Today we updated the scoring model behind the AREALME IQ Test. The October 15 release uses recent, carefully screened completions from across our language interfaces to recalibrate the score, with clearer information about the evidence behind it.
If you have taken the test before, your score in the new version may be lower, even with similar performance. We have corrected the calibration of the data model. A lower number caused by that change does not mean your cognitive ability has declined.
This is an update we want to explain openly. A useful score needs a defensible reference, and improving that reference matters more to us than preserving a flattering number.
Thousands of Recent Completions, With a Clear Definition
AREALME has been developing IQ tests for more than a decade. Our lifetime participation figures exceed seven million test takers across languages. That reach gives us a substantial practical advantage: we can gather fresh evidence quickly, investigate how questions behave, and revisit the model as the test evolves.
Lifetime participation and the sample used for a particular scoring release are different measures. For this update, we used the following eligible completions collected from September 26 through October 11, 2026:
| Test mode | Eligible completed records |
|---|---|
| Quick, 49 questions | 7,563 |
| Complete, 119 questions | 2,993 |
That is 10,556 qualifying completion records in a 16-day window. Some participants completed both modes, so these figures must not be presented as 10,556 different people. Each mode uses the first eligible completion per identifiable browser; browser identification also cannot establish a unique person across devices.
The two modes are calibrated separately. All interface languages contribute to a pooled reference, and the same approach serves all 34 language versions. This lets smaller language communities benefit from the wider evidence base without building fragile scoring tables from tiny local samples.
Data Quality Is Part of the Model
A large audience only becomes useful evidence when the records are suitable for the question being studied. For this release, our calibration process checks that:
- The questions match the current test. Records must contain the expected question set for their mode.
- The test was completed without interruption. Resumed sessions and technical skips are excluded from the modeling sample. Users can still resume a test and receive a result.
- Repeat completions do not dominate the reference. We select the first eligible completion for each identifiable browser and mode.
- Timing and recorded answers pass quality checks. Very fast completion patterns are filtered, and replayed score calculations are checked against the recorded result.
- Later data is used to evaluate the method. We compare the updated scoring with the previous model instead of judging it only on the data used to fit it.
This combination of scale, screening, and public documentation is central to AREALME’s approach. It gives us a stronger basis for improving the test than a fixed scoring table that is never checked against actual participants. It also makes our work open to scrutiny: readers can see the sample dates, counting rules, model version, and validation results.
What Improved in the Validation
We fitted the evaluation model using data through October 5, then evaluated it on October 6-11 completions: 2,228 Quick records and 882 Complete records. We also removed identifiable browser overlap across the training and evaluation samples, including overlap between modes.
The evaluation data included 55 pairs of Quick and Complete completions. The average absolute difference between the two displayed scores fell from 11.7 points to 7.0 points, a reduction of about 41%.
That is encouraging evidence that the two modes give more consistent results. Most of the improvement came from pooling and recalibrating the reference; the additional contribution from age adjustment was small.
The scope of that result matters. Fifty-five paired completions are a limited sample, and agreement between our own two modes is not a measurement of accuracy against a professionally administered IQ test. The earlier evaluation dates had also informed exploratory checks; October 11 provided fresh confirmation. After this evaluation, the release reference was refitted using the eligible September 26-October 11 records.
You can inspect the aggregate model and validation summary and the explanation in our Data Disclosure Center.
Why Your New Score May Be Lower
A comparison score depends on both your answers and the reference used to interpret them. The new model maps performance onto an updated reference scale centered at 100, with a standard deviation of 15.
When we scored the same evaluation records with the previous model and the updated method, the averages changed as follows:
| Evaluation measure | Previous model | Updated method |
|---|---|---|
| Quick average score | 110.8 | 98.6 |
| Complete average score | 109.1 | 98.6 |
These are group averages from the evaluation, not a conversion formula for an individual result. You should not subtract a fixed number from an old score to estimate a new one. The size and direction of a change depend on performance and the applicable reference.
Your older saved results keep their original scores and model. We do not rewrite your history. When comparing results, check the scoring version as well as the date. A difference across model versions is not, by itself, evidence that you have become better or worse at the test.
Age Is Used Carefully; Gender Does Not Affect Scoring
The update uses limited age adjustments for the broad groups 13-17, 18-34, and 35-54. Those adjustments are pulled toward the overall reference and capped at plus or minus five points. Other ages, and people who choose not to provide an age, use the overall reference.
Gender is not a scoring input. Providing demographic information remains optional. Your age group stays private by default and appears on a certificate only if you choose to include it.
A Clearer Record Behind Every Score
The test page now brings author information, public data links, and a collapsible Current version · score reference explanation together. It identifies model n20261015a, the sample period, and the completion counts. The certificate also includes a brief reference note in its footer.
The data center separates the current scoring reference from the older, separately dated question statistics. That distinction makes it easier to understand which evidence supports which claim. Public downloads contain aggregate information, without individual response records, personal identifiers, or answer keys.
Our standard for professional work is to make the method inspectable, test changes against evidence, and explain the consequences to the people taking the test. These are self-selected online participants, so the reference does not represent the entire world population. The result remains an estimate for learning and self-exploration, rather than a clinical assessment.
Take the updated AREALME IQ Test or explore the data behind it. Your score, cognitive profile in Complete mode, and optional certificate remain free, with no email address or account required.
Comments could not be loaded. Check your connection and try again.