Live projects · AI-built, AI-tested

What today's AI models
can actually build.

Side-by-side model comparisons, a studio of playable browser games, preregistered research experiments and explainer videos. Every project below is live and one click away.

3,000+recorded model runs
28models compared
43playable browser games
38experiments with results
Model Reviewer gallery: the same prompt rendered by many AI models
Model comparison

AI Model Reviewer

The same creative prompts given to every model and coding harness: Blender scenes, Three.js worlds, SVG animation, dashboards and games. Each result opens live in the browser, with the cost, time and harness behind it.

  • 3,003 runs
  • 2,472 finished builds
  • 28 models
  • 22 harnesses
  • 562 prompts
Source on GitHub, private · access on request
Ashen Vow: a 3D duel in a ruined courtyard
Game studio

AI Game Studio

Browser games built end to end by AI agents, from one-prompt arcade games to deeper builds with levels, saves, music and touch controls. Highlights: Ashen Vow, a 3D sword duel in a ruined courtyard, and Ember Hollow, a pixel action-RPG with a procedural cave and a two-phase boss.

  • 43 games
  • 3D and pixel art
  • Desktop and touch
  • No install
Source on GitHub, private · access on request
JevLab: when does a model answer beat code?
Research lab

JevLab

Preregistered experiments on where a narrow, typed model answer helps software, and where plain code or a simple baseline does better. Every experiment ships with recorded calls, results with intervals, and an interactive playground.

  • 38 experiments with results
  • 756 recorded calls
  • 35 playgrounds
Source on GitHub, private · access on request
Video

AI Coding Arena on YouTube

Visit the channel →