CL-SN55 941 —
GPT-61S 885 —
GPT-56T 861 —
GPT-6S 825 —
CL-OP55X 818 -0.1%
MUSE-SPK 793 —
QWEN-38X 779 —
GPT-6A 778 —
CL-OP5H 711 -0.1%
CL-OP5X 709 -0.1%
KIMI-K3X 699 -0.1%
CL-OP46H 694 -0.1%
CL-FAB5H 692 -0.1%
CL-OP55H 692 -0.1%
CL-OP47H 683 -0.1%
GEM-37FH 648 -0.2%
GEM-38FH 633 -0.2%
INKL 531 —
CL-OP47 529 -0.2%
GPT-56S 529 -5.9%
GEM-3P 493 —
CL-OP46 488 —
CL-OP48 481 —
GEM-31P 453 -0.2%
GPT-6L 436 —
CL-SN55 941 —
GPT-61S 885 —
GPT-56T 861 —
GPT-6S 825 —
CL-OP55X 818 -0.1%
MUSE-SPK 793 —
QWEN-38X 779 —
GPT-6A 778 —
CL-OP5H 711 -0.1%
CL-OP5X 709 -0.1%
KIMI-K3X 699 -0.1%
CL-OP46H 694 -0.1%
CL-FAB5H 692 -0.1%
CL-OP55H 692 -0.1%
CL-OP47H 683 -0.1%
GEM-37FH 648 -0.2%
GEM-38FH 633 -0.2%
INKL 531 —
CL-OP47 529 -0.2%
GPT-56S 529 -5.9%
GEM-3P 493 —
CL-OP46 488 —
CL-OP48 481 —
GEM-31P 453 -0.2%
GPT-6L 436 —
← Back to feed

Qwen Image 2.1: Alibaba Ships Open 7B Image Model That Runs on a 3090 and Claims to Beat Closed Rivals

Alibaba’s Qwen AI team released Qwen-Image-2.1 on September 20, an open-weight model targeting image generation and editing. The visual generation component carries 7 billion parameters, a count that Qwen says is enough to outscore most closed competitors on its internal benchmark. Independent evaluations have not yet run.

What it does

Qwen-Image-2.1 natively generates and edits transparent images in RGBA format, letting users isolate subjects or manipulate text on transparent layers without a separate compositing step. The model accepts up to ten reference images simultaneously, covering use cases like group portraits assembled from individual photos, virtual try-ons, and room design with multiple source materials. Local edits are guided by circles, masks, or painted strokes rather than text descriptions alone.

On the inference side, Qwen says architecture changes and KV cache reuse cut latency when multiple reference images are involved. The team designed the model for consumer GPU compatibility via frameworks including LightX2V, with third-party reports noting it running on RTX 3090 hardware.

Availability and licensing

Weights are published on Hugging Face, GitHub, and ModelScope, with a live demo on Hugging Face Spaces. The release ships under a research license that bars commercial use. Business users must apply to Qwen for a separate commercial licence.

What to watch

Qwen-Image-2.1’s self-reported benchmark scores were the main evidence at launch. Alibaba published a Qwen-Image-Bench evaluation chart naming specific targets including GPT Image 1.5, GPT Image 2.5 Sunburst, and Nano Banana 2.0. Independent third-party evaluations have not yet run. Until results appear on standard external benchmarks, the relative rankings should be read as vendor-reported figures.

The 7-billion-parameter visual component is notably small. If the benchmark holds under independent scrutiny, the efficiency story becomes significant: smaller models that match or exceed closed systems at consumer GPU memory budgets reshape the economics of local image generation.