Qwen image model takes first place among open models on Arena
Arena scored Qwen-Image-2.1 at 1,228 points, 17th overall and 4 points behind Google's Gemini 3 Pro image preview on the text-to-image board.
Alibaba's Qwen-Image-2.1 took first place among open models on Arena's text-to-image leaderboard with 1,228 points, Arena said on Tuesday. The model placed 17th overall. Arena runs the leaderboard and posted the ranking to its own site, where the full table is published.
The gap to the closed models above it is small. Qwen-Image-2.1 sits 4 points behind Gemini-3-pro-image-preview at 16th and 11 points behind GPT-Image-1.5-high-fidelity at 15th, according to the figures Arena posted. Arena gave no margin of error for any of the three.
Arena scores are crowd preference votes rather than a fixed test set, so the ordering can move as votes accumulate. A 4-point gap on that scale may not survive more comparisons, and Arena has not said how many votes sit behind this ranking.
Alibaba released Qwen-Image-2.1 with open weights earlier this month, and the model generates and edits images in one pass. Arena keeps a separate leaderboard for image editing, which it linked alongside the text-to-image one. It did not give the model's placing on that second board.