DesignArena raises $7.9 million for human AI feedback

DesignArena’s parent company, Intelligence, has raised $7.9 million in seed funding to turn human preferences into evaluation data for AI-generated media. The round was led by Index Ventures, with participation from Conviction — including Sarah Guo and Mike Vernal — A*, Valkyrie, and others.

The company was founded by co-founder Grace Li and several college friends a few weeks before their graduation in 2025. Their original project was an AI game engine that could produce functional games, but not necessarily enjoyable ones. That exposed a problem automated systems could not solve: determining whether a game was actually fun.

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

How DesignArena collects human feedback

For individual users, DesignArena resembles a model-routing service. A Chat-GPT-style interface accepts prompts, while dropdowns let users choose websites, images, and about a dozen other visual formats. The platform then presents outputs in repeated “A vs. B” comparisons, asking users to rank a group of results from best to worst.

The rankings are largely model-agnostic. Users typically care about receiving the strongest output, not which system produced it. That makes their choices useful to AI companies training and refining media-generation models, according to Li.

Users must log in to retrieve their results. Intelligence can therefore track how preferences vary across regions and change over time. Li said, for example, that web dashboards in Asia tend to favor a more maximalist design style.

Recommended reading

Gemini Spark can now book through Chrome

$60 million in ARR and an uncertain market

The consumer-facing tool helps generate the feedback, but the company’s main business is enterprise access to that evaluation data. Li said DesignArena is currently generating $60 million in annual recurring revenue, although the company did not provide additional financial details or explain how that figure is calculated.

Human rankings also give AI companies a complement to automated benchmarks. Automated tests can evaluate systems at much greater scale, but the source says they can be gamed or manipulated, a weakness highlighted by the Hugging Face breach last week.

The model has not worked for every startup. Yupp shut down earlier this year, less than a year after launching, despite raising $33 million from a16z crypto’s Chris Dixon, signing frontier-model customers, and claiming more than 1.3 million users.

Another comparable company has had a different trajectory. LM Arena, which evaluates text responses using a similar human-preference approach, raised $150 million in Series A funding in January, four months after formally launching its paid product. The source does not state when Intelligence plans to use its new funding or whether DesignArena’s enterprise revenue is profitable.