Skip to content
IQ Octopus
Start the free test

Press kit

Everything you need to write about IQ Octopus or our research. Questions get a reply from a person.

About IQ Octopus (boilerplate)

IQ Octopus is a free online IQ test made of 25 culture-fair visual and number puzzles, generated fresh for every attempt and scored with a Rasch model that reports an estimated IQ with a confidence range. Taking the test is free; the full result costs $4.95 once, with no subscription. IQ Octopus is operated by Sourdough AI LLC in Warren, Rhode Island, and publishes original research, including a benchmark of how AI models score on the same test people take.

Key facts

  • 25 puzzles in five sections (pattern recognition, abstract reasoning, spatial reasoning, quantitative reasoning, logical reasoning). About 12 minutes for most people, with a hard limit of 20 minutes.
  • Puzzles are generated from rule templates for every attempt, so no two tests are identical and there is no answer key.
  • Scoring uses a Rasch (one-parameter item response) model. Results show an estimated IQ on the standard scale (mean 100, standard deviation 15) with a 95% range. The highest possible estimate is 141.
  • Free to take. The full result is $4.95, one time. No subscription, no account. Refunds within 7 days on request.
  • No ads and no sale of personal data.
  • Not a clinical or diagnostic test. It cannot be used for Mensa admission, school placement, disability assessment or employment screening.
  • Operated by Sourdough AI LLC, 223 Child St, Warren, RI 02885.

Story angles from our own data

  • Top AI models score an estimated 135 on a culture-fair IQ test. 5 models took the same 25-puzzle test people take, 5 times each, scored like a person. Claude Opus 5.5 and GPT-6-Astra averaged 135; the lowest mean was 124. The study.
  • AI is weakest at spatial reasoning. Across models, Spatial Reasoning was the hardest section (74% correct on average, from 52% to 88%), against 95% for Quantitative Reasoning.
  • No model got a perfect score. Across 25 tests, the best single result was 24 of 25 correct.
  • Can you beat an AI on an IQ test? Readers can take the same test the models took. The full result shows where they land against each model.
  • An IQ test that shows its uncertainty. Every result comes with a 95% range, and the methodology publishes the full raw score to IQ table, including the ceiling of 141.

Free tools you can embed

Publishers can embed our IQ percentile calculator or an interactive IQ bell curve with one line of code. See embeddable widgets.

Logos

Please do not recolor, stretch or crop the mark. Write the name as “IQ Octopus”, two words.

Citing our research

IQ Octopus (2026). AI vs human IQ 2026: how AI models score on a culture-fair IQ test. Sourdough AI LLC. https://iqoctopus.com/research/ai-vs-human-iq-2026

Press contact

Email support@iqoctopus.com with “Press” in the subject. We can explain the method and answer questions about the data.