Drive-P2D - High-Risk Scenarios: leaderboard

Metric: Average score (%; unweighted mean of the six task accuracies on the high-risk split, scenes whose mean rating by two annotators is at least 4 on a five-point risk scale, about 1.6K questions; zero-shot, mean of two runs). Source: arxiv.org. Saturation forecast: Estimated already saturated. 11 models tracked.

Top models

#ModelScore
1GPT-4.171.82
2Gemini 2.5 Pro64.66
3Llama 3.2 90B Vision Instruct59.24
4Qwen 2.5 VL 72B Instruct58.25
5Llama 3.2 11B Instruct49.52
6Qwen 2.5 VL 7B Instruct49.4
7Gemma 3 27B (IT)48.01
8Phi-4 Multimodal Instruct46.77
9Gemma 3 4B (IT)36.84

Interactive version: theaggregate.ai/benchmark?slug=drive-p2d-high-risk-scenarios · How It Works · Data refreshed daily, snapshot 2026-09-26.