Design Arena Raises $7.9M to Bring Human Taste Into AI Model Evaluation

Intelligence, the company behind Design Arena, announced a $7.9 million seed round led by Index Ventures. The round also included participation from Conviction, A*, Valkyrie, and others.

The platform is already being used by 5.3 million people worldwide, according to the article. Its pitch is straightforward: AI labs need scalable human judgment, especially when model quality depends on subjective ideas like taste, design, and fun.

For non-enterprise users, Design Arena works like a model-routing experience: users enter a prompt, select visual formats such as websites or images, and compare outputs through a series of “A vs. B” choices. Those rankings help sort outputs from best to worst.

The enterprise value comes from giving participating models a steady stream of human feedback on media-generating performance. Users may not care which model produced an output; they typically want the best result, making their choices useful signals for what people actually prefer.

Design Arena grew out of co-founder Grace Li’s early work with college friends on an AI game engine, where the models could produce functional games but not necessarily fun ones. That problem led to a broader question: how do you measure whether AI-generated work is actually appealing?

The company’s answer is that human judgment remains essential. Automated benchmarks can scale, but the article notes they can also be gamed or manipulated, making human-led evaluation an important complement for frontier labs.

Li says Design Arena is currently generating $60 million in ARR, positioning it as a significant provider of human-led evaluation data for AI companies. The platform also requires users to log in, which lets Intelligence track how preferences shift across regions and over time.

The market is not risk-free. TechCrunch notes that Yupp shut down after raising $33 million and claiming more than 1.3 million users, while LM Arena raised $150 million in a Series A after launching its paid product.