multimodal benchmark
MMBench
A bilingual benchmark for assessing multi-modal capabilities of vision-language models through multiple-choice questions in both English and Chinese, providing systematic evaluation across diverse vision-language tasks with robust metrics.
Updated Aug 7, 2026
Published models9
Registry coverage9
MetricScore
EvidenceB
MMBench leaderboard
Sorted by the source-provided rank. Higher score is better according to the registry.
| 1 | ST | 91.8% | 100.0% | 9 | C | |
| 2 | AC | 88.0% | 87.5% | 9 | C | |
| 3 | MI | 86.7% | 75.0% | 9 | C | |
| 4 | AC | 86.5% | 62.5% | 9 | C | |
| 5 | AC | 84.3% | 50.0% | 9 | C | |
| 6 | MI | 81.9% | 37.5% | 9 | C | |
| 7 | DE | 80.3% | 25.0% | 9 | C | |
| 8 | DE | 79.6% | 12.5% | 9 | C | |
| 9 | DE | 69.2% | 0.0% | 9 | C |
Score distribution
Top published rows on the benchmark's original scale.
MMBench highlights
The top published results on this benchmark's own scale.
What is MMBench?
Definition and scoring fields from the benchmark registry.
A bilingual benchmark for assessing multi-modal capabilities of vision-language models through multiple-choice questions in both English and Chinese, providing systematic evaluation across diverse vision-language tasks with robust metrics.
Scores are shown in ratio. The current registry marks this benchmark as not independently verified with evidence level B.
- Family
- MMBench
- Modality
- multimodal
- Primary category
- multimodal
- Score direction
- higher
- LLMBoard eligible
- Yes
- Evaluation key
- mmbench|llm-stats-current
Source-native results are preserved. Eligibility for the overall LLMBoard score is a separate policy decision.
FAQ
Common questions about MMBench.
Which model scores highest on MMBench?
Step3-VL-10B is currently ranked first with 91.8%.
What does MMBench measure?
A bilingual benchmark for assessing multi-modal capabilities of vision-language models through multiple-choice questions in both English and Chinese, providing systematic evaluation across diverse vision-language tasks with robust metrics.
Is a higher score better?
Yes. Higher values rank better for this benchmark.
How many models are compared?
9 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.