Intelligence lands $7.9M seed funding as its human-taste platform hits $60M ARR

4d ago0 views

Key takeaways

Intelligence raised $7.9M in seed funding led by Index Ventures for its Design Arena platform.
The company reports $60M in ARR and 5.3 million users ranking AI-generated design outputs.
Frontier AI labs pay for access to aggregated human preference data to improve their visual models.

Intelligence, the startup operating the AI design evaluation platform Design Arena, announced Monday it has raised $7.9 million in seed funding led by Index Ventures, with participation from Conviction — the fund run by Sarah Guo and Mike Vernal — alongside A*, Valkyrie, and other investors. The company says it is currently generating $60 million in annual recurring revenue, a striking figure for an early-stage startup that launched just months ago. Co-founder Grace Li says the idea grew directly out of a problem she and a group of college friends ran into while building an AI game engine shortly before their 2025 graduation: the models could produce technically functional games, but none of the outputs were actually enjoyable to play.

That subjective gap — what makes something genuinely good rather than merely correct — led Li and her co-founders to explore how human judgment could be gathered at scale. The result was Design Arena, a platform that presents users with side-by-side comparisons of AI-generated visual content, asking them to pick the better output across categories including websites, images, and other formats. Users log in, submit a prompt, and work through a series of A-versus-B rankings until outputs are sorted from best to worst. The platform now counts 5.3 million users worldwide.

For individual users, Design Arena functions as a polished model router with a familiar chat-style interface. The deeper business, however, sits on the enterprise side, where frontier AI labs pay to access the aggregated preference signals that millions of user rankings generate. Because users simply want the best output available and have no particular loyalty to any underlying model, their judgments provide relatively clean feedback about what real people actually value in AI-generated design — something automated benchmarks struggle to capture reliably.

Intelligence also benefits from persistent user accounts, which allow the company to track how aesthetic preferences shift across regions and over time. Li has noted, for instance, that web dashboards in Asia tend to skew toward more maximalist design styles. This kind of longitudinal, geographically segmented data gives the platform an edge over one-time evaluation snapshots, and it serves as a complement to automated benchmarks that, as a recent security incident at Hugging Face illustrated, can be manipulated or gamed.

The human feedback market is not without cautionary precedents. Yupp, a competitor that raised $33 million from investors including a16z crypto's Chris Dixon and claimed over 1.3 million users, shut down earlier this year after failing to build a sustainable business. But LM Arena, which applies a similar ranking approach to text-based AI responses, raised $150 million in a Series A in January just four months after launching its paid product, suggesting the sector still has real momentum for companies that can demonstrate durable demand.

The bigger picture

Intelligence's $7.9 million raise is small in absolute terms, but the $60 million ARR figure attached to it is the number that matters most to anyone watching the AI evaluation space. That kind of revenue at seed stage signals that frontier labs — the Anthropics, OpenAIs, and Googles of the world — are actively writing checks for human preference data rather than waiting to build proprietary feedback infrastructure themselves. That dependency is a strategic opening Intelligence should move quickly to lock in through long-term contracts before larger players decide to replicate the model in-house.

The competitive risk is real but probably not immediate. Building a platform that attracts millions of users willing to voluntarily rank AI outputs is harder than it sounds — Yupp proved that scale alone does not guarantee sustainability. Intelligence appears to have threaded a specific needle: users get useful outputs from a multi-model router, and the company monetizes the behavioral exhaust of that interaction. LM Arena did something similar for text and is now valued far beyond seed territory. If Intelligence can establish comparable dominance in the visual and design category, the comparison to LM Arena will become very favorable for future fundraising conversations.

The Hugging Face breach mentioned in the background context is worth flagging specifically because it illustrates a systemic vulnerability that benefits Intelligence's pitch. When automated benchmarks get compromised or gamed, human-in-the-loop evaluation becomes a more credible alternative — not just a philosophical preference but a practical risk management tool for labs that need trustworthy model rankings. Regulators in the EU and elsewhere who are scrutinizing AI evaluation methodology will also be paying attention to whether human feedback platforms like Design Arena can offer more auditable, manipulation-resistant signals than leaderboard systems that have proven brittle.

LagPing's take

We're covering Intelligence and Design Arena because the question of how you measure whether an AI output is actually good — not just technically correct — is one of the most consequential and underreported problems in the AI industry right now. Most coverage of AI evaluation focuses on benchmark scores and automated metrics, but this story gets at something messier and more interesting: taste. The fact that a company born from a college project about video game fun has turned that question into $60 million in annual revenue and a seed round from Index Ventures tells us something real about where the AI toolchain is heading. We also think the Yupp cautionary tale is important context here — this market has already claimed one well-funded casualty, and readers deserve the full picture rather than a pure fundraising announcement. We'll be keeping an eye on whether Intelligence can hold its position as the design-focused counterpart to LM Arena's text dominance, and whether the major labs eventually decide to bring this capability in-house.

X / Twitter WhatsApp

View at TechCrunch AI

Shop AI & tech on Amazon

As an Amazon Associate, LagPing earns from qualifying purchases. Product links are affiliate links.