WildRoadBench - mAP: leaderboard
Metric: mAP (%) averaged over IoU thresholds 0.5 to 0.95, VLM Track, zero-shot box grounding on the full 1,061-image aerial set (1,394 ground-truth boxes); higher is better. Source: arxiv.org. Saturation forecast: Around 2031. 25 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3 Pro | 14.7 |
| 2 | Qwen 3.6 Plus | 12.4 |
| 3 | Gemini 3.1 Flash Lite | 11.4 |
| 4 | Qwen 2.5 VL 7B Instruct | 9 |
| 5 | Claude Sonnet 4.6 | 8.7 |
| 6 | Qwen 3 VL 8B Instruct | 8.2 |
| 7 | Kimi K2.5 | 7 |
| 8 | GPT-5.4 (2026-03-05) | 6.9 |
| 9 | Gemini 2.5 Flash | 5 |
| 10 | Qwen 3 VL 8B (Thinking) | 4.7 |
| 11 | InternVL3-8B | 2.9 |
| 12 | Claude Opus 4.6 | 2.8 |
| 13 | GPT-4o (2024-05-13) | 2.6 |
| 14 | InternVL3-14B | 2 |
| 15 | InternVL2-8B | 0.9 |
Interactive version: theaggregate.ai/benchmark?slug=wildroadbench-map · How It Works · Data refreshed daily, snapshot 2026-10-07.