LinkS2Bench - Occlusion Attribution: leaderboard

LinkS2Bench (LinkS²Bench, 2026) task on drone videos paired with satellite imagery: multiple-choice accuracy (%) on whether two marked satellite regions occlude each other in the video; higher is better.

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 August 2027. 18 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Gemini 3.1 Pro (Preview)47.3#102
2Qwen 3.5 35B A3B41.7#494
3Qwen 3.5 397B A17B41.2#270
4Claude Sonnet 4.641#179
5Seed 1.840.7#258
6GPT-5.437.1#138
7Qwen 3.5 9B36#838
8GLM-4.6V-Flash32.5#1049
9Kimi K2.527.3#272
10Qwen 3.5 4B26.4#1270
11Gemini-3.1-Flash [LinkS2Bench checkpoint unspecified]25.7
12GPT-5.4 Mini25.3#409
13Ministral 3 3B18.9#3606
14LLaVA-OneVision-1.5-8B18.1#1876
15Ministral 3 14B17.9#1885

No result here: #2 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=links2bench-occlusion-attribution · How It Works · Data refreshed daily, snapshot 2026-11-22.