Fairness rules

  • Every model receives the exact same prompt text.
  • Runs use the same prompt, with model-specific API limits recorded alongside each output.
  • The first response is kept. No retries, repair prompts, or follow-up.
  • Outputs retain model version, timestamp, token usage, and latency when available.
  • Model names remain hidden until after the visitor votes.

Security rules

Model-generated HTML and JavaScript are untrusted code. Build outputs render inside sandboxed iframes with scripts allowed but without same-origin privileges.

<iframe sandbox="allow-scripts" referrerpolicy="no-referrer"></iframe>

Generated HTML is never injected into the main page. SVG outputs are isolated or sanitized before publication.

What this is not

This is not an ELO leaderboard, a scientific benchmark, or a claim that one model is globally better than another. It is a clear, repeatable way to inspect real model behavior side by side.