot 2026-09-26.