General

Bumblebee intelligence: how bees solved a chimp-level puzzle with zero training

5 min read

Scientists long assumed real intelligence needed a big brain. A 2026 Science study broke that: bumblebees with brains smaller than a sesame seed solved a chimp-level tool puzzle with zero training. 73% of untrained bees rolled a ball to reach a reward, and some found shortcuts the team never designed. Here is what the test showed, how a million neurons power insect cognition, and why bees now rank beside crows and octopuses in animal intelligence.

Bumblebee intelligence: how bees solved a chimp-level puzzle with zero training

For most of the last century, scientists agreed on one rule: real cleverness needs a big cerebral cortex. A 2026 study on bumblebee intelligence just broke that rule. Bees with brains smaller than a sesame seed solved a puzzle built to test chimpanzees, and they did it without a single training session.

The work, published in Science, is the strongest evidence yet that bumblebee intelligence is not just reflex dressed up as learning. The bees appeared to think the problem through before they acted. Here is what happened, why it matters, and how a creature with under a million neurons earns a place next to crows and apes in any honest discussion of animal intelligence.

A buff-tailed bumblebee (Bombus terrestris) resting, the same species tested in the 2026 study

The classic box-and-banana test, rebuilt for bees

In the 1910s, psychologist Wolfgang Kohler gave chimpanzees a banana hung out of reach and a pile of crates. The chimps paused, looked around, then stacked the crates to climb up. Researchers called this "insight", the sudden grasp of a solution without trial and error. For decades it was a high bar that only large-brained vertebrates were thought to clear.

A team led by behavioural ecologist Olli Loukola, working at the University of Oulu and the University of Turku, rebuilt that test for insects. They swapped the banana for a sugar-water reward inside an artificial flower, and the crates for small yellow balls. The flower sat on a raised platform the bees could not directly reach. To win the reward, a bee had to roll a ball underneath the platform to lift it, then climb the ball. That is a spatial puzzle, not a scent cue. You can read the full setup in the original ScienceDaily summary of the study.

The hard part was ruling out shortcuts. Earlier insect-cognition results got dismissed because the subjects had been pretrained or had watched other bees. Loukola's team used "fully naive" bees with no prior exposure to the balls or the mechanism, so any success had to come from the bee itself.

16 out of 22 bees cracked it on their own

This is where the data gets hard to wave away. Of 22 naive bumblebees, 16 solved the puzzle on a first or early attempt. That is roughly 73%, and the researchers were careful to control for luck. The bees did not thrash around randomly. They paused, inspected the apparatus, then made one targeted move. That pause-then-act pattern is the fingerprint of insight rather than blind conditioning.

73% of untrained bees solved a multi-step tool puzzle on their own. Some even found a shortcut the researchers had not designed.

Then came the detail that surprised even the team. A few bees ignored the intended ball-rolling route and found a quicker way to reach the sugar, what the researchers called "cheating". This is cognitive flexibility, and it is exactly what you would expect if bumblebee intelligence were real rather than rehearsed. An animal running on fixed instinct would repeat the trained movement even when it stopped working. These bees instead held on to the goal (get the sugar) and dropped the method. That gap between goal and method is what separates genuine problem solving from rote response, and it is covered in detail by Scientific American's write-up of the experiment.

Put simply, the bees were not memorising a trick. They were reading the situation and choosing a route to the reward. That single result is why bumblebee intelligence is now hard to dismiss as a fluke.

So if you have ever wondered, are bees smart, this is a serious yes. Not smart in the human sense, but smart in a way that resists the old "reflex machine" label. You can dig into the broader science of learning and recall over on our guide to gamified active recall.

Inside the bee brain: where bumblebee intelligence actually lives

Close-up of a Bombus terrestris bumblebee's head, where roughly one million neurons handle the puzzle solving

The intuitive objection is obvious. How can a brain this small do this much? A human brain carries about 86 billion neurons. A bumblebee carries roughly one million. Yet the bee is solving a task structured like the one Kohler gave apes.

The answer is not size. It is wiring. The bee's heavy lifting happens in paired structures called mushroom bodies, long treated as the seat of insect cognition. Mushroom bodies pull together several streams of information at once: where the ball is, what the legs feel, where the reward sits, and what worked last time. A 25-year review of the field, published in Learning & Memory, lays out how these structures turn into the integration hubs that drive flexible behaviour. The mushroom body review is worth reading directly.

What this tells us is that insect brains are not budget versions of vertebrate brains. They are a separate engineering solution, parallel and compressed, tuned for low energy use. Bumblebee intelligence comes from how efficiently those million neurons are wired, not from raw headcount. The same logic shows up when you study genetics and the blueprints behind neural systems.

This is also why the field of insect cognition has been quietly gaining ground for two decades. Each new test keeps shrinking the gap between what a tiny brain can do and what we once assumed required a cortex.

Are bees smart? How they rank among the most intelligent invertebrate

Place this result next to the others and a pattern appears. Crows bend wire into hooks. Elephants move logs to reach food. Octopuses open jars from the inside. Bumblebees now roll balls to lift platforms. All four solved tool problems, and none of them share a recent ancestor with the others.

The last common ancestor of insects and primates lived more than 600 million years ago. That means this kind of problem solving was not inherited once and passed down. It evolved independently, more than once, through a process called convergent evolution. If you want the longer view, our guide to descent with modification and the theory of evolution covers the mechanics. The short version: when different animals face the same ecological pressure, natural selection keeps arriving at similar mental tools using very different hardware.

That reframes the bee. It is not a lesser primate. It is arguably the most intelligent invertebrate we have now caught in the act of insight, alongside the octopus, and it makes bumblebee intelligence one of the more uncomfortable facts for anyone who ranks minds purely by size. The Bee Conservation Trust has a readable take on how this challenges the big-brain assumption.

What bumblebee intelligence does to our definition of animal intelligence

For decades, insects were filed under "reflex-based machines", a phrase that shaped early robotics and early artificial intelligence. The idea was that a bee's behaviour was a chain of fixed responses triggered by smells and shapes, nothing more. Loukola's result makes that filing hard to defend. You can hear the researchers push back on the reflex-machine framing in this CBC interview about the bumblebee study.

If bees can mentally simulate an outcome before they act, then complex reasoning can grow out of small, decentralised nerve networks. That has two knock-on effects. First, it adds moral weight to conservation, because the creatures we are protecting turn out to be problem solvers, not biological automata. Second, it gives AI researchers a real template for efficiency. A bee does remarkable work on the energy of a sip of nectar, and bumblebee intelligence is now a concrete case study in doing more with less. Neuromorphic chips built on that principle could run on a fraction of the power that today's large models burn through.

The University of Oulu's own announcement of the findings is blunt about the takeaway. Advanced cognition is more common across the animal kingdom than the old hierarchy (primate, then bird, then everything else) allowed.

FAQ

What is bumblebee intelligence? It is the capacity of bumblebees to learn, remember, and solve novel problems, including spontaneous tool use, despite having a brain of about one million neurons. The 2026 study is now a central piece of evidence for it.

Did the bees in the study get trained first? No. The researchers used naive bees with no prior exposure to the balls or the mechanism. The 73% success rate came from untrained subjects, which is exactly what makes this count as insight rather than conditioning.

How does bumblebee intelligence compare to other animals? It sits in the same category as tool use in crows, elephants, and octopuses. The bee is now one of the strongest examples of insight in an invertebrate.

Do bees really use tools? Yes. Rolling a ball under a platform to reach food meets the behavioural definition of tool use, and some bees even found shortcuts the team had not designed.

Why does brain size matter here? It shows that wiring efficiency, not raw neuron count, can support advanced reasoning. A million well-connected neurons can outperform what the size-only theory predicted.

Test what stuck

The cleanest way to learn a topic like this is to answer questions on it before you forget the details. Drop the key facts above into our free quiz playground, generate a quick set of multiple-choice questions, and run an instant test with no signup. Active recall beats rereading every time, and bumblebee intelligence is a fun one to quiz yourself on.

Enjoyed this article?

Join Mind Hustle to discover more learning content and gamified education.

Join Mind Hustle More Articles