CrossVid - Multi-View Reasoning: leaderboard
Mean accuracy over CrossVid's two multi-view tasks: spatial reasoning and object counting across drone videos of one scene from different viewpoints.
Metric: Accuracy (%). Source: raw.githubusercontent.com. Saturation forecast: Estimated already saturated. 16 models tracked. Updated 17 Nov 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | InternVL3-8B | 40.7 | #1756 |
| 2 | InternVL3-38B | 38.6 | #927 |
| 3 | GPT-4.1 | 38.4 | #505 |
| 4 | GLM-4.1V-9B (Thinking) | 37.8 | |
| 5 | Kimi-VL-A3B (Thinking) | 32.7 | |
| 6 | Qwen 2.5 VL 32B Instruct | 31.7 | #1074 |
| 7 | MiniCPM-o-2.6 | 31.4 | #2417 |
| 8 | Qwen2.5-Omni-7B | 29.6 | #1883 |
| 9 | Gemini 2.5 Pro | 28.7 | #274 |
| 10 | InternVL3-78B | 28.1 | #755 |
| 11 | LLaVA-Video-72B-Qwen2 | 27.9 | #1344 |
| 12 | Phi-3.5-Vision Instruct | 27.6 | #6271 |
| 13 | Video-R1-7B | 26.9 | #2053 |
| 14 | LongVA-7B-DPO | 26.3 | #2177 |
| 15 | VideoLLaMA3-7B | 19.8 | #2015 |
No result here: #1 Claude Fable 5.2, #2 GPT-6.2 Astra, #3 Gemini 4 Carbon.
Interactive version: theaggregate.ai/benchmark?slug=crossvid-multi-view-reasoning · How It Works · Data refreshed daily, snapshot 2026-11-17.