Grace Li’s company began a few weeks before her graduation in 2025. She and a small group of university friends were trying to improve an AI game engine. The software could generate functional games, but none were enjoyable. This raised a specific question: how do you know if a game will be fun?
Human judgment was the only answer. The team started brainstorming ways to gather honest feedback from people at scale.
The result was DesignArena. The tool now has 5.3 million users worldwide. Many AI firms needed scalable feedback, and they were prepared to pay for it.
“It was the missing bottleneck for a lot of these models to make improvements in the design space,” Li says. “About a week later, we closed our first major deal with a frontier lab, and the rest is kind of history.”
Intelligence, the company behind DesignArena, announced a $7.9 million seed round on Monday. Index Ventures led the investment. Conviction, A*, Valkyrie, and others also participated.
Regular users treat DesignArena like a sophisticated model router. There is a Chat-GPT-style window for prompts, with separate dropdowns for websites, images, and a dozen other visual formats. After entering a request, format, and style, the system presents a series of “A vs. B” choices. Users rank the handful of outputs from best to worst.
The real value lies in the enterprise side. Participating models use the platform as a source of endless instant feedback for their media-generating systems. Users tend to be indifferent to which models they are ranking. As Li puts it, they just want the best output they can get. Their rankings provide critical input into what users actually want.
For frontier labs, that is a service worth paying for. The site is currently generating $60 million in annual recurring revenue. It has solidified its position as a key source of human-led evaluation data for the AI industry.
Users must log in to receive their output, so Intelligence can track how tastes change across different continents and over time. Dashboards in Asia, for example, tend to feature a more maximalist design style. These measures are an important complement to automated benchmarks. Those benchmarks can operate at a greater scale but are often subject to being gamed or otherwise manipulated, as the Hugging Face breach demonstrated last week.
That is not to say that crowdsourced human feedback will be an automatic winning market. Less than a year after launching, Yupp closed its doors earlier this year. It raised $33M from a16z crypto’s Chris Dixon. The company did secure some frontier models as customers and claimed over 1.3 million users, but it could not build a sustainable long-term business.
Other startups based on human evaluation seem to be thriving. LM Arena, which takes a similar approach to text-based responses, raised $150 million in a Series A in January. This happened just four months after formally launching its paid product.
What it means
DesignArena proves that human preference data is valuable enough to sustain a business. While some attempts fail, the demand for real-world taste checks is driving significant investment in the sector.
