Benchmark
MMBench
A bilingual benchmark for assessing multi-modal capabilities of vision-language models through multiple-choice questions in both English and Chinese…
- Modality
- multimodal
- Categories
- multimodal, reasoning, vision
- Openness
- unknown
- Source
- llm_stats
- Reported scores
- 9
Reported scores
llm_stats
| Model | Organization | Reported value | Reported | Evidence |
|---|---|---|---|---|
| DeepSeek VL2 | DeepSeek | 0.796 | 2024-12-13 | evidence |
| DeepSeek VL2 Small | DeepSeek | 0.803 | 2024-12-13 | evidence |
| DeepSeek VL2 Tiny | DeepSeek | 0.692 | 2024-12-13 | evidence |
| Phi-3.5-vision-instruct | Microsoft | 0.819 | 2024-08-23 | evidence |
| Phi-4-multimodal-instruct | Microsoft | 0.867 | 2025-02-01 | evidence |
| Qwen2-VL-72B-Instruct | Qwen | 0.865 | 2024-08-29 | evidence |
| Qwen2.5 VL 72B Instruct | Qwen | 0.88 | 2025-01-26 | evidence |
| Qwen2.5 VL 7B Instruct | Qwen | 0.843 | 2025-01-26 | evidence |
| Step3-VL-10B | StepFun | 0.918 | 2026-01-15 | evidence |
Scores are partitioned by the source that reported them and are never merged into a single cross-source ranking, because the sources measure different things and say so.