Intelligence raises $7.9M for Design Arena.
Intelligence has raised $7.9 million for Design Arena, a platform generating $60 million in ARR by crowdsourcing human aesthetic feedback to help frontier AI labs refine their visual models.

Intelligence, the startup behind the crowdsourced evaluation platform Design Arena, has secured $7.9 million in seed funding. The round was led by Index Ventures, with participation from Conviction (led by Sarah Guo and Mike Vernal), A*, Valkyrie, and other investors. Founded by Grace Li and her college friends just before graduating in 2025, the company originally sought a way to evaluate whether AI-generated games were actually fun. They realized that human judgment was irreplaceable, leading them to build Design Arena. Today, the platform has attracted 5.3 million users who rank visual outputs, such as images and websites, in an A-versus-B format.
This crowdsourced ranking system has become a goldmine for frontier AI labs seeking high-quality human feedback to train their media-generation models. Intelligence has successfully monetized this demand, already generating $60 million in annual recurring revenue (ARR). By requiring users to log in, the company can track how design preferences shift across different regions and over time. This structured preference data offers a vital alternative to automated benchmarks, which are increasingly vulnerable to manipulation and gaming.
For AI practitioners and developers, this development highlights a growing shift toward human-in-the-loop evaluation as a critical component of model alignment. While some competitors have struggled—such as Yupp, which raised $33 million from a16z crypto's Chris Dixon and gathered 1.3 million users before shutting down—others are seeing massive investor interest. For instance, the text-focused LM Arena raised $150 million in a Series A round this past January. For developers building generative media, platforms like Design Arena provide the nuanced, real-world aesthetic feedback that automated metrics simply cannot replicate, helping them fine-tune models to match actual human taste.
This is our own summary of reporting by TechCrunch AI



