LinkS2Bench - Event Geo-localization: leaderboard
Metric: Event Geo-localization: multiple-choice accuracy (%) on where in the satellite image a video event happens 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 April 2027. 18 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Seed 1.8 | 50.3 |
| 2 | Gemini 3.1 Pro (Preview) | 44.1 |
| 3 | GPT-5.4 | 44 |
| 4 | Claude Sonnet 4.6 | 32.7 |
| 5 | Kimi K2.5 | 32.7 |
| 6 | Qwen 3.5 397B A17B | 31 |
| 7 | GPT-5.4 Mini | 30 |
| 8 | Qwen 3.5 35B A3B | 29.8 |
| 9 | Qwen 3.5 9B | 18 |
| 10 | Ministral 3 3B | 16.9 |
| 11 | Ministral 3 14B | 15 |
| 12 | Ministral 3 8B | 13 |
| 13 | Qwen 3.5 4B | 8.7 |
| 14 | Llama 4 Scout | 4.1 |
Interactive version: theaggregate.ai/benchmark?slug=links2bench-event-geo-localization · How It Works · Data refreshed daily, snapshot 2026-10-07.