Urban Airspace Navigation - SPL: leaderboard

A 2026 benchmark of goal-oriented aerial navigation in urban 3D scenes: from RGB views, a model issues discrete horizontal and vertical flight actions to reach a described goal, over 5,037 human-annotated trajectories. This board is success weighted by path length in percent, averaged over short, middle and long trajectories; higher is better.

Metric: Success weighted by path length (%) averaged over short, middle and long goal-oriented aerial navigation episodes in urban 3D scenes; higher is better. Source: arxiv.org. Saturation forecast: Around April 2027. 12 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1GPT-5.132#240
2GPT-4.122.2#505
3O4 Mini18.5#368
4Gemini 2.5 Pro14.4#276
5Gemini 2.5 Flash14.2#463
6Gemini 2.0 Flash13.3#723
7GPT-4o12.8#751
8Qwen 3 VL Plus12.7#512
9Qwen-VL-Max12.6#1400
10GLM-4.6V11.9#710
11Qwen 2.5 VL 7B4.2#1862
12Qwen 2.5 VL 3B4#2639

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=urban-airspace-navigation-spl · How It Works · Data refreshed daily, snapshot 2026-11-21.