Design Arena · judged on OpenVibe.Bench

Which AI designs it best?

Two designs from the same brief, the models hidden. Pick the better one and that kind's board moves. No account needed.

6 votes so far across 9 kinds.

Blind pair · Screens

Which is better?

Design a full-page dark analytics dashboard: left icon sidebar, top bar with a date-range picker and a search field, four KPI cards, one wide line chart, one bar chart, and a table of recent events. Real labels, no lorem ipsum. Inline CSS only, no network, no icon fonts, no charts from a CDN.

Design A

Phone · 390 px

Desktop · 1280 px

Design B

Phone · 390 px

Desktop · 1280 px

You are not signed in: your vote counts with the signed-out voters, up to 30 a day. Sign in to vote as yourself.

  • Pages

    Judge

    2 votes · 2 signed in · 0 signed out

    1. 1gpt-5-nano-2025-08-07 new10000
  • Screens

    Judge

    1 vote · 1 signed in · 0 signed out

    1. 1gpt-5-nano-2025-08-07 new10000
  • Systems

    Judge

    1 vote · 1 signed in · 0 signed out

    1. 1gpt-5-nano-2025-08-07 new10000
  • Components

    Judge

    0 votes · 0 signed in · 0 signed out

    No votes yet. Be the first to judge it.

  • Illustration

    Judge

    0 votes · 0 signed in · 0 signed out

    No votes yet. Be the first to judge it.

  • Motion

    Judge

    1 vote · 1 signed in · 0 signed out

    1. 1gpt-5-nano-2025-08-07 new10000
  • Charts

    Judge

    1 vote · 1 signed in · 0 signed out

    1. 1gpt-5-nano-2025-08-07 new10000
  • 3D models

    Judge

    0 votes · 0 signed in · 0 signed out

    No votes yet. Be the first to judge it.

  • Shaders

    Judge

    0 votes · 0 signed in · 0 signed out

    No votes yet. Be the first to judge it.

Elo from blind pairwise votes, K = 32. A tie is half a win; both bad is a game for neither model; new means fewer than 10 games. Signed-out votes are capped per voter and per day.

Start a round

Describe a page or a component once; three models build it, and the round joins the blind pairs. You cannot judge your own round.

Start a round

Everything is open

Every brief, every output and every vote can be downloaded from Bench, per kind. No voter is named.

Pages dataset The gallery on Bench

What each model costs 2 kinds

Design page arena · Cost

#ModelEntriesUsableTotal costSolvedCost per solvedMedian costMedian time
1 gpt-5-nano-2025-08-07 3 3 $0.0042 3 $0.0014 $0.0014 19.6 s