Skip to content

QVQ-72B-Preview

An experimental Qwen model for visual reasoning

Visit official site

What QVQ-72B-Preview is

The model is described as an experimental research model from the Qwen team with stronger visual reasoning. The material states it does not fully replace Qwen2-VL-72B and can lose focus on image content in multi-step reasoning.

What you can do with it

  • Work on multidisciplinary reasoning tasks
  • Apply the model to mathematical reasoning
  • Try challenging problems on OlympiadBench

What to watch out for

  • Multi-step visual reasoning may lead to hallucination
  • Basic recognition tasks show no clear gain over Qwen2-VL-72B
  • Only single-turn dialogue and image output are supported

Pros & cons

✓ What we like

  • Multidisciplinary understanding and reasoning are named
  • Results on OlympiadBench are reported

! What to watch out for

  • Multi-step visual reasoning may lose focus on image content and cause hallucination
  • Only single-turn dialogue is supported and video input is not

FAQ

What is the MMMU score?

The material states the model scored 70.3 percent on MMMU.

Does it replace Qwen2-VL-72B?

The material states QVQ does not fully replace the abilities of Qwen2-VL-72B.

Last reviewed: 2026-11-15

More LLM API platform tools

View all →

How we review