DesignArena's parent company Intelligence has raised a $7.9 million seed round led by Index Ventures, with Conviction, A*, and Valkyrie joining. The crowdsourced ranking platform is already generating $60 million in ARR and counts 5.3 million users, numbers that put it well past the traction most seed-stage startups show at announcement.
The product is a marketplace for human taste. Users submit prompts for websites, images, or one of a dozen other visual formats, then rank the resulting outputs through a series of A/B comparisons. Frontier labs pay to plug their image and design models into that stream, treating it as a live source of preference data that automated benchmarks can't produce.
Co-founder Grace Li started the company a few weeks before graduation in 2025 with college friends building an AI game engine. The models could produce functional games, but none were fun — and there was no automated way to tell whether a given output would land with a human. That question turned into DesignArena.
“It was the missing bottleneck for a lot of these models to make improvements in the design space.”— Grace Li, DesignArena co-founder
Key facts
- 01Intelligence, the company behind DesignArena, raised a $7.9M seed round led by Index Ventures with Conviction, A*, and Valkyrie participating.
- 02DesignArena is generating $60M in ARR and has 5.3 million users worldwide.
- 03Yupp, a similar human-feedback startup backed by a16z crypto's Chris Dixon, shut down earlier this year after raising $33M and reaching 1.3M users.
- 04LM Arena, which applies the same model to text responses, raised a $150M Series A in January, four months after launching its paid product.
Conviction partners Sarah Guo and Mike Vernal joined the round. Index Ventures led it. The syndicate reads as a bet that human preference data is a durable input to model training, not a temporary bridge until benchmarks improve.
Enterprise revenue is where the business works. Non-enterprise users get a ChatGPT-style prompt window and a model router; the labs on the other side get ranked outputs across geographies and formats. Because users log in, Intelligence can track how preferences shift across regions and over time.
“About a week later, we closed our first major deal with a frontier lab, and the rest is kind of history.”— Grace Li, DesignArena co-founder
That geographic layer is a specific asset for labs shipping design-generating models globally. A model tuned to Western minimalism ships differently into markets with different visual conventions, and Intelligence sits on the data that quantifies the gap.
The $60M ARR figure lands in a market that has already produced one high-profile failure. Yupp raised $33 million from Chris Dixon's a16z crypto fund, signed frontier labs as customers, and reached 1.3 million users before shutting down earlier this year. Human-feedback marketplaces are not automatic winners.
“web dashboards in Asia tend to have a more maximalist design style.”— Grace Li, DesignArena co-founder
LM Arena is the counterexample. Its text-response ranking platform raised a $150 million Series A in January, four months after formally launching its paid product. Text and design appear to be separate markets with room for a leader in each, and DesignArena is positioning to be that leader on the visual side.
The pitch to labs got sharper last week, when a breach at Hugging Face highlighted how automated benchmarks can be gamed or manipulated. Human rankings are slower and more expensive, but they're harder to spoof at scale. That's the argument Intelligence is selling, and at $60M ARR, enough labs are buying to make the case.
The open question is whether human-feedback data compounds into a moat or commoditizes. Yupp's collapse suggests the latter is possible even with real customers and real users. DesignArena's lead on scale — four times Yupp's user base at shutdown — is the argument that the winner in this category will be the platform with the deepest preference graph, not the one with the best interface.
For frontier labs racing to ship image and design models that actually get used, the seed round matters less than what it signals: labs are willing to pay eight figures a year for taste data they cannot generate internally. Whichever platform accumulates the largest, most granular preference graph across regions and formats becomes a bottleneck vendor to the entire model layer above it — a position worth considerably more than $7.9 million buys today.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
