DesignArena Creators Secure $7.9M to Give AI Models a Sense of Taste

Grace Li, co-founder, recounts the genesis of her company several weeks prior to her 2025 graduation. She and a group of college friends were developing an AI game engine. While their models could produce functional games, they lacked an essential element: enjoyment. This led them to a fundamental question: how does one objectively assess the "fun" factor of a game?

Concluding that human judgment was irreplaceable, the team promptly began conceptualizing methods to acquire authentic human feedback on a large scale.

This endeavor culminated in DesignArena, an AI-powered tool currently utilized by 5.3 million individuals globally. It quickly became apparent that numerous AI companies were actively seeking scalable user feedback solutions, and a significant portion of them were prepared to invest in such services.

Li commented, "It was the missing bottleneck for a lot of these models to make improvements in the design space." She further elaborated, "About a week later, we closed our first major deal with a frontier lab, and the rest is kind of history."

This past Monday, the company responsible for DesignArena, named Intelligence, revealed the successful closure of a $7.9 million seed funding round. The round was spearheaded by Index Ventures, with additional contributions from Conviction (including Sarah Guo and Mike Vernal), A*, Valkyrie, and other investors.

For individual, non-enterprise users, DesignArena functions akin to a sophisticated model router. It features a prompt interface reminiscent of Chat-GPT, complemented by distinct dropdown menus for specifying formats such as websites, images, and numerous other visual types. After inputting a request, including desired format and style, users are guided through a series of "A vs. B" comparisons until they have ranked a selection of outputs from most to least preferred.

While providing a valuable service, the platform's primary worth resides on the enterprise side. Here, participating models leverage DesignArena as an inexhaustible source of immediate feedback for their media-generation capabilities. Users, as Li explains, are typically unconcerned with the specific models they are evaluating, focusing solely on obtaining the optimal output. This impartiality ensures their rankings deliver crucial insights into genuine user preferences.

Li states that for frontier laboratories, this constitutes a service of significant value, noting that the site currently boasts an Annual Recurring Revenue (ARR) of $60 million. This financial performance firmly establishes DesignArena as a vital provider of human-led evaluation data for the artificial intelligence sector.

A critical aspect is the mandatory user login to access outputs, which enables Intelligence to monitor evolving aesthetic preferences across various continents and throughout time. (Li specifically observes a tendency towards a more maximalist design style in Asian web dashboards.) These qualitative metrics serve as an important complement to automated benchmarks. While automated systems offer greater scalability, they are frequently susceptible to manipulation or "gaming," a vulnerability dramatically highlighted by the Hugging Face breach last week.

However, the success of crowdsourced human feedback as a market is not guaranteed. Yupp, for instance, ceased operations earlier this year, less than a year after its launch, despite securing $33 million in funding from a16z crypto’s Chris Dixon. The company had attracted frontier models as clients and claimed over 1.3 million users, yet ultimately failed to establish a sustainable long-term business model.

Despite these challenges, other startups focused on human evaluation are demonstrating significant success. LM Arena, which employs a comparable methodology for evaluating text-based responses, successfully raised $150 million in a Series A funding round in January, a mere four months after the official launch of its paid product.