Qwen-Image-2.1-Turbo: "7B" Label Hides a 16.2B, 32.4 GB Pipeline

Qwen released the accelerated Qwen-Image-2.1-Turbo checkpoint on October 9, 2026, generating and editing images in 8 denoising steps instead of the base model's 40. Despite the "7B" label, the full pipeline holds 16.2B parameters and 32.4 GB of weights, with over half in the Qwen3-VL 8B text encoder.

• DiT transformer 7.115B (14.23 GB), text encoder 8.767B (17.53 GB), VAE 0.338B
• Full BF16 needs 32.4 GB of weights; GGUF Q4_K_M with a 4-bit encoder fits in 11.19 GB
• 8 steps instead of 40, CFG=1, fixed sigma schedule baked into model_index.json
• License rules out paid use; Qwen publishes no Turbo speed or quality benchmarks