OmniTraffic: leaderboard
OmniTraffic (2026) tests traffic-scene understanding in multimodal models with 3,204 human-verified multiple-choice questions over reconstructed intersections and real surveillance footage, from vehicle perception through multi-view and temporal reasoning to signal-phase decision support. The score is overall accuracy; higher is better.
Metric: Overall accuracy (%) over all eleven task categories, 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 | 72.4 | #137 |
| 2 | Grok 4 | 59.8 | #307 |
| 3 | Claude Sonnet 4.5 (Thinking) | 59.7 | #250 (Claude Sonnet 4.5) |
| 4 | Qwen 3 VL 235B A22B (Thinking) | 59.1 | #451 |
| 5 | GPT-5.2 | 57.1 | #176 |
| 6 | GPT-4o | 55 | #741 |
| 7 | Gemini 2.5 Pro | 51 | #266 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=omnitraffic · How It Works · Data refreshed daily, snapshot 2026-11-13.