The image generation track in the AI field has once again enjoyed a highlight moment. On November 10, SuperCLUE officially released its November 2026 updated text-to-image benchmark ranking. This evaluation not only covered mainstream products from home and abroad, but also for the first time introduced China's latest model ArtArk, as well as the GPT series, Nano Banana series, and others for supplementary testing. In the end, the domestic model ArtArk delivered an outstanding performance, taking first place on the overall ranking with a composite score of 95.14, successfully surpassing GPT Image2.5Sunburst.
Core metric stuns the industry: text-image consistency becomes the only perfect score in the entire field
In this highly competitive evaluation, ArtArk's championship did not rely on a slight edge in a single category; instead, it demonstrated exceptionally strong overall comprehension and reconstruction capabilities. In particular, in the "text-image consistency" test item that the industry closely watches, ArtArk achieved the only perfect score among all tested models.
Text-image consistency essentially tests how well AI truly understands users' written instructions and how accurately it turns complex textual descriptions into images. Whether it involves a specific main character, props or objects, or complex spatial scenes and their relationships, the model must not only ensure the image looks good but also restore every detail strictly according to the instructions. ArtArk's perfect score marks that it has reached a top level in complex semantic understanding and image alignment.
Competition among leading tier heats up: gap between domestic and overseas products is almost negligible
Looking at the overall ranking distribution, the technical gap between the world's top text-to-image models has narrowed to an extreme degree.
Among overseas models, GPT Image2.5Sunburst ranked second overall with 94.27 points, the highest score among overseas products tested; the same-series GPT Image2.5Flare followed closely with 93.45 points. On the domestic side, SenseNova U1Pro scored a high 93.81 points to rank second in China, while Doubao-Seedream-5.0-pro and Qwen-Image-3.0-Pro tied for third domestically. Overall, the highest and lowest scores among the top five differ by only 1.91 points, showing that leading products at home and abroad are locked in an extremely close competition.
Each model has its own strengths: functional segmentation and technology iteration continue to accelerate
It is worth noting that although their overall scores are close, mainstream models show clear differentiated advantages in specific capabilities.
Details and Chinese character performance: GPT Image2.5Sunburst won the highest scores across the board in the two categories of Chinese character generation and realistic reproduction; Flare performed excellently in overall image quality and text-image consistency (93.94 points).
Creation and layout advantages: GPT-Image-2 performed strongly in creativity and reasoning, scoring 96.81 points, and its complex information layout score reached 94.04 points; meanwhile, China's SenseNova U1Pro stood out most in complex information layout, ranking first among all tested models with 95.80 points.
Rapid series iteration: The Nano Banana series also made significant progress, with Nano Banana2.1 becoming the highest-scoring version in the series (92.88 points); its text-image consistency and complex information layout capabilities improved by a substantial 9.09 points and 6.87 points, respectively, compared with the previous generation.
These test results show that the evaluation criteria for current text-to-image models have long gone beyond simply "looking beautiful." Whether a model can accurately understand human instructions, handle complex image layouts, and precisely generate specific text has become a core hard metric for measuring AI image generation strength. ArtArk's successful rise to the top also proves to the industry that domestic text-to-image models now have the strong capability to compete head-on with leading overseas products.
