VideoMME (w sub.)Multimodal

Video-MME (Video Massive Multimodal Evaluation) is a general video multimodal evaluation framework spanning short, medium, and long videos of up to one hour. Its 900 questions cover perception, understanding, and reasoning, with multiple evaluation modes including visual-only and subtitle-assisted settings.

Ranking Sources

#ModelScore

1

Qwen3.8-MaxAlibaba

90.4%

2

Kimi K3Moonshot AI

90.0%

3

GPT-5.6 SolOpenAI

89.5%

4

GPT-5.5OpenAI

89.4%

5

Seed2.1 ProByteDance

89.2%

6

Seed2.1 TurboByteDance

89.0%

7

Gemini 3.1 ProGoogle

88.4%

8

Gemini 3 ProGoogle

88.4%

9

Qwen3.7-PlusAlibaba

88.0%

10

Qwen3.6-PlusAlibaba

87.8%

23 models total

Back to benchmarks