CVSBench - CVUSA Satellite-to-Ground: leaderboard
CVSBench (2026) tests cross-view spatial reasoning with paired street-view and satellite images: models infer attributes of the unseen view and localize the street-view camera on the satellite image, on four-choice and binary questions. This board is accuracy in percent on CVUSA-subset questions that infer street-view attributes from a satellite image; higher is better.
Metric: Accuracy (%) on CVUSA-subset satellite-to-ground questions: from a satellite image, infer spatial attributes of the unseen street view; four-choice and binary-choice questions over paired street-view and satellite images (Gemini-2.5-flash-generated, human-verified), test split; higher is better. Source: arxiv.org. Saturation forecast: Not forecast. 14 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3 VL 4B Instruct | 62.4 | #1279 |
| 2 | GPT-4o | 62.1 | #741 |
| 3 | Qwen 3 VL 8B Instruct | 61.9 | #981 |
| 4 | GPT-5 Chat | 59.5 | #444 |
| 5 | Gemma 3 4B (IT) | 52.88 | #3831 |
| 6 | DeepSeek V3.2 | 47.03 | #406 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=cvsbench-cvusa-satellite-to-ground · How It Works · Data refreshed daily, snapshot 2026-11-13.