GeoDrive-Bench (Rule-Given): leaderboard
GeoDrive-Bench (2026) overall accuracy when the relevant clause of the country traffic handbook is placed in the prompt before a rule-application chain of thought, an upper-bound setting for rule knowledge. Percent; 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; Rule-Given prompting (the question-relevant rule clause from the country's traffic handbook is prepended, followed by a four-step rule-application chain); 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 | 77.2 | #1688 |
| 2 | Gemma 3 12B | 72.2 | #2000 |
| 3 | Qwen 2.5 VL 7B Instruct | 70.3 | #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-rule-given · How It Works · Data refreshed daily, snapshot 2026-11-13.