Alibaba's Qwen-Image-2.0 ranks near top in blind tests
Alibaba's Qwen team has released Qwen-Image-2.0, a single model for image generation and editing that placed third for text-to-image and second for editing in blind AI Arena tests.
Alibaba's Qwen team has released Qwen-Image-2.0, a foundational image model that handles both generation and editing. The details come from a PixVerse write-up that summarizes the Qwen team's own blog post.
The model pairs an 8B Qwen3-VL encoder with a 7B diffusion decoder, which the team describes as "7B-class" efficiency. It accepts prompts of up to 1,000 tokens and outputs at native 2K resolution. Qwen is pitching it at text-heavy work: slides, posters, comics, bilingual layouts and multi-column infographics with numerical data. Demonstrations also show several classical Chinese calligraphy styles, including Slender Gold Script (瘦金体) and Small Regular Script (小楷).
On the AI Arena blind-evaluation leaderboard as of February 9, 2026, Qwen-Image-2.0 placed:
- Text-to-image: third with an Elo of 1029, behind Gemini-3-Pro-Image-Preview (1050) and OpenAI's GPT Image 1.5 (1043).
- Image editing: second with 1034, behind Gemini-3-Pro-Image-Preview (1042) and ahead of ByteDance's Seedream 4.5 (1011) and Alibaba's earlier Qwen-Image-Edit-2511 (1002).
These rankings are nine months old, and the write-up does not say where the model stands today.
The write-up also leaves out several practical details. It does not say whether the weights are open, how the model is licensed or priced, or how developers can access it. The technical report it cites, arXiv 2508.02324, was not reviewed for this article.