Alibaba has unveiled a new generative artificial intelligence (AI) model that can produce complex workplace images such as newspaper layouts, infographics and storyboards in one go. The model prioritises productivity that can be used immediately in real work settings over aesthetic polish.

On July 22, blockchain outlet Decrypt reported that Alibaba launched an image-generation model called Qwen Image 3.0 that can process instructions of up to 4,500 tokens.

Alibaba said 4,500 tokens is 4.5 times more than the previous generation. Users can create a finished layout by entering multiple panels, charts, formulas, explanatory text and fine print in a single prompt, without generating each scene and element separately.

Alibaba said it can generate newspaper pages, storyboards and complex infographic grids at once. It also said demo images it released were produced in a single generation process, not stitched together from multiple outputs.

The model targets practical production rather than flashy images. In an official announcement, the Qwen development team said Qwen Image 3.0 aims to be a deployable productivity tool rather than merely producing "good-looking results."

It also improved text rendering. Alibaba said Qwen Image 3.0 can precisely render small letters around 10 pixels and finely reproduce details such as skin texture and hair. It also supports LaTeX notation for expressing complex mathematical formulas, researchers said. Alibaba said this allows it to generate in one go images in the form of academic documents or educational materials that include formulas, explanatory text and charts.

Multilingual rendering and interface implementation functions have also been expanded. Alibaba said the model can natively render 12 languages and can simulate key digital interface formats such as web pages, games and livestreams. It also offers a function that pulls the latest information from the internet and reflects it in images. If a user requests a weather visualisation by entering a specific city and date, it can produce results based on the latest real data rather than estimates.

Alibaba presented design studios, content production teams, e-commerce operations organisations and the education sector as key use cases for the model. The move is seen as a strategy aimed at companies and organisations that repeatedly need to produce large volumes of visual materials and document-style images.

However, the announcement did not include material that can verify actual performance. Qwen Image 3.0 was released without open model weights, and no performance comparison table or technical report was provided. The model can currently be tried via Qwen's official chat service, but application programming interface (API) pricing has not yet been disclosed.

That differs from how earlier models were released. Qwen Image 1.0 released open weights and a technical report under an Apache 2.0 licence on launch day, but the new model introduced performance mainly through demo images selected by the company.

Comparison with existing rival models also needs confirmation. In Alibaba's own evaluation, Qwen Image Bench, the previous flagship model Qwen Image 2.0 Pro ranked fifth among 18 models. OpenAI's GPT Image 2 ranked first at the time.

It is difficult to quantify how much the new model has improved over the previous generation based only on the materials currently available. Future user evaluations, API pricing and whether technical reports are released are expected to be key criteria for judging competitiveness.

Whether Qwen Image 3.0 succeeds will depend on how reliably it delivers results in complex document-style images and workplace production environments. As Alibaba attempts to differentiate itself in the image-generation AI market by emphasising real-world usability, attention is focused on additional performance data and actual usage results to be released later.