Original research · October 21, 2026
AI vs human IQ 2026: how AI models score on a culture-fair IQ test
Abstract
We gave 5 AI models the same 25-puzzle culture-fair IQ test people take on IQ Octopus, 5 times each, and scored them with the same Rasch scorer. Mean estimated IQs ranged from 124 to 135. The top score, 135, was shared by Claude Opus 5.5 and GPT-6-Astra. No model reached the test ceiling of 141.
Take the same test the models took and see where you land.
Why we ran this
AI models are often described with human test scores. Most of those claims use a different test for each model, an unpublished prompt or a verbal test the model may have seen in training. We wanted a comparison that anyone can check: one fixed set of puzzles that did not exist before the test was generated, the same scorer we use for people, and every prompt and setting published.
Results
| Model | Provider | Mean estimated IQ | Range |
|---|---|---|---|
| Claude Opus 5.5 | Anthropic | 135 | 129 to 137 |
| GPT-6-Astra | OpenAI | 135 | 133 to 137 |
| Claude Sonnet 5.5 | Anthropic | 132 | 126 to 137 |
| Claude Haiku 5.5 | Anthropic | 124 | 120 to 129 |
| GPT-5.6-Terra | OpenAI | 124 | 120 to 129 |
Mean is the average of a model's 5 estimated IQs, rounded. Range is its lowest and highest single test. The test's ceiling is 141: a perfect 25 of 25 maps to 141, so scores near the top are compressed. See the methodology for the full raw score table.
Accuracy by section
Pattern Recognition (25 puzzles per model)
Abstract Reasoning (45 puzzles per model)
Spatial Reasoning (25 puzzles per model)
Quantitative Reasoning (15 puzzles per model)
Logical Reasoning (15 puzzles per model)
| Model | Pattern | Abstract | Spatial | Quantitative | Logical |
|---|---|---|---|---|---|
| Claude Opus 5.5 | 92% | 98% | 84% | 100% | 100% |
| GPT-6-Astra | 88% | 100% | 88% | 100% | 93% |
| Claude Sonnet 5.5 | 88% | 96% | 80% | 100% | 93% |
| Claude Haiku 5.5 | 92% | 80% | 64% | 93% | 87% |
| GPT-5.6-Terra | 96% | 96% | 52% | 80% | 67% |
| All models | 91% | 94% | 74% | 95% | 88% |
Section sizes per test: Pattern Recognition 5, Abstract Reasoning 9, Spatial Reasoning 5, Quantitative Reasoning 3, Logical Reasoning 3. The all-models row is the simple average of the model accuracies.
Every test
| Model | Test 1 | Test 2 | Test 3 | Test 4 | Test 5 |
|---|---|---|---|---|---|
| Claude Opus 5.5 | 137 (24/25) | 137 (24/25) | 137 (24/25) | 129 (22/25) | 137 (24/25) |
| GPT-6-Astra | 133 (23/25) | 137 (24/25) | 137 (24/25) | 133 (23/25) | 137 (24/25) |
| Claude Sonnet 5.5 | 129 (22/25) | 137 (24/25) | 133 (23/25) | 126 (21/25) | 137 (24/25) |
| Claude Haiku 5.5 | 123 (20/25) | 129 (22/25) | 126 (21/25) | 123 (20/25) | 120 (19/25) |
| GPT-5.6-Terra | 126 (21/25) | 129 (22/25) | 126 (21/25) | 120 (19/25) | 120 (19/25) |
Each cell is one test's estimated IQ with the number correct. The same test number means the same puzzles for every model (seeds 1001, 1002, 1003, 1004, 1005).
Try one · Pattern Recognition
Which tile comes next?
Tap the option you think is right, or show the answer.
Show the answer
The missing tile shows a tiny solid circle (option 3).
How to solve it
- Compare neighboring tiles and name what changes each time.
- From tile to tile, the circle gets smaller.
- The fill cycles solid, hatched and outline, then starts again.
- Everything else stays the same from tile to tile.
- The last tile shows a small outlined circle. Apply each change once more.
- So the missing tile shows a tiny solid circle.
That was 1 puzzle. The full test has 25.
How this compares with people
On the standard IQ scale (mean 100, standard deviation 15), 135 sits at about the 99.02th percentile and 124 at about the 94.5th. Read those with care. Our scale is anchored by design difficulties, not yet by a sample of adults, and the models had no time limit while people have 20 minutes. You can take the same test and compare directly: the full result shows your estimate next to each model.
How we tested
Each model took the same test a person takes on IQ Octopus: 25 puzzles in five sections (Pattern Recognition 5, Abstract Reasoning 9, Spatial Reasoning 5, Quantitative Reasoning 3, Logical Reasoning 3). The puzzles are generated from a seed, so we fixed 5 seeds (1001, 1002, 1003, 1004, 1005) and gave every model the same 5 tests. That is 125 puzzles per model.
What the model saw
Each puzzle was rendered as one PNG image, the way it looks on screen: the question at the top, the puzzle below it, and the answer options underneath, each labeled with a number from 1. Black lines on a white background. For the number series puzzles we also sent the numbers and options as plain text alongside the image. Each puzzle was a separate request with no memory of the others, and the model never learned whether it was right.
The exact prompt
System prompt:
You are taking a visual reasoning test. Each question shows one puzzle as an image: the puzzle on top and numbered answer options underneath. Exactly one option is correct.User turn: the image, then the same question a person sees above that puzzle, followed by the answer instruction:
Which tile comes next? Answer with the option number only.
Which tile completes the grid? Answer with the option number only.
Which option is the same shape, only rotated? Answer with the option number only.
Which one does not belong? Answer with the option number only.
Which number comes next? Answer with the option number only.
Which tile completes the second pair? Answer with the option number only.The OpenAI models ran through the Codex command line tool, which takes one message. For them the system prompt, the line “The attached image is the puzzle.” and the question above were sent together as that message, with the same image attached.
Settings
- Claude Opus 5.5 (Anthropic): Claude API, adaptive thinking, effort medium.
- GPT-6-Astra (OpenAI): Codex CLI, reasoning effort medium.
- Claude Sonnet 5.5 (Anthropic): Claude API, adaptive thinking, effort high.
- Claude Haiku 5.5 (Anthropic): Claude API, adaptive thinking, effort medium.
- GPT-5.6-Terra (OpenAI): Codex CLI, reasoning effort medium.
- Temperature: provider default (these models do not accept a temperature setting).
- Each model could reason before it answered, up to 16,000 output tokens for Claude models. Only the final reply was read.
- Tests run on 2026-10-21.
Scoring, identical to people
Most replies were a single number. When a model added an explanation, we took the number on its own line at the start or the end of the reply, and otherwise the first number in the reply. A reply with no valid option number counts as a skipped puzzle, and a skipped puzzle counts as wrong, exactly as it does for a person. The answers then went through the same scorer that scores every human attempt: a Rasch model that weighs each puzzle by its difficulty and turns the pattern of right and wrong answers into an estimated IQ. See how the test works for the details. A model's score on this page is the mean of its 5 estimated IQs, rounded to a whole number the way the scorer rounds.
Limitations
- Models see a rendered image and get no time limit. People have 20 minutes for the whole test. The models had no clock, and a person sees the same picture on a screen, not a file.
- The scale was built for people. The puzzle difficulty values are set for human test takers. An estimated IQ for a model translates its right and wrong answers onto that human scale. It is not a measure of intelligence in the human sense.
- Results vary by prompt and settings. A different prompt, more reasoning time or a different image size could move these numbers. We used one fixed prompt and the settings listed above.
- We did not test every model. We tested the models we could run on 2026-10-21. Gemini models are not included because we had no access to them for this run.
- 5 tests per model is a small sample. The range column shows how much a model's score moved between tests. Treat differences of a few points as noise.
Data and reuse
Download the per-test results as CSV: ai-vs-human-iq-2026.csv. One row per model per test, with the seed, number correct, estimated IQ and the model's section accuracies. The data is licensed CC BY 4.0: reuse it freely with credit to IQ Octopus and a link to this page. The live version of the table is the AI leaderboard.
How to cite
Suggested citation:
IQ Octopus (2026). AI vs human IQ 2026: how AI models score on a culture-fair IQ test. Sourdough AI LLC. https://iqoctopus.com/research/ai-vs-human-iq-2026BibTeX:
@misc{iqoctopus2026ai,
author = {{IQ Octopus}},
title = {AI vs human IQ 2026: how AI models score on a culture-fair IQ test},
year = {2026},
publisher = {Sourdough AI LLC},
url = {https://iqoctopus.com/research/ai-vs-human-iq-2026},
note = {Data: https://iqoctopus.com/research/ai-vs-human-iq-2026.csv, CC BY 4.0}
}Journalists: the press kit has key facts and logos, and questions go to support@iqoctopus.com.
Versions
- October 21, 2026: first publication. 5 models, 5 tests each.