VEBench: leaderboard

VEBench (2026) tests large multimodal models on real-world video editing tasks. The overall score is the mean of Technique Recognition accuracy and Footage Selection accuracy with subtitles, in percent; higher is better.

Metric: Overall score (%): mean of Technique Recognition accuracy and Footage Selection accuracy with subtitles; higher is better. Source: arxiv.org. Saturation forecast: Around February 2028. 9 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Gemini 2.5 Pro39.55#280
2Qwen 3 VL 32B Instruct34.36#607
3GPT-4o33.29#754
4Qwen 3 VL 8B Instruct32.39#1043
5Qwen 3 VL 4B Instruct30.2#1348
6InternVL3-14B29.12#1227
7Qwen 3 VL 30B A3B Instruct28.84#903
8Qwen 2.5 VL 7B Instruct28.74#1919
9InternVL3-8B27.99#1795

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

Interactive version: theaggregate.ai/benchmark?slug=vebench · How It Works · Data refreshed daily, snapshot 2026-11-22.