LinkS2Bench - Zone Counting: leaderboard

LinkS2Bench (LinkS²Bench, 2026) task on drone videos paired with satellite imagery: mean relative accuracy (%, MRA over thresholds 0.50-0.95 with a zero-count rule) on the number of vehicles or pedestrians in a marked satellite region; higher is better.

Metric: Zone Counting: mean relative accuracy (%, MRA over thresholds 0.50-0.95 with a zero-count rule) on the number of vehicles or pedestrians in a marked satellite region 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: Not forecast. 18 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Qwen 3.5 9B74.8#838
2Kimi K2.573.5#272
3GLM-4.6V-Flash71.1#1049
4Gemini-3.1-Flash [LinkS2Bench checkpoint unspecified]68.8
5Claude Sonnet 4.667.3#179
6Ministral 3 14B64.1#1885
7Gemini 3.1 Pro (Preview)63.7#102
8Qwen 3.5 4B63.7#1270
9GPT-5.463.6#138
10Ministral 3 3B61.9#3606
11LLaVA-OneVision-1.5-8B61.5#1876
12Ministral 3 8B59.1#2166
13GPT-5.4 Mini54.4#409
14LLaVA-OneVision-1.5-4B53.3#1874
15Qwen 3.5 35B A3B42.6#494

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

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