Step-by-step reasoning helped QVQ-72B-Preview solve visual problems, but could also make it lose track of the image. Qwen built this experimental open weightsA model whose trained parameters are published for anyone to download and run — unlike API-only models you can access but never possess.Full definition → model on Qwen2-VL-72B and reported improved results on visual mathematics benchmarks. Its release account also identifies a specific failure: during extended reasoning, the model can drift away from the visual evidence and invent details. That limits what a fluent explanation establishes. An answer to a diagram-based problem still depends on whether the model preserves the diagram's actual relationships throughout its reasoning.
Terms in this piece · Glossary
open weights — A model whose trained parameters are published for anyone to download and run — unlike API-only models you can access but never possess.
Key quotes
“Language and vision intertwine in the human mind, shaping how we perceive and understand the world around us.”
