Alibaba / Qwen · Released 2026-09-20
Generation, transparent assets and multi-reference editing. Image-preference evaluation is distinct from the text, coding and physics boards. Weights are limited to non-commercial research/evaluation; commercial use needs separate permission.
Original model explanation ↗No comparable leaderboard score is included yet. Explore the evaluations and examples below.
Independent image evaluation and version distinction
Artificial Analysis read October 10 lists base Qwen-Image-2.1 at Elo 1035 for text-to-image v2.0, with a 95% interval of 1026–1044 and 5,489 samples.
Image-preference evaluation, not a general index
Official example explained
Released September 20, Qwen-Image-2.1 accepts up to ten reference images.
Official work and method
The model unifies text-to-image generation and editing, with native transparent-image support.
The 0902 snapshot updates coding, tool orchestration and vision while retaining explicit version identity.
Original release ↗The announced default Max endpoint transition points to 0902; this is a routing change, not another new model.
Original release ↗Generation, transparent assets and multi-reference edits share one visual model; check the terms for available weights.
Original release ↗The Singapore International lifecycle table lists hosted Pro access; the repository's October 9 link announcement is not its initial release.
Original release ↗An eight-step checkpoint and hosted Turbo access continue multi-reference editing; denoising steps are not measured end-to-end latency.
Original release ↗