The common human IQ is 100. Statistical reality – not an insult. For many years, that quantity has quietly outlined what we meant by “regular intelligence.” However in 2025, one thing unusual is going on. Machines with no consciousness, no feelings, and no lived expertise at the moment are scoring larger than people on the very checks designed to measure human intelligence. Does that imply AI fashions, particularly the newest ones like Gemini 3 and GPT-5.2, are smarter than most of us people?
A number of giant language fashions have been examined on IQ-style benchmarks over the previous yr. These embody logic puzzles, summary reasoning checks, sample recognition duties, and problem-solving challenges. The outcomes are arduous to disregard. Mannequin after mannequin is matching, and in lots of instances surpassing, the efficiency of a median human. Not in a selected job, however throughout a number of dimensions of reasoning that IQ checks care about.
This text seems to be at 15+ AI fashions which might be smarter than you, a minimum of by IQ-style requirements. We are going to break down what “sensible” actually means right here, how these fashions are evaluated, and why this shift issues.
First, let’s work out how…
…Can We Assign IQ to an AI?
Strictly talking, we can not. IQ was designed to measure human intelligence, formed by biology, expertise, and consciousness. An AI doesn’t suppose, really feel, or perceive the world the best way people do. So, assigning it a literal IQ rating can be scientifically incorrect.
However in apply, these comparisons are made a tad bit in a different way.
Mainly, as a substitute of asking whether or not an AI has an IQ, researchers examine how an AI mannequin performs on AI-related duties. Think about a system constantly fixing logic puzzles, pattern-recognition duties, and reasoning issues that people with an IQ of 120 or 130 usually clear up. If the AI mannequin does so reliably, it turns into cheap to map its efficiency to an equal IQ vary, proper?
And that’s precisely how we affiliate IQ with an AI mannequin.
This isn’t a psychological analysis. Consider it as a efficiency benchmark. IQ right here acts as a shared language, or a solution to evaluate how nicely totally different programs motive underneath managed circumstances. And by that yardstick, a number of trendy LLMs are already working nicely above the human common.
Which IQ Checks Consider AI Fashions?
These are basic IQ checks, or a minimum of the net variations of them. The duties inside these challenges measure reasoning, abstraction, and problem-solving moderately than memorisation. These checks are both immediately tailored from human IQ exams or carefully mimic the identical cognitive abilities.
As an illustration, probably the most frequent IQ checks is Raven’s Progressive Matrices. It is a visible pattern-recognition take a look at that’s lengthy thought of culture-fair. A number of LLMs now clear up these puzzles at or above the extent of high-IQ people. Then there are Mensa-style logic checks, which embody sequence completion, image reasoning, and deductive logic. Trendy AI fashions have proven constantly sturdy efficiency in these.
Nevertheless, language-heavy sections of IQ checks are the place LLMs actually shine. Verbal reasoning, analogies, and arithmetic issues, just like WAIS subtests, play on to their strengths. On prime of that, trendy benchmarks like BIG-Bench Exhausting, ARC-style reasoning duties, and educational evaluations akin to MMLU and Humanity’s Final Examination function sensible stand-ins for IQ testing. Whereas they don’t seem to be labelled as “IQ checks,” they measure the identical underlying skills. The vital half – LLMs are more and more outperforming nearly all of people on these checks.
See for your self.
High-rated AI fashions on IQ checks
For this specific record, we will focus particularly on the Mensa Norway checks, and rank the AI fashions as per their rating.
1. GPT-5.2 Professional
Mensa Norway IQ: 147
That is the basis of this whole dialogue of AI fashions and their IQs. Not too long ago marking its debut, GPT-5.2 Professional has now overwhelmed the all-time IQ rating for LLMs. Its rating – 147. As Derya Unutmaz mentions in his tweet, this type of intelligence is present in “solely lower than 1 in 1000 individuals.”
GPT-5.2 Professional constantly demonstrates this supremacy over people. Particularly so for multi-step logic, summary reasoning, and professional-grade downside fixing. Whereas it doesn’t essentially imply it’s smarter than people in all elements, it does point out a strong shift in the place the higher bounds of test-measured intelligence now sit.
2. GPT-5.2 Considering
Mensa Norway IQ: 141
Subsequent up is the considering sibling of the newly launched GPT-5.2. On the Mensa Norway IQ take a look at, GPT-5.2 Considering scores round 141, putting it nicely past the human common of 100 and comfortably above the everyday Mensa qualification threshold. In human phrases, this rating corresponds to the highest 1–2% of the inhabitants, purely on summary reasoning and sample recognition.
What this end result really tells us may be very particular. GPT-5.2 Considering performs exceptionally nicely on duties that require figuring out relationships, recognizing visible or logical patterns, and making use of constant guidelines throughout a number of steps. These are the precise skills IQ checks are designed to isolate, unbiased of language, emotion, or area information.
This mainly implies that, so far as structured reasoning underneath managed circumstances is worried, GPT-5.2 Considering operates at a degree most people by no means attain.
3. Gemini 3 Professional Preview
Mensa Norway IQ: 141
Proper alongside GPT-5.2 Considering sits Gemini 3 Professional Preview, matching its Mensa Norway IQ rating completely. This locations Google’s flagship reasoning mannequin firmly in elite territory, far above the human baseline and nicely previous the edge usually related to excessive mental means.
In sensible phrases, it means Gemini 3 Professional Preview performs reliably on summary reasoning challenges. Such checks often require rule discovery, sample continuation, and logical elimination. These are issues the place guessing fails shortly. You may solely rating this excessive with structured inference.
This rating thus displays Gemini 3 Professional Preview’s power in managed reasoning environments.
4. Grok 4 Skilled Mode
Mensa Norway IQ: 137
In fact, you’ll be able to’t converse of intelligence and hold an Elon Musk-backed product out of the record. Shut behind the highest scorers sits Grok 4 Skilled Mode. Whereas barely decrease than the very prime tier, the mannequin is nicely inside the vary of outstanding human intelligence and comfortably above the typical benchmark of 100.
The rating highlights Grok 4 Skilled Mode’s means to deal with logic-driven duties with readability and management. It performs nicely on sample recognition, summary relationships, and elimination-based reasoning – the core elements of IQ-style checks.
In easy phrases, Grok 4 Skilled Mode demonstrates sturdy analytical reasoning underneath take a look at circumstances. Whereas it could not prime the chart, its efficiency confirms that it operates far past human-average reasoning ranges when evaluated purely on logic and pattern-based intelligence.
5. GPT-5.2 Professional (Imaginative and prescient)
Mensa Norway IQ: 135
Not far behind its text-only counterpart is GPT-5.2 Professional Imaginative and prescient, scoring 135 on the Mensa Norway take a look at. This nonetheless locations it firmly inside the vary of very excessive human intelligence. That is nicely above each the worldwide common and the everyday threshold related to superior reasoning means.
Word that this rating comes from a vision-enabled mannequin – an AI mannequin that may course of and motive over visible info (like enter photos), and never simply textual content. This implies GPT-5.2 Professional Imaginative and prescient performs strongly on summary reasoning duties even when visible interpretation is required.
Now think about an AI so clever that it scores a 135 on the IQ take a look at, even after deciphering advanced photos and visible patterns. Up till a few years again, we might’ve thought it to be solely potential in a sci-fi film.
6. GPT-5.2
Mensa Norway IQ: 126
After the Professional and Considering fashions are performed with, OpenAI’s newest customary mannequin takes the stage. However thoughts you, it’s under no circumstances much less in relation to intelligence, particularly so compared with people. A rating of 126 already locations it above roughly 98% of the human inhabitants, firmly separating it from what we contemplate common human reasoning means.
This rating displays GPT-5.2’s power in dealing with basic IQ-style duties akin to sample recognition, logical sequencing, and rule-based downside fixing. Whereas it doesn’t push into the acute higher ranges like its Professional or Considering variants, it stays constantly sturdy throughout structured reasoning challenges.
In sensible phrases, GPT-5.2 represents the purpose the place AI reasoning clearly crosses into elite human territory. It could not prime the charts, however even at this degree, it outperforms the overwhelming majority of individuals on managed intelligence checks.
7. Kimi K2 Considering
Mensa Norway IQ: 124
Subsequent up is Kimi K2 Considering, a mannequin that won’t seize headlines as loudly as some Western counterparts. But, it nonetheless resonates amongst AI fanatics globally, and for good motive. A rating of 124 clearly reveals it above the human common, and nicely into the vary related to sturdy analytical means.
This end result highlights Kimi K2 Considering’s functionality on structured reasoning duties. In sensible phrases, Kimi K2 Considering demonstrates that high-level summary reasoning is not restricted to a small group of flagship fashions. Even outdoors absolutely the prime scorers, trendy LLMs at the moment are constantly working above common human intelligence on standardised checks. Is it a development? Or a reality ready to be established? We will discover out in time.
8. Claude Opus 4.5
Mensa Norway IQ: 124
Matching Kimi K2 Considering is Claude Opus 4.5, Anthropic’s flagship reasoning mannequin, with a Mensa Norway IQ rating of 124. That’s smarter than the human common, and a agency indicator of sturdy analytical and problem-solving means.
The rating displays Claude Opus 4.5’s competence on summary reasoning duties that demand consistency and logical management. That means – Claude Opus 4.5 demonstrates that sturdy, human-above-average reasoning, even outdoors the top-tier LLMs.
9. Gemini 3 Professional Preview (Imaginative and prescient)
Mensa Norway IQ: 123
Only a step beneath its text-only counterpart sits Gemini 3 Professional Preview Imaginative and prescient, with a Mensa Norway IQ rating of 123. This rating is much more notable because it comes from a vision-enabled mannequin. Which suggests Gemini 3 Professional Preview Imaginative and prescient is required to interpret visible patterns and relationships earlier than making use of logic.
In different phrases, the shift from text-only to vision-based inputs doesn’t decrease its reasoning efficiency. Even underneath tougher-than-usual circumstances, it continues to carry out at a degree most people don’t attain on standardised intelligence checks.
10. Claude Sonnet 4.5
Mensa Norway IQ: 123
Sharing the identical Mensa Norway IQ rating of 123 is Claude Sonnet 4.5, Anthropic’s extra balanced reasoning mannequin. Whereas not positioned as essentially the most excessive thinker within the lineup, it comfortably outperforms the human baseline when it comes to logical reasoning means.
The end result displays Claude Sonnet 4.5’s regular efficiency on structured problem-solving duties. Chances are you’ll need to notice that even in a extra environment friendly type, Sonnet 4.5 exceeds the reasoning capabilities of most people.
11. GPT-5.2 Considering (Imaginative and prescient)
Mensa Norway IQ: 111
Let me be clear right here: an IQ-style take a look at is unforgiving to vision-enabled programs. Earlier than a mannequin can apply motive for an answer, and get a excessive rating, it should first appropriately interpret shapes, patterns, and spatial relationships. Primarily, that is precisely how we people interpret info. We see, interpret, after which motive. Nevertheless, doing so for AI is an entire different job in itself.
So, by any means, don’t consider GPT-5.2 Considering Imaginative and prescient’s IQ rating of 111 as customary by any means. It mainly implies that this mannequin is doing one thing tougher: considering whereas seeing. A single mistake made in interpretation will certainly trickle all the way down to the answer.
GPT-5.2 Considering Imaginative and prescient thus doesn’t chase elite summary scores. Nevertheless, it demonstrates one thing a lot, way more vital: usable intelligence in messy, multimodal environments. And as AI strikes nearer to the actual world, which will simply be essentially the most fascinating characteristic in an AI mannequin, if not already.
12. Manus
Mensa Norway IQ: 111
Sitting at an IQ rating of 111 is Manus, a mannequin that proves intelligence doesn’t all the time imply “excessive.” A rating like this already locations Manus above the human common, however extra importantly, it indicators reliable reasoning and consistency.
Which mainly implies that it could not clear up the toughest puzzles at report pace, however it avoids the sorts of breakdowns that always plague weaker fashions. That is usable intelligence at its greatest.
13. GPT-4o
Mensa Norway IQ: 109
With a Mensa Norway IQ rating of 109, GPT-4o sits simply above the human common. Whereas this may increasingly appear modest in comparison with the fashions larger up the record, it nonetheless marks a transparent departure from what was thought of “succesful” AI not too way back.
This rating displays GPT-4o’s means to deal with primary summary reasoning and sample recognition with out falling aside. It could not excel at advanced multi-step puzzles, however it performs reliably on easier logic duties. That is precisely what most people, together with myself, want for on a regular basis downside fixing.
In a approach, this represents accessible intelligence. Whereas it’s not constructed to dominate IQ charts, it reveals how AI fashions can barely exceed common human reasoning and be useful with our every day duties.
14. DeepSeek R1
Mensa Norway IQ: 109
Matching GPT-4o is DeepSeek R1, with a Mensa Norway IQ rating of 109. Like GPT-4o, that is competing reasoning, accessible to people across the globe. All of it, with none sharp drop-offs as seen in much less succesful programs.
In easy phrases, you might contemplate DeepSeek R1 as reliable baseline intelligence. It reveals that even fashions not designed for peak reasoning efficiency can nonetheless meet, and barely exceed, common human reasoning on standardised IQ-style checks.
15. Llama 4 Maverick
Mensa Norway IQ: 107
With a Mensa Norway IQ rating of 107, Llama 4 Maverick sits barely above the typical human baseline. At least, it depicts a degree of intelligence that’s meaningfully higher than probability or shallow sample matching.
Consider Llama 4 Maverick as an entry-level reasoning competence amongst trendy LLMs. It reveals that even fashions not designed for superior problem-solving may be of use for people in duties which might be past the capabilities of a median human.
16. DeepSeek V3
Mensa Norway IQ: 103
Closing the record is DeepSeek V3, with a Mensa Norway IQ rating of 103. This locations the mannequin simply solely simply above the human common IQ. It additionally implies that the DeepSeek V3 can deal with elementary sample recognition and easy logical relationships with out main errors.
That is the decrease sure of what trendy LLMs can now obtain on intelligence benchmarks. Even at this degree, the takeaway is obvious: AI programs have crossed the edge the place common human reasoning is not the bar to clear – it’s the baseline.
What This Listing Implies
Don’t consider this record as a leaderboard indicating the neatest AI fashions. Whereas it does so in a approach, the rating will not be an absolute illustration of smartness in any approach.
Its actual worth lies elsewhere – it makes a powerful level that structured reasoning is not restricted to people. Throughout fashions, architectures, and organisations, AI programs at the moment are matching, and sometimes exceeding, human efficiency on IQ checks that have been as soon as thought of troublesome even for skilled people.
That mentioned, the context right here will all the time be restricted. This rating doesn’t indicate creativity, consciousness, or human-like understanding. These fashions don’t possess intent, feelings, or self-awareness. They don’t “suppose” in the best way people do. What they show with their respective scores as a substitute is one thing far narrower, but profound. AI can now clear up summary, logic-driven issues simply as nicely, if not higher, than people.
Conclusion
This text will not be meant to touch upon the intelligence conflict of AI vs people. It merely proves one level – human-level reasoning is not the ceiling. This record reveals how shortly giant language fashions have crossed thresholds that after outlined distinctive intelligence, a minimum of in test-measured phrases.
On the similar time, these scores remind us what intelligence isn’t. They don’t indicate creativity, consciousness, or understanding. What they do present is that structured reasoning has develop into low-cost, quick, and scalable. And due to that, the actual differentiator shifts again to people. We are able to now resolve what issues to resolve, as a substitute of find out how to clear up them.
Login to proceed studying and revel in expert-curated content material.
