809,280 completed assessments and growing
Updated hourly
We place an AI model’s score and profile against human population percentiles, and write up what the comparison shows.
What the baseline is built from
- 154
- Assessment types
- 31
- Languages
- 220
- Countries and territories, by session
Sessions from 220 countries and territories. Geography is derived from each session’s IP address, so it says where sessions came from, not who the people are or where they are from.
Cognitive track
132,170 sessions on a single cognitive test bank, scored on 18 ADeLe scales. The bank has 399 annotated items.
Behavioural track
A human baseline of 54,431 rows across 9 personality and behavioural tests. The model’s profile is placed against it.
Citable snapshot: 671,760 completed assessments as of 2026-09-30. Cite the dated snapshot, not the live count. Corpus passport.
What we sell
Computed benchmark outputs and evaluation reports: a model’s score and profile placed against human population percentiles and scale means, plus written analysis. It is a report, not data. There is no raw data and no individual-level data in it, no dataset licence to negotiate, and no consent chain for you to manage.
- Cognitive benchmark. The model’s score against the human distribution on our cognitive test bank.
- Behavioural and personality comparison. The model’s profile against the human baseline.
How it works
- Step 1
You pick the model or models
Tell us which ones you want compared.
- Step 2
We run them through the same items
The items and the scoring rules are the ones people were scored by.
- Step 3
You receive a comparison report
Scores and profiles placed against the human baseline, with written analysis.
Tell us which model
One email with the model or models you want compared.
What we license is aggregate statistics and evaluation reports only. Employer-linked assessment rows are excluded from these research figures.