'Design Arena' judges anonymous AI design competitions and creates 'AI rankings with high design ability'



AI performance is measured from a variety of perspectives, such as how accurately it can solve difficult mathematical problems and whether it can converse so naturally that it is indistinguishable from a human. ' Design Arena ' is a test that focuses on AI design capabilities, measuring the design capabilities of AI through repeated tests that 'determine the superiority or inferiority of anonymous AI.' Anyone can participate in the test without registering an account, so we actually took part in the Design Arena test and checked the rankings of each AI.

Design Arena

https://www.designarena.ai/

When you access Design Arena, an input area will appear at the top of the screen. Enter the design you want to request from the AI in this input area and click the send button. Four randomly selected AIs will then generate a design for you. In this example, I entered 'Create a website for a bakery. The store name is 'GIGA Bread'.'



If you are asked to agree to the terms and conditions, read them carefully and then click 'I Understand and Agree.'



First, the generation process is carried out using two of the four randomly selected AIs.



When the generated results are displayed, click 'Vote as Winner' for the one you like better. The identity of the AI is not revealed at this point. In this case, the AI on the right was not able to generate a design in the first place, so the one on the left was declared the winner.



The remaining two AI types will then be generated, so please wait a while.



Once the generated results are displayed, you can choose a winner. Since design preferences vary from person to person, you can get results that reflect reality, not just a numerical evaluation, but 'how easily you can create a design that people will like.'



You will be asked to change the combination and choose whether to win or lose, so select whether to win or lose again.



More selection.



The victory/loss selection screen will be displayed a total of five times.



After selecting the winner for each of the five matches, the results screen will be displayed. This time, 'GPT-5 (Minimal)' came in first place.



If you scroll down, you'll see a list of which AI created which design. Further down, you'll see the results of questions submitted by other users.



At the bottom of

the Design Arena homepage , a ranking based on past wins and losses is displayed. The top spot went to 'Claude Opus 4.1 (Thinking),' with a win rate of 73.7%.



The rankings were also calculated by category. In terms of website design ability, 'GPT-5 (Minimal)' came in first with a win rate of 73.6%.



in AI,   Software,   Review,   Web Application, Posted by log1o_hf