CrossVid - Comparative Analysis: leaderboard

Mean accuracy over CrossVid's four comparative tasks (behaviour, narrative, culinary and procedural-error comparison across videos).

Metric: Accuracy (%). Source: raw.githubusercontent.com. Saturation forecast: Estimated already saturated. 16 models tracked. Updated 17 Nov 2026.

Top models

#ModelScoreOverall rank
1Gemini 2.5 Pro54.7#274
2GPT-4.147.6#505
3GLM-4.1V-9B (Thinking)44.7
4Qwen 2.5 VL 32B Instruct38.3#1074
5LLaVA-Video-72B-Qwen233.9#1344
6Kimi-VL-A3B (Thinking)33.4
7InternVL3-78B33.1#755
8InternVL3-38B27.8#927
9Qwen2.5-Omni-7B26.7#1883
10MiniCPM-o-2.626.2#2417
11InternVL3-8B26.1#1756
12Phi-3.5-Vision Instruct25.9#6271
13LongVA-7B-DPO23.5#2177
14VideoLLaMA3-7B20.8#2015
15Qwen 2.5 VL 7B Instruct19.3#1893

No result here: #1 Claude Fable 5.2, #2 GPT-6.2 Astra, #3 Gemini 4 Carbon.

Interactive version: theaggregate.ai/benchmark?slug=crossvid-comparative-analysis · How It Works · Data refreshed daily, snapshot 2026-11-17.