Alibaba has released Qwen-Image-2.1-Turbo, an accelerated 8-step checkpoint of its open-weight Qwen-Image-2.1 image generation and editing model. The new model reduces denoising steps from the base model's forty down to just eight on the same 7B architecture, significantly speeding up processing times.
Built on a single-stream DiT design with 7B parameters, the checkpoint features a Qwen3-VL 8B text encoder and a 64-channel RGBA autoencoder supporting native transparency. It utilizes prefix KV caching to reuse text and reference image context across denoising steps, achieving 2K generation and editing capabilities. Developers can run the model using CUDA GPUs in BF16 via Diffusers, or access it through Alibaba Cloud Model Studio.
The hosted API version is priced at CNY 0.1 per image with a 120 requests-per-minute limit, making it 2.5 times cheaper and allowing six times the request rate of the Pro version. However, because the weights are distributed under a research license, commercial self-hosting requires separate permission. This release provides developers with a faster, cost-effective tool for high-resolution image creation and multi-reference editing.