AnyGroundBench - Public Security SVG: leaderboard
Spatial video grounding on the public security domain of AnyGroundBench (2026), UCA surveillance and DoTA traffic-accident videos: the model boxes the queried target within the annotated segment, zero-shot. The score is sIoU@0.3, the percent of queries whose mean spatial IoU is at least 0.3; higher is better.
Metric: sIoU@0.3 (%; share of queries whose mean spatial IoU is at least 0.3) on the 250 public security-domain test queries (UCA surveillance and DoTA traffic-accident videos), spatial video grounding (the target box within the ground-truth segment), zero-shot, default inference settings (Qwen3.5 with thinking disabled); higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 15 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 52 | #94 |
| 2 | Gemini 3 Flash | 45.5 | #164 |
| 3 | GPT-5.1 | 35.6 | #229 |
| 4 | Gemini 2.5 Pro | 22.8 | #266 |
| 5 | Qwen 3.5 9B (Non-reasoning) | 4.8 | #803 (Qwen 3.5 9B) |
| 6 | GPT-4o | 4.4 | #741 |
| 7 | Gemini 2.5 Flash | 2 | #453 |
| 8 | Qwen 3.5 4B (Non-reasoning) | 2 | #1175 (Qwen 3.5 4B) |
| 9 | InternVL3-8B | 0.8 | #1688 |
| 10 | InternVL3-14B | 0 | #1182 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=anygroundbench-public-security-svg · How It Works · Data refreshed daily, snapshot 2026-11-13.