Model
Inference & Serving
Qwen2-VL
1Alibaba's 2vision-language 3model 4handles 5images, 6video
the six Ws · specification
W1
Who
Alibaba's Qwen team developed the Qwen2-VL model family.
W2
What
A multimodal model handling arbitrary image resolutions and long videos alongside text.
W3
Where
Released on Hugging Face and ModelScope under the Qwen organization.
W4
When
Qwen2-VL launched in August 2024.
W5
Why
It advanced open multimodal AI with dynamic-resolution vision processing and strong video reasoning.
W6
With
Runs on transformers, vLLM, or SGLang, using a redesigned ViT paired with the Qwen2 language model.
W7
Watch
Open-weight release by Alibaba; smaller sizes use Apache 2.0, the 72B variant uses the Qwen license with usage restrictions.
alternatives
works with