multimodal benchmark
MME
A comprehensive evaluation benchmark for Multimodal Large Language Models measuring both perception and cognition abilities across 14 subtasks. Features manually designed instruction-answer pairs to avoid data leakage and provides systematic quantitative assessment of MLLM capabilities.
Updated Aug 7, 2026
Published models3
Registry coverage3
MetricScore
EvidenceB
MME leaderboard
Sorted by the source-provided rank. Higher score is better according to the registry.
| 1 | DE | 22.5% | 100.0% | 3 | C | |
| 2 | DE | 21.2% | 50.0% | 3 | C | |
| 3 | DE | 19.1% | 0.0% | 3 | C |
Score distribution
Top published rows on the benchmark's original scale.
MME highlights
The top published results on this benchmark's own scale.
What is MME?
Definition and scoring fields from the benchmark registry.
A comprehensive evaluation benchmark for Multimodal Large Language Models measuring both perception and cognition abilities across 14 subtasks. Features manually designed instruction-answer pairs to avoid data leakage and provides systematic quantitative assessment of MLLM capabilities.
Scores are shown in ratio. The current registry marks this benchmark as not independently verified with evidence level B.
- Family
- MME
- Modality
- multimodal
- Primary category
- multimodal
- Score direction
- higher
- LLMBoard eligible
- Yes
- Evaluation key
- mme|llm-stats-current
Source-native results are preserved. Eligibility for the overall LLMBoard score is a separate policy decision.
FAQ
Common questions about MME.
Which model scores highest on MME?
DeepSeek VL2 is currently ranked first with 22.5%.
What does MME measure?
A comprehensive evaluation benchmark for Multimodal Large Language Models measuring both perception and cognition abilities across 14 subtasks. Features manually designed instruction-answer pairs to avoid data leakage and provides systematic quantitative assessment of MLLM 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 marked as eligible for the current LLMBoard capability methodology.