multimodal benchmark
MMBench-Video
A long-form multi-shot benchmark for holistic video understanding that incorporates approximately 600 web videos from YouTube spanning 16 major categories, with each video ranging from 30 seconds to 6 minutes. Includes roughly 2,000 original question-answer pairs covering 26 fine-grained capabilities.
Updated Aug 7, 2026
Published models3
Registry coverage3
MetricScore
EvidenceB
MMBench-Video leaderboard
Sorted by the source-provided rank. Higher score is better according to the registry.
Score distribution
Top published rows on the benchmark's original scale.
MMBench-Video highlights
The top published results on this benchmark's own scale.
What is MMBench-Video?
Definition and scoring fields from the benchmark registry.
A long-form multi-shot benchmark for holistic video understanding that incorporates approximately 600 web videos from YouTube spanning 16 major categories, with each video ranging from 30 seconds to 6 minutes. Includes roughly 2,000 original question-answer pairs covering 26 fine-grained capabilities.
Scores are shown in ratio. The current registry marks this benchmark as not independently verified with evidence level B.
- Family
- MMBench-Video
- Modality
- multimodal
- Primary category
- multimodal
- Score direction
- higher
- LLMBoard eligible
- No
- Evaluation key
- mmbench-video|llm-stats-current
Source-native results are preserved. Eligibility for the overall LLMBoard score is a separate policy decision.
FAQ
Common questions about MMBench-Video.
Which model scores highest on MMBench-Video?
Qwen2.5 VL 72B Instruct is currently ranked first with 2.0%.
What does MMBench-Video measure?
A long-form multi-shot benchmark for holistic video understanding that incorporates approximately 600 web videos from YouTube spanning 16 major categories, with each video ranging from 30 seconds to 6 minutes. Includes roughly 2,000 original question-answer pairs covering 26 fine-grained capabilities.
Is a higher score better?
Yes. Higher values rank better for this benchmark.
How many models are compared?
3 unique published model results are currently shown.
Does this benchmark affect the overall score?
This benchmark is preserved as source-native evidence but is not eligible for the current overall score.