Alibaba's Qwen team has released Qwen-Image-2.1-Turbo, an accelerated checkpoint of its open-weight Qwen-Image-2.1 model. The new version generates and edits images in eight denoising steps, compared with the base model's default of 40 steps.
For developers, this means roughly five times fewer denoising steps, which can translate into faster inference and lower compute costs. The model retains the 7B parameter size of the original, according to the release.
The checkpoint is open-weight, continuing the Qwen team's pattern of releasing models that developers can run themselves. The announcement does not detail exact wall-clock speedups or quality trade-offs, but the step reduction is the headline change.