Alibaba Qwen-Image-2.1-Turbo and missing evaluations

Alibaba released Qwen-Image-2.1-Turbo on October 9, 2026, an accelerated open-weights image generation model that produces 2K images in 8 denoising steps. The Qwen team stated the model "still generates strong 2K images from text, and supports continued creation through natural-language edits, from adding accessories to changing a scene." Alibaba describes the base architecture as one that "outperforms most closed-source models" with "drastically accelerated inference for multi-image inputs." Open weights and hosted APIs for both the Pro and Turbo variants launched simultaneously.

Orcarouter.ai analyzed both model cards and found the Turbo card contains no evaluation numbers, no Qwen-Image-Bench result, and no ablation data. The two checkpoints share identical resolution presets, from 1:1 at 2048x2048 up to 16:9 at 2752x1536, and load through the same QwenImage21Pipeline in Diffusers. As orcarouter.ai put it, "The Turbo card's quick start is the base card's quick start with the checkpoint name changed and bfloat16 spelled dtype instead of torch_dtype." Both models carry identical license metadata under the Qwen-Image-2.1 Research License Agreement.