Home
About

Which AI wrote this?

ModelGuessr is a free browser game where you read anonymous LLM answers and guess which model produced each one. Under the hood, it's a crowdsourced study of how we recognise different AI models.

How it works

Every prompt is sent ahead of time to a handful of large language models — GPT, Claude, Gemini by default, plus any OpenRouter model a host wants to mix in. Players see the prompt and the anonymised answers, then assign each answer to the model they think wrote it.

When you guess, you can optionally tap a few style tags (verbose, uses lists, hedges, confident…) explaining your reasoning. Those tags aggregate into a public Model signatures dataset showing exactly how players recognise each model.

Daily Challenge
One fixed set of prompts per day, one shot. Global leaderboard. Solo-friendly.
Rooms
Create a private code to play with friends, or join a public room. Up to 12 players per room.
Arena
Blind head-to-head comparisons. Pick which answer is better, contribute to a ranking of LLM quality.
Stats
Leaderboards, confusion matrix, model signatures. The dataset grows with every game played.

Why it matters

As language models get better, their outputs blur together — but real stylistic differences remain. ModelGuessr turns that into a game, then into open data. Every guess + reason contributes to a corpus that didn't exist before: a quantified map of which tells humans actually use to tell GPT from Claude from Gemini.

Nothing here is official. Answers are generated via OpenRouter and belong to their respective providers. Prompts are either seeded by us or suggested by players and approved by an admin.

What we do and don't store

See the Privacy page. In short: a single anonymous cookie (no email, no account), your nickname for leaderboards, and your guesses + style tags so we can compute fair stats. No third-party tracking.

Try today's challenge Back to home