Skip to content

Qwen-Image 2.1

Alibaba's open-weight image generation and editing model, with a 7B-parameter image generator, released in September 2026 under the Qwen Research License. It can output transparent images and accepts up to 10 reference images.

Alibaba (Qwen)Image modelReleased September 2026Data checked October 18, 2026

This model is not in the current people's rankings: it is either new and still collecting votes, or ranked below the top 12. See the AI leaderboard for the models people rate highest.

Strengths and weaknesses

  • Open weights on Hugging Face, with community ComfyUI and quantized builds
  • Highest-scoring open-weight entry on the LMArena text-to-image board (19th of 82 entries)
  • Built-in transparency (RGBA) output and multi-reference editing
  • The Qwen Research License is not a standard open-source license, so check it before commercial use
  • Far behind the closed leaders on both image arenas

Best for

  • Running image generation on your own hardware
  • Transparent PNG assets
  • Experimenting with an open image model

Every score we have

Each bar compares this model with the best result any model on this page has on the same test.

  • LMArena text to image1,222.9Arena score 1222.9 (95% CI 1216.3 to 1229.6); 11,441 votes; entry "qwen-image-2.1"; position 19 of 82 on the full board
  • LMArena image editing1,368.7Arena score 1368.7 (95% CI 1362.8 to 1374.6); 21,941 votes; entry "qwen-image-2.1"; position 18 of 59 on the full board
  • Artificial Analysis image arena1,041.7Elo 1041.7 (95% CI 1029.7 to 1053.7); 5,468 appearances; position 21 of 104 on the Artificial Analysis text-to-image arena

Compare with