Alibaba Qwen has released a new visual reasoning model, QVQ-Max. The Alibaba Qwen team today released the visual reasoning model QVQ-Max on the social platform X. As an upgraded version of QVQ-72B-Preview, the Alibaba Qwen team said that the new model is targeted at...

Alibaba Qwen Releases New Visual Reasoning Model QVQ-Max

The Alibaba Qwen team today released the visual reasoning model QVQ-Max on the social platform X. As an upgraded version of QVQ-72B-Preview, the Alibaba Qwen team said that the new model has optimized for the shortcomings of traditional artificial intelligence in visual information processing, enhancing the ability from visual perception to cognitive reasoning. QVQ-Max supports joint reasoning of images, videos, and text. In the MathVision benchmark test, QVQ-Max exhibits the characteristic that "thinking length" is positively correlated with accuracy, verifying the model's potential in complex multimodal tasks. The official said that QVQ-Max stands out in three aspects: first, meticulous observation, capable of accurately identifying details and text markings in images; second, in-depth reasoning, analyzing and reasoning by combining background knowledge; third, flexible application, supporting creative generation and content creation.

—— NetEase Technology, Alibaba Qwen, Github

via Windvane Reference Express - Telegram Channel