Embodied-BenchClaw Aerial-Bench: leaderboard
Aerial-Bench (2026) is a UAV and aerial-view spatial understanding benchmark built automatically by the Embodied-BenchClaw multi-agent system, covering object recognition, counting, image-plane relations, area and scale comparison and depth-based reasoning. The board is the unweighted mean in percent of the five question-type scores in the vision-language setting; higher is better.
Metric: Overall score (%), the unweighted mean of the five question-type scores; questions over aerial and UAV images from the 5,000-question benchmark built by the Embodied-BenchClaw pipeline (object and category recognition, counting with interval choices, image-plane spatial relations, visible-area and scale comparison, depth-based near-far reasoning), vision-language setting; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 14 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 | 74.13 | #49 |
| 2 | Gemini 3 Pro (Preview) | 71.83 | #113 |
| 3 | Kimi K2.5 | 71.8 | #256 |
| 4 | Qwen 3.6 27B | 71.27 | #341 |
| 5 | Qwen 3.6 35B A3B | 69 | #462 |
| 6 | Grok 4.20 0309 (Reasoning) | 68.57 | #230 (Grok 4.20 0309) |
| 7 | Claude Opus 4.7 | 67.23 | #79 |
| 8 | Claude Sonnet 4.6 | 67.07 | #161 |
| 9 | GPT-5.2 | 65.87 | #176 |
| 10 | GPT-4.1 | 63.13 | #495 |
| 11 | Claude Haiku 4.5 (20251001) | 51.33 | #536 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=embodied-benchclaw-aerial-bench · How It Works · Data refreshed daily, snapshot 2026-11-13.