About The IQ Gym
The IQ Gym is an educational IQ training and practice platform. It covers 49 question families across six cognitive domains, verifies every answer before a question can enter the bank, and reports every score as a confidence band rather than a single number.
Key facts
- Free test
- 20 questions in 12 minutes, across abstract, numerical, verbal and spatial reasoning. No sign-up.
- Full tests (free account)
- A fixed test of 49 questions in about 50 minutes, and an adaptive test of 24 questions in 30 minutes. Both add working memory and processing speed, for six areas.
- How a score is reported
- A 90% confidence band on the familiar IQ scale (average 100, standard deviation 15), never a single number: about ±18 points on the free test, ±13 on the fixed test and ±16 on the adaptive one. Takers aged 16 and over also get a modelled percentile.
- Timing
- Enforced by the server, not the browser.
- Answers
- Every question comes with a worked explanation, shown free after the test.
- Norms
- Provisional: built from the questions' design, to be recalibrated from real answers. A reliability figure will be published once there is enough data.
- Price
- Free. There is no paid tier, no subscription and no checkout.
- Limits
- An educational estimate, not a clinical assessment, and not accepted by Mensa or other high-IQ societies. Practice improves test performance; it does not raise underlying intelligence.
How each of these works, with sources: methodology.
Why we built it
Most online IQ tests are built to produce a flattering number and a paywall. Very few will tell you how confident they are, what their questions actually measure, or where their norms come from. We wanted the opposite: a training tool that is straight with you about what it can and cannot do, and that shows its working.
What makes the questions different
Every question in the bank is original, written to have exactly one correct answer, and comes with a worked explanation. Questions arrive one of two ways:
- Generated procedurally, then checked by a solver that re-derives the answer from the puzzle and rejects anything ambiguous or under-determined. For a few families that solver shares code with the generator, so it cannot catch every generator mistake.
- Hand-authored, then lint-checked against rules from the Haladyna & Downing multiple-choice item-writing guidelines: no all-of-the-above options and no length cues that give the key away. The check is mechanical: it cannot prove that the key is the only defensible answer.
Published tests and puzzle books informed the question formats we cover. None of their content was copied — the bank is entirely our own.
How we measure
Scoring uses an Elo-as-Rasch calibration: each question carries a difficulty rating and each response updates both that rating and your ability estimate. The ratings in the bank today are the ones our authors assigned — live calibration begins as responses accumulate, and until then they are estimates, which is why our norms are labelled provisional. Ability is estimated with a Bayesian EAP method and converted to the familiar 100 ± 15 scale. Because every estimate carries uncertainty, we report a 90% confidence band and a per-domain profile instead of a single headline number.
What we promise, and what we don't
We will help you get better at IQ-style tests: the patterns, the strategies, the speed, the confidence. That improvement is real and it is measurable. We will not tell you we are raising your underlying general intelligence, because the evidence for that kind of far transfer is weak — and every score here is an educational estimate, never a clinical or professional assessment.
When we get it wrong
Verification catches most bad questions before they reach you, but not all of them. Every question carries a 'Report a problem' control while you answer it — on the free test, on both full-length assessments, and in drills — and every scored test's worked review carries one beside each question afterwards. Every report lands in a review queue that a person works through. If a question turns out to be wrong or ambiguous, it comes out of the bank.