DriveCombo-Text - L3 Dynamic Interactions: leaderboard
Metric: Accuracy (%) on DriveCombo Level 3 (two coexisting rules about other road participants) four-option multiple-choice questions on the 20% test split, rules from five countries' traffic codes; text-only variant in which the frames are replaced by a textual scene description; zero-shot, greedy decoding; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 14 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4.5 | 77.9 | #138 |
| 2 | GPT-5 Pro | 77.53 | #75 |
| 3 | Gemini 2.5 Pro | 75.41 | #145 |
| 4 | GLM-4.5V | 75.07 | #339 |
| 5 | GPT-5 Mini | 73.44 | #176 |
| 6 | GPT-5 Nano | 72.42 | #415 |
| 7 | Gemini 2.5 Flash | 72.04 | #237 |
| 8 | Gemma 3 27B | 69.51 | #596 |
| 9 | Gemma 3 12B | 59.39 | #666 |
| 10 | Llama 3.2 11B | 51.36 | #1183 |
| 11 | Gemma 3 4B | 50.93 | #1084 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=drivecombo-text-l3-dynamic-interactions · How It Works · Data refreshed daily, snapshot 2026-10-11.