Drive-P2D - Optimal Action (Decision-1): leaderboard

Metric: Accuracy (%; select the ego vehicle's optimal action (single choice); zero-shot, all scenarios, mean of two runs). Source: arxiv.org. Saturation forecast: Estimated already saturated. 11 models tracked.

Top models

#ModelScore
1GPT-4.156.95
2Qwen 2.5 VL 72B Instruct54.69
3Gemma 3 27B (IT)50.4
4Gemini 2.5 Pro49.47
5Gemma 3 4B (IT)45.78
6Llama 3.2 90B Vision Instruct45.14
7Phi-4 Multimodal Instruct44.58
8Llama 3.2 11B Instruct41.84
9Qwen 2.5 VL 7B Instruct40.87

Interactive version: theaggregate.ai/benchmark?slug=drive-p2d-optimal-action-decision-1 · How It Works · Data refreshed daily, snapshot 2026-09-26.