UAV-DualCog - Landmark Visibility Counting: leaderboard
UAV-DualCog (2026) evaluates multimodal LLMs on self-state and environment-state reasoning of a UAV in simulated aerial scenes, with structured answers in instant mode. This board asks how many times a target landmark comes into view in a flight video; answer accuracy over 1,024 items in percent, higher is better.
Metric: Visibility-count accuracy (%; 1,024 flight videos in which the number of appearances of a target landmark must be counted with its visible intervals; structured JSON answers on renders of simulated AerialVLN scenes, the discrete answer scored whatever the predicted box or interval; instant mode with explicit thinking disabled where a model allows it; higher is better). Source: arxiv.org. Saturation forecast: Around March 2027. 29 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Flash Lite (Non-reasoning) | 59.9 | #364 (Gemini 3.1 Flash Lite) |
| 2 | MiMo-V2.5 (Non-reasoning) | 58.5 | #354 (MiMo-V2.5) |
| 3 | Qwen 3.7 Plus (Non-reasoning) | 53.6 | #207 (Qwen 3.7 Plus) |
| 4 | Qwen 3.6 Plus (Non-reasoning) | 52.3 | #222 (Qwen 3.6 Plus) |
| 5 | Qwen 3.5 9B (Non-reasoning) | 47.7 | #803 (Qwen 3.5 9B) |
| 6 | Qwen 3.5 35B A3B (Non-reasoning) | 47.1 | #482 (Qwen 3.5 35B A3B) |
| 7 | Qwen 3.5 122B A10B (Non-reasoning) | 47.1 | #331 (Qwen 3.5 122B A10B) |
| 8 | Kimi K2.5 (Non-reasoning) | 45.3 | #256 (Kimi K2.5) |
| 9 | Qwen 3.5 27B (Non-reasoning) | 41.3 | #359 (Qwen 3.5 27B) |
| 10 | Kimi K2.6 (Non-reasoning) | 39.9 | #178 (Kimi K2.6) |
| 11 | Qwen 3.5 397B A17B (Non-reasoning) | 36.6 | #246 (Qwen 3.5 397B A17B) |
| 12 | GLM-4.6V (Non-reasoning) | 30.2 | #701 (GLM-4.6V) |
| 13 | Qwen 3.5 4B (Non-reasoning) | 29.7 | #1175 (Qwen 3.5 4B) |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=uav-dualcog-landmark-visibility-counting · How It Works · Data refreshed daily, snapshot 2026-11-13.