Qwen Image 3

Qwen Image 3.0 is the latest image model from the Qwen team at Alibaba. It launched on 21 July 2026, first to invited developers and then to everyone on 5 August. Alibaba's goal was an image model that works as a practical tool rather than one that just makes pretty pictures. It can take very long, detailed prompts and still get the small things right, including text as small as ten pixels, mathematical formulas and writing in twelve languages, all in a single pass. Alibaba pitches it for jobs like newspaper layouts, storyboards and exam papers.

Who makes it

Qwen is Alibaba's family of AI models. It began in 2023 as a set of large language models, the kind that power chatbots, and quickly became one of the most widely used open model families in the world, with developers everywhere building on it. Image generation was added in 2025.

History

The first Qwen-Image came out on 4 August 2025. It was a large 20-billion-parameter model released under the Apache 2.0 licence, which meant anyone could download it, run it and build on it for free. It was especially good at text, Chinese included, and an editing version followed two weeks later. Hobbyists took to it quickly and produced a wave of fine-tuned versions.

Qwen-Image 2.0 arrived in February 2026. It was much smaller than the original but scored higher, and could make 2048-pixel images natively. This time Alibaba published a technical report but not the model itself. Qwen Image 3.0 goes further in the same direction: it shipped without downloadable weights, benchmark scores or a technical report, and is available only through Alibaba's cloud services. That is a clear step away from the openness that made the first version so popular.

Using it

Developers outside China reach Qwen Image 3.0 through Qwen Cloud, which offers Standard and Pro tiers priced per image.