OmniTraffic - Vehicle Counting: leaderboard
An OmniTraffic task: vehicle counting (count vehicles per image or lane). OmniTraffic (2026) renders reconstructed real intersections as controllable 3D scenes and adds real roadside footage; accuracy in percent, higher is better.
Metric: Accuracy (%) on vehicle counting (count vehicles per image or lane), multiple-choice VQA on the 3,204-item human-verified OmniTraffic test set (12 reconstructed simulated intersections plus real roadside surveillance from South Korea and Tianjin), greedy decoding (at most 500 tokens), exact match against the answer letter with up to three attempts for invalid output (still invalid counts as wrong); higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro | 77.2 | #137 |
| 2 | Claude Sonnet 4.5 (Thinking) | 59 | #250 (Claude Sonnet 4.5) |
| 3 | GPT-5.2 | 55.6 | #176 |
| 4 | Qwen 3 VL 235B A22B (Thinking) | 55.5 | #451 |
| 5 | Gemini 2.5 Pro | 51.9 | #266 |
| 6 | Grok 4 | 47.6 | #307 |
| 7 | GPT-4o | 41.4 | #741 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=omnitraffic-vehicle-counting · How It Works · Data refreshed daily, snapshot 2026-11-13.