Metric
elo (Elo)
Status
Still discriminating
Category
Design & Generation
Maintainer
Design Arena (designarena.ai)
What it tests
Which model produces the best a native Android app, built agentically, judged by blind human pairwise voting. Design Arena shows a voter two anonymous models' outputs for the same prompt and asks which is better, so it measures real output quality on a generation task rather than a multiple-choice proxy. This is directly the kind of work a generative product is asked to do.
How it works
A voter submits a prompt, two anonymous models each produce an output, and the voter picks the winner. Votes are aggregated into an Elo rating per model in this category, alongside a win rate and a rank, using the same Bradley-Terry method as chat arenas applied to design and code generation.
How to read a score
Higher Elo is better, and ratings are relative, so a gap of tens of points is within noise and only large gaps separate models. Read it next to the win rate. It reflects human taste and the category's prompt distribution, not objective correctness.
Limitations
Human preference is subjective and can reward flashy output; each category's prompt set is curated and evolves; vote volume differs by category, so lower-traffic categories are noisier.
Leaderboard
Checked Nov 10, 2026, 07:00 UTC · scores last changed Nov 10, 2026Best Design Arena: Android Native (agentic) score per model, ranked. Each score links to its source.
Run your own prompt on the top 3
Claude Opus 5.5, Grok 4.6, Muse Spark 1.3, side by side on the same prompt, with each one's time and cost. A free account includes starter credit.
Compare the leaders
More Design & Generation benchmarks
Design Arena: Website
Blind human preference for models generating a complete, styled website.
Design Arena: Code
Blind human preference for models generating code across common front-end tasks.
Design Arena: Game Dev
Blind human preference for models generating a playable browser game.
Design Arena: Data Viz
Blind human preference for models generating a data visualization.
Design Arena: UI Component
Blind human preference for models generating a polished UI component.
Design Arena: 3D
Blind human preference for models generating an interactive 3D scene.
Design Arena: SVG
Blind human preference for models generating an SVG illustration.
Design Arena: ASCII Art
Blind human preference for models generating ASCII art.
Design Arena: Logo
Blind human preference for models generating a logo.
Design Arena: Image
Blind human preference for models generating an image.
Design Arena: Graphic Design
Blind human preference for models generating a piece of graphic design.
Design Arena: Full-stack (agentic)
Blind human preference for models generating a full-stack app, built over multiple agentic steps.
Design Arena: Mobile Apps (agentic)
Blind human preference for models generating a mobile app, built agentically.
Design Arena: Web Apps (agentic)
Blind human preference for models generating a web app, built agentically.
Design Arena: PPTX Slides (agentic)
Blind human preference for models generating a PowerPoint slide deck (python-pptx), built agentically.
Design Arena: Godot Game Dev (agentic)
Blind human preference for models generating a Godot-engine game, built agentically.
Design Arena: Game Dev (agentic)
Blind human preference for models generating a game, built over multiple agentic steps.
Design Arena: HTML Slides (agentic)
Blind human preference for models generating an HTML slide deck, built agentically.
Design Arena: Slides (agentic)
Blind human preference for models generating a slide deck, built agentically.
LMArena: Text to Image
Blind human preference between models generating images from a text prompt.
LMArena: Text to Image (Photorealistic)
Blind human preference between models generating photorealistic images from a text prompt.
LMArena: Text to Image (Text Rendering)
Blind human preference between models generating images that contain written text from a text prompt.
LMArena: Text to Image (Commercial Design)
Blind human preference between models generating product shots, ads and other commercial design from a text prompt.
LMArena: Text to Image (Portraits)
Blind human preference between models generating portraits of people from a text prompt.
LMArena: Image Editing
Blind human preference between models generating edited images from an image and an instruction.
LMArena: Multi-Image Editing
Blind human preference between models generating edited images that combine several input images from several images and an instruction.
LMArena: Text to Video
Blind human preference between models generating video clips from a text prompt.
LMArena: Image to Video
Blind human preference between models generating video clips from a starting image.
Design Arena: Android Native (agentic) is maintained by Design Arena (designarena.ai). Scores on this page are collected from openly-licensed sources and each links to its origin. See the full matrix to compare every model across every benchmark.