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

ModelOrganizationReported valueReportedEvidence
DeepSeek VL2DeepSeek0.7962024-12-13evidence
DeepSeek VL2 SmallDeepSeek0.8032024-12-13evidence
DeepSeek VL2 TinyDeepSeek0.6922024-12-13evidence
Phi-3.5-vision-instructMicrosoft0.8192024-08-23evidence
Phi-4-multimodal-instructMicrosoft0.8672025-02-01evidence
Qwen2-VL-72B-InstructQwen0.8652024-08-29evidence
Qwen2.5 VL 72B InstructQwen0.882025-01-26evidence
Qwen2.5 VL 7B InstructQwen0.8432025-01-26evidence
Step3-VL-10BStepFun0.9182026-01-15evidence

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.