MVBenchMultimodal

MVBench is a general video-understanding benchmark with 20 subtasks covering action recognition, temporal localization, video question answering, and other dimensions at varying difficulty levels.

Ranking Sources

#ModelScore

1

Gemini 3.7 FlashGoogle

82.2%

2

GPT-5.2OpenAI

78.1%

3

GLM-5.3-FlashZhipu AI

77.8%

4

Qwen3.5-397B-A17BAlibaba

77.6%

5

GPT-5.6 TerraOpenAI

75.0%

6

Qwen3.5-35B-A3BAlibaba

74.8%

7

Qwen3.6-35B-A3BAlibaba

74.6%

8

Gemini 3 ProGoogle

74.1%

9

DeepSeek-V4-Flash-Vision-ExpDeepSeek

69.4%

10

Claude Opus 4.5Anthropic

67.2%

11 models total

Back to benchmarks