Intelligence, the startup behind the AI evaluation platform Design Arena, has raised a $7.9 million seed round led by Index Ventures, with participation from Conviction, A*, Valkyrie and other investors.

Design Arena works by having users submit prompts and then choose between competing AI-generated outputs, ranking them from best to worst. That crowdsourced preference data is then sold to AI companies looking to refine how their models handle subjective, taste-driven tasks like visual design.

The company has grown quickly since launching. Design Arena now serves 5.3 million users and generates roughly $60 million in annual recurring revenue, according to figures disclosed alongside the funding announcement in early August.

Conviction’s participation in the round came through its general partners Sarah Guo and Mike Vernal, adding two well-known AI-focused investors to the startup’s cap table alongside Index Ventures, which led the deal.

The new capital will fund an expansion beyond Design Arena’s original focus on visual design evaluation into adjacent categories, including user interface and user experience evaluation, content layout preferences and subjective writing style assessment. The push reflects a broader bet that AI companies will keep needing structured human feedback to fine-tune models on tasks where there is no single objectively correct answer.

Design Arena’s growth points to a wider trend in AI development, where model quality increasingly hinges on human preference data rather than raw benchmark performance alone. Startups built around gathering and structuring that feedback have drawn growing investor interest over the past year, as foundation model companies compete on the subjective quality of their outputs as much as their technical capabilities.

The company has not disclosed a valuation tied to the new round or detailed hiring plans following the raise.

The raise adds Design Arena to a growing list of startups positioning themselves as infrastructure for AI taste rather than as model builders themselves, a category investors have increasingly backed as differentiation between large language models narrows on raw capability and shifts toward output quality and style.