VEBench - Technique Recognition: leaderboard

Five-option multiple choice from VEBench (2026) on which editing technique (J-cut, L-cut, cutaway, match, jump, invisible or smash cut) a real edited video uses. Accuracy in percent at each model default frame sampling; higher is better.

Metric: Technique Recognition accuracy (%): five-option multiple choice on the editing technique (J-cut, L-cut, cutaway, match, jump, invisible and smash cuts) used in a real edited video, each model at its default frame sampling; higher is better. Source: arxiv.org. Saturation forecast: Around 2028. 11 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Gemini 2.5 Pro34.65#276
2Qwen 3 VL 32B Instruct28.99#606
3Qwen 3 VL 8B Instruct28.64#1029
4Qwen 3 VL 4B Instruct28.6#1332
5Qwen 3 VL 30B A3B Instruct28.3#895
6Qwen 2.5 VL 7B Instruct27.47#1902
7InternVL3-14B27.08#1214
8InternVL3-8B25.21#1782
9GPT-4o24.68#751
10VEBench VideoLLaMA3 (checkpoint unspecified)23.2
11VITA-1.522.16#3083

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

Interactive version: theaggregate.ai/benchmark?slug=vebench-technique-recognition · How It Works · Data refreshed daily, snapshot 2026-11-21.