Which is better, Qwen2 VL 72B Instruct or Qwen3 VL 4B (Reasoning)?

Refresh Models —Deals —Regulatory —Sources —

Questions this page answers

Which is better, Qwen2 VL 72B Instruct or Qwen3 VL 4B (Reasoning)?: Across 14 shared benchmarks, Qwen2 VL 72B Instruct scores higher on 8 and Qwen3 VL 4B (Reasoning) on 6. The widest gap is mathvision, where Qwen3 VL 4B (Reasoning) scores 60 against 25.9.

Qwen2 VL 72B Instruct vs Qwen3 VL 4B (Reasoning)

Across 14 shared benchmarks, Qwen2 VL 72B Instruct scores higher on 8 and Qwen3 VL 4B (Reasoning) on 6. The widest gap is mathvision, where Qwen3 VL 4B (Reasoning) scores 60 against 25.9.

AlibabavsAlibaba14 shared benchmarks8–6 head-to-head

Benchmark	Qwen2 VL 72B Instruct	Qwen3 VL 4B (Reasoning)
CC-OCR	68.7	73.8
chartqa	88.3	84
DocVQA_test	96.5	94.2
InfoVQA_test	84.5	83
mathvision	25.9	60
MathVista	70.5	79.5
MMMU (val) (Pass@1)	64.5	70.8
MMMU-Pro	46.2	57
MMStar	68.3	73.2
MV-Bench	73.6	69.3
OCRBench	877	81.6
RealWorldQA	77.8	73.2
TextVQA-val	85.5	80.5
Video-MME	77.8	59.7

Best tracked score per model per benchmark (default configuration; source-attributed). ↓ marks lower-is-better metrics. Open either model for its full surface, provenance and pricing.