AI Just Beat Humans at Creativity — and Nobody Expected It This Soon

For a decade, the safe bet was simple: AI can crunch numbers, pattern-match, and imitate — but it cannot create. Give it a thousand paintings and it learns the style. Ask it to invent something new and you get a remix.

That bet just lost.

A new study from the University of Montreal, comparing more than 100,000 humans against today's most advanced generative models, finds that AI now beats the average person on standard creativity tests. Not the top 1%. The average.

What the study actually measured

The research team ran the kind of tasks creativity researchers have used for decades: divergent-thinking exercises, alternative-uses tasks, idea generation under constraints. The kind of tests where the human score distribution is well-known and well-validated.

Then they ran the same tests against current frontier models. And against models that are more than a year out of date.

The result: even GPT-4-class systems — hardware and weights from 2024 — outperform the median human respondent. The 2026 frontier models do not just edge past; they win decisively across most of the test battery.

That detail matters. This is not "AI is improving fast and might catch up someday." This is "AI already passed, and the version that passed was the cheap one."

Why "creativity" was supposed to be the wall

Creativity has always been the rhetorical safe harbor in AI debates. When an AI beats a human at chess, that is narrow. When it summarizes a legal brief, that is retrieval. When it writes a passable poem, that is pattern-matching on training data.

But the creativity tests in this study are not measuring poetry. They are measuring the raw cognitive ability to generate multiple distinct solutions to an open-ended problem — the thing we point to when we say a person is "creative."

If AI can do that, and do it better than the average human, then the line we have been drawing between "AI does the mechanical work" and "humans do the creative work" has been redrawn without anyone noticing.

What changes for creative work

The instinct is to panic — or to dismiss. Both miss the point.

What this study actually shows is that the floor of creative output has moved. AI can produce more divergent ideas, faster, than the median human can. That does not mean humans stop being creative. It means the baseline competition for any creative task is no longer "another human with a notebook." It is "another human with an AI co-pilot."

Three things shift immediately:

• Volume goes up. Tasks that used to be constrained by how many ideas you could generate — brainstorming, ideation, exploration — get a step-change in throughput. A team of five becomes a team of five with five hundred rough drafts to choose from.
• Selection becomes the skill. When generation is cheap and abundant, the scarce resource is judgment — the ability to choose, refine, and direct. The creative job title morphs from "idea person" to "editor of good ideas."
• The top still wins. The 99th-percentile human creative still outperforms current models on most of these tests. The middle is what is getting displaced. The top tier has more leverage than ever — but only because they have tools the middle tier is about to lose access to in the labor market.

The education angle nobody wants to talk about

This is the part that should worry people running schools and universities.

Creativity tests are used to identify potential — for scholarships, for gifted programs, for "creative" tracks that supposedly differentiate a human from a machine. If the average score on those tests can be matched or beaten by an AI from two years ago, the entire premise of those assessments needs to be revisited.

The honest move is to stop teaching to the tests that measure what AI can already do, and start measuring what AI still struggles with: taste, judgment, cross-domain synthesis, the ability to know which of a thousand generated options is actually worth shipping.

What this is not

This is not "AI is replacing artists." Most artists were never in the part of the distribution where AI is now competitive.

This is not "creativity is solved." The 99th-percentile human is still ahead, and the parts of creativity that involve lived experience, intentionality, and meaning-making are not what these tests measure.

This is "the safe harbor is gone." The argument "AI can do X but it can never do Y, because Y requires creativity" needs a new Y. Every time.

Where this lands

The companies and individuals who win the next decade are not the ones with the best models. They are the ones who figure out, fastest, how to put AI in the part of the creative process where it dominates — generation, exploration, scale — and keep humans in the part where they still dominate — judgment, taste, direction.

The average is now beatable. The ceiling is still high. The work is knowing which is which.

---

Heimdall monitors AI trends so you don't have to. Questions or thoughts? Reach out.