Miju Labs

← All specialists

Design Arena

5.3M users, $60M ARR, under a year old — the consumer-arena route into the same data. Nobody has said whether that number is gross or net.

medium confidence4 minupdated 2026-08-30design · arena · preference data · consumer
Vertical
Design
Founded
2025 (Arcada Labs, YC S25)
Headquarters
Not disclosed
Raised
$7.9M seed
Last valuation
Not disclosed
Revenue
$60M ARR claimed; gross or net not stated
Status
Active — less than a year old
Who runs it · 2 people in the index

Latest (Sep 2026): On 3 Aug 2026 the company, now branded Intelligence (intelligence.ai; earlier Arcada Labs, YC S25), announced a $7.9M seed led by Index Ventures with Conviction, A*, Valkyrie and YC. It claims growth from $5M to $60M ARR in six months with a team of about 10, 5.3–5.5M users in 190+ countries, and says it is profitable. Index names Google, OpenAI, Thinking Machines Lab, xAI and Meta as users of Design Arena for model evaluation. Through mid-2026 it has shipped sister arenas (Prediction Arena, audio realism, slides, mobile/fullstack agents); whether $60M ARR is gross or net is not disclosed.

What Design Arena is saying
MiniMax H3 Max by @fal takes the #1 spot on Video Editing by Design Arena with an Elo of 1373. The model ranks ahead of MiniMax H3 by @MiniMax and @GoogleDeepMind's Gemini Omni Flash and Gemini Omni Flash 1.1. It's also in a league of its own on the Pareto frontier for Preference vs Speed and Preference vs Price. Congratulations to the @fal team on post-training an exceptional model.
154818550KView on X ↗
BREAKING: MiMo-V2.6-Pro by @XiaomiMiMo lands at #8 overall (#3 open-weight) on Design Arena with an Elo of 1338. This is an impressive 54-point and 22-position increase from MiMo-V2.5-Pro. MiMo-V2.6-Pro also reaches #4 overall in Website (#2 open-weight) and #6 overall in Agentic Frontend Development (#2 open-weight), showing strong performance across both direct generation and agentic coding. This places @XiaomiMiMo's new model among leading models such as GPT-5.6 Sol by @OpenAI, Claude Opus 5 by @Anthropic, and Kimi K3 by @MoonshotAI. Congratulations to the @XiaomiMiMo team on returning to a top-10 placement on Design Arena!
221418530KView on X ↗

Design Arena — corporate name Arcada Labs, YC S25, founded by Grace Li and Kamryn Ohly — raised a $7.9M seed led by Index Ventures with Conviction, A* and Valkyrie in August 2026. It reports 5.3 million active users and $60M ARR, at less than a year old (TechCrunch, 3 Aug 2026).

The mechanism is the opposite of Contra Labs and Taste Labs. Those companies pay a small number of vetted professionals for considered judgement. Design Arena gives free frontier-model access to a very large consumer crowd, watches which output they prefer, and sells the resulting feedback data to enterprise AI labs (TechCrunch). The consumer side is the supply acquisition channel and the product is the exhaust.

The $60M number, and what it is not

$60M ARR on $7.9M of seed capital in under a year is the hardest single figure in the Design and UI/UX vertical, and the reason design is described as validated rather than speculative.

Gross or net is not stated

The reporting does not say whether $60M is gross revenue or net revenue. The business is two-sided — consumers receive free model access, labs pay for data — so there is a real possibility that the top line includes model-access cost passed through, or credits, or both. Every other number on this page is sourced; this one is a company-reported figure relayed by press with no accompanying margin, cost of revenue, or customer count. Treat it as gross unless proven otherwise, and see GMV is not revenue before putting it in a comparison table. The atlas has no basis for a margin estimate here.

The arena model also carries an unusual cost structure. Free frontier-model access at 5.3M users is not free to provide. Whoever bears that inference bill — Design Arena, or the labs supplying models in exchange for the data — determines whether this is a high-margin data business or a subsidised distribution channel. Nothing public settles it.

The comparable that died

The same reporting names the case against: Yupp AI shut down in 2026 despite having frontier-lab customers, with 1.3M users and $33M raised. In the same period LM Arena raised $150M in January 2026 (TechCrunch).

Read those three facts together and the arena category looks winner-take-most rather than merely competitive. Yupp had customers and capital and still failed; LM Arena raised at a scale that implies the market believes one arena wins. A crowd-preference arena has no supply moat — anyone can run the same offer tomorrow — so the defensible asset is the traffic and the reference status of the leaderboard, both of which concentrate.

That is the structural weakness. Contra Labs' narrated trajectories are hard to reproduce; 5.3M consumer sessions are hard to reproduce quickly, which is not the same thing.

What the crowd can and cannot judge

Design Arena publishes a public leaderboard (designarena.ai/leaderboard). The crowd is large enough to resolve preferences with tight error bars, and consumer preference is a legitimate signal for consumer-facing generation.

But the design vertical's most cited empirical finding cuts directly against a consumer crowd. Amplify's memo on Taste Labs records that Runway, Figma and Adobe found optimising against general user preference degraded quality as judged by experts (Amplify Partners). And the TASTE work built by Lica World and Contra.Work Inc. shows nine pre-trained systems all falling below 0.55 macro agreement with a five-designer majority against a 0.741 single-rater ceiling (arXiv 2605.20731) — a professional standard that a consumer vote does not measure.

So the two routes are not competing for the same contract. Arena data is cheap, abundant and directionally consumer-correct. Expert panels are expensive, scarce and the only thing that measures the professional standard. A lab plausibly buys both.

What is not known

Whether $60M is gross or net. Customer names, customer count and One customer is a binary event. Who pays for the free model access. Cost of revenue and therefore any margin. Valuation at the seed. Headquarters. Whether the consumer crowd is compensated at all, and on what terms.

What to learn from it: the arena route buys scale and a leaderboard for very little capital, and it buys neither professional-standard signal nor a supply moat — so the number to chase in your own niche is not user count, it is whether your buyer's failure mode is consumer preference or expert standard.