Qwen-Image-3.0, released July 21, 2026 by Alibaba's Qwen team, targets production work: newspapers, storyboards, exam papers, and academic figures. The model's signature advance is 4.5k-token prompts enabling one-shot complex layouts — demonstrated by a nine-panel infographic grid with physics derivations, medical diagrams, and cell-biology comparisons, each panel retaining readable small text.

Text rendering reaches 10px legibility, handling LaTeX-grade equations, dense newspaper copy, and tiny annotation labels without collapse. Detail extends to pore-level skin texture and faithful restoration of damaged ink paintings. The model renders 12 languages natively, supports 100+ art styles, and produces simulated UIs for web and games. It can build reference infographics from photos and pull live web data, such as a weather-forecast graphic for a specific date.

Compared to Qwen-Image-2.0, Nano Banana Pro, Seedream 5 Pro, and GPT-Image-2, Qwen-Image-3.0 leads on small-type legibility, multi-panel structure, and script fidelity. Access is via Qwen Studio (chat.qwen.ai) and Qwen Cloud API; third-party endpoints may be unstable at launch. Pricing and open-weight availability remain undisclosed.