Design Arena

Design Arena is a crowdsourced benchmarking platform for AI-generated design, built by Intelligence. The site presents the same creative prompt to multiple leading AI models simultaneously, displays the results side by side, and lets users vote on which output is best — with those votes directly powering public leaderboards.

What It Is

Design Arena describes itself as "the world's first crowdsourced benchmark for AI-generated design." Rather than relying on curated expert opinions or automated metrics, it collects millions of pairwise human preference votes across categories like websites, slides, mobile apps, game dev, 3D design, data visualization, UI components, images, logos, SVGs, text-to-speech, ASCII art, and video. The platform covers models such as Claude Opus, GLM, and GPT variants, evaluated anonymously to prevent brand bias.

How the Tournament Works

Each voting session follows a structured tournament format:

• A user enters a prompt; a category is selected (manually or automatically).
• Four models are randomly drawn from the active pool and given the identical prompt simultaneously.
• Two randomly paired models are shown side by side, anonymously, for a blind vote.
• Winners and losers are re-matched across up to five battles, producing a complete 1st–4th ranking per session.
• Every pairwise comparison result feeds directly into leaderboard calculations with no editorial filtering.

Rankings are computed using the Bradley-Terry model, a statistical framework for pairwise comparison data. The algorithm iterates until strength estimates stabilize (threshold: 0.0001) or reaches 200 iterations, then converts normalized strengths to ratings via Rating = 400 × log₁₀(strength). Models with fewer than 50 pairwise comparisons are filtered from main charts; preliminary status is applied until ~200 comparisons are reached.

What People Are Building

According to the platform's own analytics over a recent 30-day window, the most common prompt categories submitted by users include Productivity (20.7%), Corporate (15.0%), E-commerce (14.6%), Dashboard (10.4%), and Landing Page (8.6%), with Game Dev, 3D, and UI Components also represented. This distribution reflects real-world demand patterns across agentic web development use cases.

Community and Global Reach

The platform reports a community of 5.1M+ users across 190+ countries, as stated on the homepage and About page. Community votes shape every leaderboard update in real time, with each pairwise comparison weighted equally. Design Arena maintains a blog, Discord, Twitter/X, LinkedIn, and Reddit community for ongoing discussion and methodological feedback.

Why It Matters for AI Evaluation

Design Arena's methodology is grounded in the argument that "taste is hard to measure" — good design reflects aesthetic values that automated benchmarks miss. By keeping model identities hidden during evaluation and sourcing votes from a large global community, the platform aims to surface which models "actually have taste" rather than which ones score highest on narrow technical metrics. All configurations and methodologies are publicly documented and open for community review.