DriveCombo - L5 Rule Conflicts: leaderboard
Metric: Accuracy (%) on DriveCombo Level 5 (two conflicting rules resolved by the legal priority order) four-option multiple-choice questions on the 20% test split, rules from five countries' traffic codes; input is the question with four RGB frames of the simulated driving scene; zero-shot, greedy decoding; higher is better. Source: arxiv.org. Saturation forecast: Around 2030. 14 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 Pro | 44.19 | #75 |
| 2 | Claude Sonnet 4.5 | 43.99 | #138 |
| 3 | Gemini 2.5 Pro | 43.06 | #145 |
| 4 | GLM-4.5V | 41.86 | #339 |
| 5 | Gemini 2.5 Flash | 39.33 | #237 |
| 6 | GPT-5 Mini | 38.1 | #176 |
| 7 | Gemma 3 27B | 37.42 | #596 |
| 8 | GPT-5 Nano | 33.24 | #415 |
| 9 | Gemma 3 12B | 29.54 | #666 |
| 10 | Gemma 3 4B | 26.66 | #1084 |
| 11 | Llama 3.2 11B | 25.74 | #1183 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=drivecombo-l5-rule-conflicts · How It Works · Data refreshed daily, snapshot 2026-10-11.