On July 21, Alibaba's Tongyi Qianwen team released Qwen Image 3.0. This update doesn't focus on making images "look better," but rather emphasizes whether the generated images can be directly integrated into workflows for use in design, content creation, e-commerce, and education.

Can generate complex layouts in one go

According to the official statement, the new model can receive text commands with up to 4,500 tokens, approximately 4.5 times that of the previous generation. This means that users can describe more complex page structures in a single prompt, and the model can then directly output the complete image.

Alibaba showcased examples including newspaper layouts, storyboards, and high-density infographics. According to their official statement, this type of content can be generated in one go, rather than being split into multiple images and then stitched together in post-production.

Small print and formulas are key.

In addition to its long instruction processing capabilities, Alibaba also highlights "detail reproduction" as another selling point of this upgrade. The official statement claims that Qwen Image 3.0 can accurately render small text down to 10 pixels and handle details such as hair and skin texture.

Regarding text and academic content, the model also supports LaTeX formula typesetting, which can be used to generate academic page templates with formulas, charts, and explanatory text. The official documentation also states that the model supports native rendering in 12 languages and can simulate common visual formats such as web pages, games, and live streaming interfaces.

It is now available for trial use, but key information has not been disclosed.

Qwen Image 3.0 is currently available for trial at chat.qwen.ai. However, this release did not include open-source model weights, benchmark tables, or technical reports, and the API pricing has not yet been disclosed.

This differs from the release method of Qwen Image 1.0. The latter, upon launch, simultaneously provided open weight under the Apache 2.0 license and technical reports. In contrast, Qwen Image 3.0 currently relies more on official demo examples to illustrate its capabilities.

Based on the information disclosed so far, Alibaba hopes to position this model as an image tool that can be directly used in production, rather than just a generative model that emphasizes aesthetic expression. However, in the absence of unified test results and technical details, it is currently difficult for outsiders to quantify and compare its actual differences with other mainstream models.