Chinese model ArtArk achieves top spot in superclue text-to-image benchmark

The November 2026 SuperCLUE benchmark ranking for text-to-image generation revealed a significant milestone for domestic AI. The Chinese model ArtArk secured first place overall with a composite score of 95.14, surpassing GPT Image2.5Sunburst. Notably, ArtArk achieved the sole perfect score in the highly scrutinized text-image consistency metric, demonstrating superior complex semantic understanding. Competition at the leading edge is extremely tight. GPT Image2.5Sunburst ranked second among international models with 94.27 points, while SenseNova U1Pro placed second domestically with 93.81 points. The top five models showed a narrow score spread of only 1.91 points. While ArtArk led, GPT Image2.5Sunburst excelled in Chinese character generation and realistic reproduction, and SenseNova U1Pro led in complex information layout with 95.80 points. Furthermore, the Nano Banana series showed iterative improvement, with Nano Banana2.1 scoring 92.88 points and improving its text-image consistency and complex information layout scores by 9.09 and 6.87 points, respectively, over its predecessor.