The lineup

10 image models, one prompt each.

Every model gets the identical prompt at the same aspect ratio, once, and whatever comes back first is what gets published. The percentages below are an automated check of whether the candy’s names were printed correctly — applied the same way to every model. They are not a ranking, and the site does not publish one.

ModelWeightsIn the lineup sinceJapanese nameEnglish nameRendered
GPT Image 2.5 SunburstOpenAIClosed100%13/13100%13/1313/13
Nano Banana ProGoogleClosed100%13/1385%11/1313/13
Nano Banana 2GoogleClosed100%13/13100%13/1313/13
Seedream 5.0 ProByteDanceClosed100%13/13100%13/1313/13
Ideogram V4IdeogramClosed77%10/1369%9/1313/13
Grok Imagine 2.0xAIClosed92%12/13100%13/1313/13
Qwen Image 3AlibabaOpen85%11/13100%13/1313/13
FLUX 2 ProBlack Forest LabsClosed69%9/1392%12/1313/13
Z-Image TurboAlibabaOpen46%6/1369%9/1313/13
MAI Image 2.5 ProMicrosoft · pausedClosed100%2/20%0/22/2
HunyuanImage 3TencentOpen54%7/13100%13/1313/13

Sample sizes are small and shown next to every percentage on purpose. A model that has rendered a handful of candies has not been measured in any serious sense. How the check works.

Browse the candies