Some are, most are not, and the format is not what decides it. A 2024 study administering the WAIS-IV remotely found scores that matched in-person testing at correlations above .90, so taking a test through a browser does not by itself corrupt the result. What corrupts the result is how the test was built: whether the questions were written by someone trained in psychometrics, whether the scoring was calibrated against a real sample, and whether the site has any incentive to hand you a flattering number.
What actually makes an IQ test accurate
Three properties do the work, and none of them are visible from the landing page.
Reliability is whether the test gives you a similar score when you take it again. A test with poor reliability is measuring noise. Validity is whether it measures the thing it claims to. A quiz that mostly tests reading speed can be perfectly reliable and still not be an intelligence test. Norming is the calibration step: scores only mean something relative to a reference population, and a test that has never been administered to a representative sample cannot honestly tell you your percentile.
Most free online tests fail on the third. Writing 30 plausible questions is easy. Establishing what the score distribution looks like across thousands of representative test takers is expensive, and a site giving the test away has little reason to pay for it.
The score inflation problem
There is a commercial reason so many online tests return numbers in the 120s and 130s. A flattering score gets shared. A score of 103 does not get posted to a group chat, and a site that depends on referral traffic learns this quickly.
Inflation happens two ways. The test can be too easy, so that most takers cluster near the ceiling and the scale gets stretched to make the cluster look exceptional. Or the scoring curve can simply be generous, mapping 70% correct onto 130 because nothing stops it. Neither is detectable from inside the test. The tell is population-level: if a test's own published distribution shows most users above 120, the calibration is wrong, because by definition only about 9% of people are.
A useful sanity check is to take two unrelated tests. Genuine cognitive ability is stable, so a 30-point gap between two results means at least one of them is badly calibrated.
What online tests measure less well
Even a carefully built online test loses things a clinical assessment captures. There is no examiner watching how you approach a problem, which is where a psychologist picks up on strategy, frustration tolerance, and whether a low score reflects ability or anxiety. Timed subtests that depend on manipulating physical blocks cannot be delivered through a browser at all.
Conditions are uncontrolled. A clinical assessment happens in a quiet room with a trained administrator. An online test happens wherever you were sitting, possibly with a phone lighting up beside you. That variance is not random noise either, since it systematically depresses scores for people testing in worse conditions.
Nobody is checking whether you looked anything up. Self-administered tests rely entirely on the taker not cheating, which is fine for personal curiosity and disqualifying for any official use.
How to read a score from an online test
Treat it as an estimate with a wide margin. Even professionally administered IQ tests report a confidence interval, usually around plus or minus five points; an online test deserves a wider one.
The more informative part of a good result is the category breakdown. A total of 112 tells you little. Knowing that it came from strong pattern recognition and noticeably weaker verbal reasoning tells you something you can act on, and the shape of that profile is more stable across tests than the headline number.
Take the result seriously as a signal and lightly as a fact. If a test says 95 and you were exhausted and taking it on a phone on a train, the number is not describing your ability. If three separate tests taken under reasonable conditions all land near 95, that is worth paying attention to.
Where IQ SpeedRun sits
Worth being direct about our own test. It uses the item formats standardized batteries use, which is matrix reasoning, number series, verbal analogies, and spatial rotation, and it reports on the conventional scale where 100 is the midpoint and 15 points is one standard deviation. Each session draws 25 questions from a 30-item bank in random order, so a retake is not the same paper.
It is not clinically administered, it has no supervised norm sample, and it cannot be used for diagnosis, school placement, or entry to a high-IQ society. We say this on every test page rather than in a footnote. It is built to give you a reasonable estimate and a useful breakdown of where your strengths sit, and it should be read as exactly that.
Questions people also ask
Can an online IQ test be used for Mensa?
No. Mensa accepts only supervised, standardized tests from its approved list, administered under controlled conditions. No unsupervised online test qualifies, including ours. Mensa also runs its own supervised admission test, which is the usual route.
Why did I score differently on two online tests?
Different calibration, most often. Two tests that were normed against different samples, or not normed at all, will place the same performance at different points on the scale. Content differences matter too: a test heavy on vocabulary will favor a different person than one heavy on spatial rotation.