Quick answer: each puzzle has a difficulty rating, your answers are combined with item response theory into one ability estimate, and that is shown on the usual IQ scale (average 100, standard deviation 15) with a percentile and an 80% range. The ratings are careful estimates, not yet checked against a formal norming study, and this page says what that means.
What the test measures
Altto’s IQ test uses puzzles that need no knowledge of facts, in four areas that every test includes: pattern reasoning (grids and sequences of shapes with a hidden rule), numerical reasoning (number sequences, grids and symbol puzzles), spatial reasoning (mirror images, rotated shapes, folded paper, cube nets) and verbal and logical reasoning (analogies, odd-one-out, letter codes, logic puzzles). The Complete test adds working memory and processing speed.
The puzzles are generated by code from a fixed set of puzzle types, with a new random draw each time, so every attempt is a fresh set. For pattern grids, picture sequences and number series, the code tests each puzzle against alternative rules so that only one option fits.
The four tests
- Quick: 10 questions, about 5 minutes.
- Standard: 20 questions, about 10 minutes (15-minute limit). The all-round choice.
- Complete: 40 questions in six areas, about 25 minutes.
- Top 2% challenge: 30 reasoning questions, most of them in the hard half of the range, 35-minute limit. Built for precision around IQ 130, the top-2% line. It uses the same puzzle types and difficulty ratings as the other tests, with the harder types used more than once.
How a score is worked out
Each puzzle type has three numbers: how hard it is, how sharply it separates stronger from weaker solvers, and the chance of getting it right by guessing (one over the number of options). Your answers update a probability curve over possible ability levels, starting from the assumption that most people are near average. The centre of that curve is your ability estimate, and its spread gives the likely range. The estimate is converted to the IQ scale (100 plus 15 for each unit of ability), kept between 55 and 150, and turned into a percentile using the normal curve.
Because harder puzzles count for more and guesses are allowed for, two people with the same number of right answers can get different scores. This is the same family of methods used by professional tests; what separates those tests is the large, careful sample behind their difficulty figures.
What simulated test-takers show
To check the scoring itself, we simulate test-takers whose true score we set, let them answer each puzzle with the probability the model gives, score them, and compare. This run used 1,000 simulated test-takers per cell with a fixed random seed, so it can be repeated exactly. In each cell the bold number is the typical error: the root-mean-square gap, in IQ points, between the score the test gave and the true score. Under it, “shift” is how far the scores sat above (+) or below (−) the truth on average.
| Test | True score 100 | True score 115 | True score 130 | True score 140 |
|---|---|---|---|---|
| Quick test (10 questions) | 8.5 shift +0 | 8.7 shift -2.4 | 11.1 shift -7.6 | 14 shift -12.3 |
| Standard test (20 questions) | 6.4 shift -0.6 | 7.2 shift -1.3 | 8.2 shift -3.9 | 9.5 shift -6.9 |
| Complete test (40 questions) | 5.1 shift +0 | 5.4 shift -0.7 | 6.1 shift -1.9 | 7 shift -3.9 |
| Top 2% challenge (30 questions) | 9.2 shift -4.4 | 6.5 shift -2.5 | 5.3 shift -1.8 | 5.6 shift -2.1 |
Three things stand out. The Complete test is the most precise across most of the range. The Top 2% challenge is the most precise at 130 and above and the least precise at 100. And every test places very high scorers a little low, most of all the shorter ones.
The range the page shows is meant to hold the true score about 80% of the time. In these simulations it held between 73% and 92% of the time for the Complete and Top 2% tests at every true score, and for the Quick and Standard tests up to a true score of 130. For very high true scores on the shorter tests it held less often (about 50% on the Quick test and 66% on the Standard test at a true score of 140), so a high result on a short test deserves a wider mental margin than the page shows.
The ceilings and floors: these are what a perfect run, and pure guessing, score on each test.
| Test | Random guessing, average | Random guessing, best of 300 | All questions right |
|---|---|---|---|
| Quick test (10 questions) | 68 | 91 | 134 |
| Standard test (20 questions) | 63 | 85 | 142 |
| Complete test (40 questions) | 60 | 78 | 148 |
| Top 2% challenge (30 questions) | 82 | 106 | 150 |
All right on the Top 2% challenge reaches the 150 cap. Random guessing averages about 82 on it, higher than on the other tests because it has few easy questions to pull a guesser down, so a score under about 100 there means the test is too hard to measure you, not that your ability is that low.
What real results show
These are finished Altto tests from 26 October to 12 November 2026. The people who took them chose to, so this is not a sample of the population.
| Test | Finished tests | Average score | Scored 130+ | Scored 140+ |
|---|---|---|---|---|
| Standard (20 questions) | 1,631 | 102.0 | 63 (3.9%) | 15 (0.9%) |
| Quick (10 questions) | 559 | 102.5 | 44 (7.9%) | 0 (0.0%) |
| Complete (40 questions) | 258 | 110.3 | 21 (8.1%) | 4 (1.6%) |
| Expected for the whole population | – | 100.0 | 2.3% | 0.38% |
The averages land near 100 on the Standard and Quick tests, which is a sensible sign. The share scoring 130 or more is about 4% on the Standard test and about 8% on the Quick and Complete tests, against about 2.3% expected for the whole population. Self-selection (keen puzzle-solvers choose to take an IQ test) can explain some of that, and generous difficulty ratings can too. We cannot yet tell which, which is why the next section matters. The Top 2% challenge has no real results yet.
What we do not know yet
- Whether the difficulty ratings match real people. They are estimates. No formal norming study has been done.
- Age norms. Scores are not compared with people of your own age.
- Retest reliability. The site does not identify visitors, so we cannot follow the same person across attempts.
- Whether a long unsupervised test behaves like a supervised one. We assume it does; there is no check.
How we are going to check
From 12 November 2026 the site records, for each finished test, an anonymous line for every question that was answered: the puzzle type, its difficulty rating, and whether the answer was right. It records no names, no accounts, no timings, no country and no way to link a line to a person. Once there are about 5,000 finished tests, we will compare how often people at each score level really get each puzzle type right with what its rating predicts, adjust any rating that is off, and publish what changed and why. If the ratings turn out too generous, scores will move down, and this page will say by how much.
How to read your result
Treat your score as an estimate with a margin. Look at the range, not just the number; take the test rested and in one sitting; and for a second reading, retake it a few days later. The more important thing about the number is what it is for: curiosity, practice, or a reason to try a supervised test. If it is the last one, our page on the Mensa IQ test covers the options and their costs.
Frequently asked questions
How does Altto score an IQ test?
Each puzzle has a difficulty rating, a measure of how well it separates stronger from weaker solvers, and a guessing allowance. Your answers are combined with item response theory (a three-parameter model) into a single ability estimate. That estimate is placed on the standard IQ scale, where the average is 100 and the standard deviation is 15, and shown with a percentile and a likely range. Harder puzzles count for more, and lucky guesses are allowed for.
What does the likely range mean?
It is an 80% range: about four times in five, the true score of someone with that result should fall inside it. In our simulations it did that for most scores, but less often for very high true scores on the shorter tests, so treat it as a guide rather than a guarantee.
Has Altto’s IQ test been properly norm-referenced?
Not yet. The difficulty ratings are careful estimates, not the result of a formal study of a representative sample, and there are no separate norms by age. From 12 November 2026 the site logs anonymous question-by-question results so the ratings can be checked against real answers once enough have built up.
Are the simulation results real?
They are simulations: made-up test-takers whose true scores we set, who answer each puzzle with the probability the scoring model gives. They show how well the scoring recovers a known score. They cannot show whether the difficulty ratings match real people, which is the open question above.
Why do high scorers get scored a little low?
Scoring starts from the assumption that most people are near average and lets the answers pull the estimate away from it. With few questions, that pull is not always complete, so very high true scores are on average estimated a few points low. The simulation table shows the size of the shift for each test.
Is this a clinical or official IQ test?
No. It is for personal interest. An official IQ score needs a supervised test given by a psychologist, and Mensa and other high-IQ societies accept only supervised tests.