Benchgen
Models/alibaba/

QVQ-72B-Preview

DraftPublic

Model Details

QVQ-72B-Preview

Organization Modality License Released

Quick answer: QVQ-72B-Preview is Alibaba's open-weight visual reasoning model, combining the Qwen2-VL architecture with extended chain-of-thought reasoning, and the top open-source scorer among tested models on the MedXpertQA expert-level medical benchmark.

At a Glance

Where QVQ-72B-Preview leads

  • Top open-source score on MedXpertQA among tested models
  • Extends Qwen2-VL with slow-thinking, multi-step visual reasoning

Where it lags

  • A research preview, not a production-stable release; single-round conversation only

Best for: open-weight visual reasoning tasks requiring multi-step chain-of-thought.

What QVQ-72B-Preview Is

QVQ-72B-Preview is an experimental open-weight visual reasoning model from Alibaba's Qwen team, built on Qwen2-VL and trained to produce extended chain-of-thought reasoning before answering visual questions.

Specifications

FieldValue
OrganizationAlibaba
ModalityMultimodal (text + vision)
LicenseOpen weights (research preview)
Release dateDecember 2024