AI Research Atlas

Qwen2-VL (2B, 7B, 72B)

Alibaba (Qwen) · 29 August 2024

Qwen2-VL: understands 20+ minute video and any image resolution; 2B and 7B Apache 2.0, 72B API-only at launch.

Second-generation Qwen vision-language model on Qwen2 with dynamic-resolution inputs, video question answering over 20+ minutes, device-operating agent use and multilingual text in images. The 72B was served by API at launch and opened on 2024-09-19.

Date
Thursday, 29 August 2024
Lab
Alibaba (Qwen)
Kind
open-weights
Access
open weights

Figures

MeasureValueMeasured by
Video length understood20+ minutes
Company claim; 2B and 7B open under Apache 2.0
company

Blog dated 2024-08-29. README news dates the 2B/7B release 2024-08-30 and the open 72B instruct model 2024-09-19. Architecture details (M-RoPE, dynamic resolution) and benchmark tables were not opened.

Sources

  1. qwenlm.github.io/blog/qwen2-vl/
  2. github.com/QwenLM/Qwen3-VL
  3. en.wikipedia.org/wiki/Qwen

This record was checked against its sources on 6 October 2026. How we check

Related