DesignArena creators raise $7.9 million to bring taste to AI models

According to co-founder Grace Li, the venture launched just weeks before her 2025 graduation, when a small group of college peers attempted to build an AI-powered game engine. While the models could generate functional games, none were genuinely enjoyable, prompting a critical question: how do you predict whether a game will be fun?

They concluded that human judgment is irreplaceable, leading them to brainstorm scalable methods for gathering honest user feedback.

This effort evolved into DesignArena, an AI tool now utilized by 5.3 million users globally. The founders discovered that numerous AI companies were seeking scalable user feedback and were willing to pay for access.

“It addressed a critical bottleneck for many models aiming to improve in the design space,” Li explains. “Within a week, we secured our first major partnership with a leading AI lab, and the rest is history.”

On Monday, Intelligence, the company behind DesignArena, announced a $7.9 million seed funding round led by Index Ventures, with additional investments from Conviction (Sarah Guo and Mike Vernal), A*, Valkyrie, and other investors.

For individual users, DesignArena functions similarly to a sophisticated model router. Users enter prompts into a ChatGPT-style interface, selecting from dropdowns for websites, images, and various other visual formats. After specifying the request, format, and style, users are presented with “A vs. B” comparisons to rank outputs from best to worst.

While useful for individuals, the platform’s true value lies in its enterprise segment, where AI models use it as a source of instant, continuous feedback for media generation. Users typically do not care which specific models they are ranking; they simply seek the highest quality output, making their rankings a vital indicator of user preferences.

Li states that this service is highly valuable to leading AI labs, noting that the platform currently generates $60 million in annual recurring revenue (ARR), establishing itself as a crucial source of human-led evaluation data for the AI sector.

Crucially, users must log in to access outputs, allowing Intelligence to track how preferences shift across different regions and over time. (Li observes that web dashboards in Asia often favor a more maximalist design aesthetic.) These insights complement automated benchmarks, which, despite their scale, are vulnerable to manipulation or gaming, as highlighted by the recent Hugging Face security incident.

However, crowdsourced human feedback is not guaranteed to be a winning market. Less than a year after its launch, Yupp shut down earlier this year despite raising $33 million from a16z crypto’s Chris Dixon. Although Yupp attracted several leading models as clients and claimed over 1.3 million users, it failed to establish a sustainable long-term business.

Nevertheless, other startups focusing on human evaluation are thriving. LM Arena, which applies a similar approach to text-based responses, raised $150 million in a Series A round in January, just four months after officially launching its paid product.