GeoDrive-Bench: leaderboard
GeoDrive-Bench (2026) tests region-specific driving reasoning in vision-language models: 5,053 human-validated multiple-choice questions on driving scenes from China, the US, the UK, Japan, Singapore and India, covering perception, prediction, planning and local traffic rules without country labels. Overall accuracy in percent with direct answering; higher is better.
Metric: Overall accuracy (%) on 5,053 human-validated multiple-choice questions on driving scenes from China, the US, the UK, Japan, Singapore and India (perception, prediction, planning and region-rule reasoning), without country labels; Direct prompting (the model answers with a letter in one pass, no intermediate reasoning); vLLM inference, 512 output tokens; higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | InternVL3-8B | 61 | #1688 |
| 2 | Gemma 3 12B | 60.1 | #2000 |
| 3 | Qwen 2.5 VL 7B Instruct | 57.9 | #1833 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=geodrive-bench · How It Works · Data refreshed daily, snapshot 2026-11-13.