Qwen Image 2.1 Puts a 7B Generator Next to Nano Banana 2.0
Alibaba's Qwen Image 2.1 is a 7B open-weight image model that beats Nano Banana 2.0 on one internal score.
Direct answer
Qwen Image 2.1 is a 7-billion-parameter open-weight image model that Alibaba says scored 60.2 on its image benchmark, just above Google's Nano Banana 2.0 at 59.82.
Alibaba Cloud has released Qwen Image 2.1, a 7-billion-parameter image model with open weights. Tom's Hardware reported the launch on 23 September 2026 and framed the claim directly: a small model that Alibaba says can beat Google's Nano Banana 2.0.
The size is the product point. A 7B generator is light enough, the report says, to run on older consumer GPUs such as an RTX 3090. That is a different offer from the hosted image tools most people use, where the weights stay on the vendor's side and the interface is a subscription.
The scores do not all point the same way
On Alibaba's own Qwen Image Benchmark, Image 2.1 scored 60.2. The same write-up puts Google's Nano Banana 2.0 at 59.82, Muse Image at 62.34, and OpenAI's GPT Image 2.5 Sunburst at 67. The open model clears one closed competitor and trails two others.
A preliminary AI Arena check was less kind to the internal ranking. There, Qwen Image 2.1 scored 1228 against 1260 for Nano Banana 2. The gap is small, and it is enough to keep the "beats Nano Banana" line inside one benchmark rather than every public board.
If later tests hold the quality up, the interesting part is local control: image generation without a cloud account. Until those tests land, the honest summary is a competitive 7B model, not a new leaderboard top.
Source: Tom's Hardware, 23 September 2026.
FAQ
- How small is the model?
- Tom's Hardware describes it as a 7-billion-parameter model that can run on older consumer cards such as an RTX 3090.
- Did it lead every board?
- On the internal Qwen Image Benchmark it trailed GPT Image 2.5 Sunburst and Muse Image. On AI Arena it also sat just behind Nano Banana 2.