Qwen-Image-2.1-Turbo
The 8-step Qwen image checkpoint — same 7B model, five times fewer forward passes, open weights.
About
Qwen-Image-2.1-Turbo, released October 9, 2026 by Alibaba's Qwen team, is an accelerated checkpoint of the open-weight Qwen-Image-2.1 (base model: September 20, 2026). Same 7-billion-parameter visual-generation architecture, but the denoising schedule is baked into the weights: 8 steps instead of the base model's 40-step default, at CFG=1 so each step is a single forward pass — roughly five times fewer passes per picture on the same hardware.
It handles text-to-image and image editing in one checkpoint, with up to 10 reference images, local edits via circles/painted annotations/masks, identity preservation for people and products, and native transparent (RGBA) layer generation. ComfyUI added Turbo within hours in three forms (full BF16, INT8 convrot quantization, LoRA extraction). Prefix KV caching reuses text and reference-image context across denoising steps.
Weights are on Hugging Face and ModelScope under the Qwen Research License — research-only, with commercial use requiring a separate license. Turbo and Pro APIs went live the same day on Alibaba Cloud Model Studio from 0.1 yuan per image. An independent test found ~5x speedups on Mac hardware but Turbo failing exact-count and layout tests the base model passed; Alibaba has published no Turbo speed or quality figures of its own.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Independent small test (Oct 11, 2026): ~5x faster than base on Mac hardware; failed all 12 exact-count/arrangement tests. Provider: no published Turbo speed/quality figures; base Qwen-Image-Bench 60.28 (provider-run).
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 8/10
Release facts (October 9, 2026; same 7B architecture; 8 baked-in steps vs 40; Qwen Research License; ComfyUI support; 0.1 yuan/image API) confirmed by Qwen's model repository and corroborated by independent outlets. Docked for: (1) the Qwen Research License is not a general commercial license; (2) Alibaba published no Turbo speed/quality figures, and independent testing found real quality tradeoffs on counts and layouts; (3) the fixed 8-step schedule removes the fidelity dial power users had.