Qwen-Image models
10 Qwen-Image models from Qwen on Hugging Face, from 338M to 27.4B parameters, published by Qwen in BF16, with a 256K-token context. The smallest official model, Qwen-Image-2.1, needs about 15.9 GB of VRAM at its published precision; the cheapest live fit is RTX 3090 at $0.162/hr.
Part of the Qwen series
Pick a size
One row per official Qwen-Image size: the VRAM it needs at each precision, the cheapest GPU that holds it today, and the KV cache for a 32K-token context.
| Model | Parameters | Native VRAM | FP8 VRAM | INT4 VRAM | Live GPU fit (native) | Est. $/hr | KV cache at 32K |
|---|---|---|---|---|---|---|---|
| Qwen-Image-2.1 | 7.1B | 15.9 GB | 8.0 GB | 4.0 GB | RTX 3090 | $0.162/hr | not published for this architecture |
| Qwen-Image | 20.4B | 45.7 GB | 22.8 GB | 11.4 GB | RTX A6000 | $0.550/hr | not published for this architecture |
| Qwen-Image-Bench | 27.4B | 61.1 GB | 30.6 GB | 15.3 GB | RTX PRO 6000 | $1.38/hr | 2.0 GB |
VRAM is weights times a flat 1.2 overhead; the KV cache is a separate per-model figure at 16-bit, shown where the architecture is published. See the methodology.
Official models (6)
| Model | Parameters | Published as | VRAM needed | Live GPU fit | Est. $/hr |
|---|---|---|---|---|---|
| Qwen-Image-2.1 | 7.1B | BF16 | 15.9 GB | RTX 3090 | $0.162/hr |
| Qwen-Image-2.1-Turbo | 7.1B | BF16 | 15.9 GB | RTX 3090 | $0.162/hr |
| Qwen-Image | 20.4B | BF16 | 45.7 GB | RTX A6000 | $0.550/hr |
| Qwen-Image-2512 | 20.4B | BF16 | 45.7 GB | RTX A6000 | $0.550/hr |
| Qwen-Image-Edit-2511 | 20.4B | BF16 | 45.7 GB | RTX A6000 | $0.550/hr |
| Qwen-Image-Bench | 27.4B | BF16 | 61.1 GB | RTX PRO 6000 | $1.38/hr |
Fine-tunes and community models (4)
| Model | Parameters | Published as | VRAM needed | Live GPU fit | Est. $/hr |
|---|---|---|---|---|---|
| texture-fix-vae-for-qwen-image-2.1 | 338M | F32 | 1.5 GB | P4 | $0.058/hr |
| Qwen-Image-2.1-Lightning-v2-RL | 7.1B | F32 | 31.8 GB | V100 | $0.219/hr |
| Qwen-Image-2.1-TailSteer | 7.1B | BF16 | 15.9 GB | RTX 3090 | $0.162/hr |
| Qwen-Image-2.1-viggle-turbo | 7.1B | BF16 | 15.9 GB | RTX 3090 | $0.162/hr |
VRAM is for the precision the model is published in: the weights times a flat 1.2 overhead, with the KV cache not included. See the methodology. The fit shown is the lowest-priced single GPU type that holds the model at that precision, or the lowest-priced multi-GPU set (up to 8) when none does.
