AI Judge

Lab policy

Terms

Use AI Judge as a public bench, not a private product. These terms are written in ordinary English for a personal project.

Effective 28 August 2026 · Last updated 28 August 2026 · Operator MiMs

The deal

By using https://ai-judge.genxmims.org you agree to these terms and to the Privacy and Cookies pages. If you do not agree, do not use the lab.

MiMs runs this as a hobby. It is not a company, not a paid service, and not a certified benchmark bureau.

Who may use it

You must be at least 13. Discord’s terms also apply when you sign in. If you are under 18, you need a parent or guardian’s permission before you spend money through an OpenRouter key.

What this site is

  • A shared bench that sends versioned prompt packs to models, runs validators, and scores answers with a seeded blind judge panel.
  • A public leaderboard built from scored, non-cancelled runs on this host.
  • A playground for shorter judged chats.
  • Not legal, medical, or safety advice. Not an official ranking of any vendor.

You are publishing

Runs, playground chats, published packs, and leaderboard scores are public lab records. Other people can read them. Only run controls (pause, resume, cancel, retry) are limited to the launcher. Pack drafts stay yours until you publish; published packs cannot be edited or deleted.

You grant MiMs a worldwide, royalty-free permission to host, display, and keep that material on this lab so the board stays reproducible. You still own what you wrote. You promise it is yours to publish and that it does not break the law or these terms.

Your key, your bill

Model calls go through OpenRouter. In production you bring your own key. This site does not sell credits and does not refund OpenRouter spend. You are responsible for that bill and for keeping the key safe in your browser.

OpenRouter’s terms apply to those calls. Discord’s terms apply to login.

Do not

  • Break the law or ask the lab to help you break it.
  • Write sexual content involving minors, or tasks about weapons, pathogens, or crime-as-the-job — including in custom packs.
  • Try to break, scrape destructively, or overload the host.
  • Impersonate staff, probe other people’s keys, or treat this as a private API for a product you sell without asking.
  • Paste secrets, other people’s personal data, or anything you need to stay confidential.
  • Use output as if a human signed it. Model text is untrusted.

MiMs may hide content, pause a run, or block an account that breaks this. There is no appeal process beyond Discord.

Shared knobs

Default concurrency, trials, budget, and timeouts are shared for everyone on this host. Changing them in Settings changes the lab, not only your session. Be careful.

No uptime promise

The lab can be down, slow, wiped, moved, or shut off. Runs can fail mid-stream. Scores can be wrong. Seeds and hashes are there so you can see what happened — not so we guarantee a result.

As-is

The site is provided as-is, with no warranty of any kind. To the extent the law allows, MiMs is not liable for lost money (including OpenRouter spend), lost data, or anything you do with model output. If a court ever finds a liability that cannot be disclaimed, it is limited to zero — this is a free personal project.

If we disagree

Talk to MiMs on Discord first (MiMs, id 292675388180791297).

This is not a registered company, so these terms do not name a country. If a court is ever required, the laws of the operator's place of residence apply, and that is the venue. We are saying that plainly because a hobby lab should not pretend to be a Delaware corporation.

Changes

These terms can change. The date at the top is the version you are looking at. Keep using the site after a change and you accept the new terms.