AI Research Atlas

QVQ-Max-Preview

Alibaba (Qwen) · 28 March 2025

First QVQ-Max version, a visual reasoning model whose MathVision accuracy rises as its thinking budget grows. It is hosted only.

A rebuilt visual reasoner after the experimental QVQ-72B-Preview, applying chain-of-thought to images and video for math, code and creative tasks. Alibaba shows accuracy improving with maximum thinking length.

Date
Friday, 28 March 2025
Lab
Alibaba (Qwen)
Kind
model
Access
app only

Blog dated 2025-03-28 in the feed. Access is inferred (Qwen Chat is the only surface linked in the text I read); no weights or numeric benchmarks were found.

Sources

  1. qwenlm.github.io/blog/qvq-max-preview/

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

Related