AI Benchmark Exam home

AI Benchmark Exam

Human vs AI

Attempt an AI benchmark exam

This exam draws its questions from the MMLU (Massive Multitask Language Understanding) benchmark. Modern large language models have fully saturated MMLU, scoring 100%. Take the same exam and see how you compare.

Test rules

Twenty Questions

Each test is randomly sampled from the MMLU question bank of 10,973 questions.

Thirty Minutes

Move freely between questions. The test submits automatically when time expires.

Cross-disciplinary

The benchmark spans 48 academic subjects.

Scoring Rules

Correct answers earn 1 point; incorrect answers lose 0.5 to discourage blind guessing.

Exam language
Exam type

Start a new exam?

You already have an exam in progress. Starting a new exam will begin a fresh attempt.