LinkS2Bench - Occlusion Attribution: leaderboard
Metric: Occlusion Attribution: multiple-choice accuracy (%) on whether two marked satellite regions occlude each other in the video on the full LinkS2Bench (17,903 questions over 1,500 UAV videos paired with satellite images), temperature 0; higher is better. Source: arxiv.org. Saturation forecast: Around February 2028. 18 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 47.3 |
| 2 | Qwen 3.5 35B A3B | 41.7 |
| 3 | Qwen 3.5 397B A17B | 41.2 |
| 4 | Claude Sonnet 4.6 | 41 |
| 5 | Seed 1.8 | 40.7 |
| 6 | GPT-5.4 | 37.1 |
| 7 | Qwen 3.5 9B | 36 |
| 8 | Kimi K2.5 | 27.3 |
| 9 | Qwen 3.5 4B | 26.4 |
| 10 | GPT-5.4 Mini | 25.3 |
| 11 | Ministral 3 3B | 18.9 |
| 12 | Ministral 3 14B | 17.9 |
| 13 | Llama 4 Scout | 14.3 |
| 14 | Ministral 3 8B | 12.8 |
Interactive version: theaggregate.ai/benchmark?slug=links2bench-occlusion-attribution · How It Works · Data refreshed daily, snapshot 2026-10-07.