CrossVid - Temporal Understanding: leaderboard

Mean accuracy over CrossVid's three temporal tasks: inferring plot, aligning functional steps and ordering procedural steps 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 Pro56#274
2GPT-4.146.7#505
3Video-R1-7B26.9#2053
4Qwen 2.5 VL 32B Instruct26.5#1074
5MiniCPM-o-2.626.4#2417
6GLM-4.1V-9B (Thinking)23.1
7LLaVA-Video-72B-Qwen222#1344
8Qwen2.5-Omni-7B21.6#1883
9InternVL3-8B20.3#1756
10Qwen 2.5 VL 7B Instruct20#1893
11Kimi-VL-A3B (Thinking)17.9
12Phi-3.5-Vision Instruct17.2#6271
13InternVL3-78B15.6#755
14InternVL3-38B10.1#927
15LongVA-7B-DPO7.5#2177

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

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