Qwen-Image-3.0 renders readable 10px text and LaTeX in one pass
Alibaba's new image model renders infographics, formulas and 12 languages in one pass — but ships closed and invite-only, with no open weights.
Copy markdown
Legible 10px text, LaTeX and infographic grids in one pass
Qwen-Image-3.0 draws full 3×3 infographic grids, multi-line LaTeX with sub- and superscripts, and readable ten-pixel text in a single generation — from prompts up to 4,500 tokens. It is the first image model you would trust to typeset a dense diagram or slide, not just make it look pretty.
The catch: closed, no weights this time
Where Qwen-Image 1.0 shipped Apache-2.0 weights and a technical report, 3.0 arrives with no weights, no benchmark table, no parameter count, and no paper. If your pipeline relies on self-hosting Qwen's image model, this generation ends that.
How to try it right now
Access is invite-only API for now, rolling into Qwen Chat where you can test prompts in the browser. No public pricing has been posted yet — plan for an API tier, not a free download.
What you could ship this weekend
One-shot marketing visuals, cheat-sheets, UI mockups and multi-window app screenshots — with live internet data baked in, like a weather panel rendered inside the image — across 12 languages including Japanese, Korean and Spanish, no separate typesetting step.