Honeybees can count to 4, and grasp zero, study finds

Honeybees can count up to 4 and recognize zero as a quantity lower than one, according to a body of research that has been building over the past several years and continues to attract scientific attention. The findings position bees among a select group of non-human animals demonstrated to possess basic numerical cognition.

What the research shows

Studies conducted by researchers at RMIT University in Melbourne trained honeybees ( Apis mellifera ) to associate visual stimuli carrying different numbers of elements with sugar rewards or bitter quinine solutions. Bees consistently chose images bearing fewer elements when rewarded for doing so, and they extended this behavior to novel stimuli — indicating they were tracking quantity, not memorizing specific images.

The zero result was particularly notable. When bees were shown a blank stimulus alongside one bearing a single element, they treated the blank as numerically lesser, placing it correctly at the low end of a number line. This behavior mirrors responses seen in primates and young human children before formal mathematical instruction .

Counting up to 4 aligns with what cognitive scientists call the "subitizing" range — the quantity threshold within which humans and many animals can perceive number instantly, without serial counting. Beyond 4, bees appear to lose precision, which also mirrors human perceptual limits under subitizing conditions.

Why the insect brain matters here

A honeybee brain contains roughly 1 million neurons. The human brain contains approximately 86 billion. That a structure six orders of magnitude smaller can support numerical discrimination is the finding that draws sustained scientific interest, because it implies numerical cognition does not require a large or structurally complex brain.

Researchers have proposed that bees use their mushroom bodies — paired lobes in the insect brain associated with learning and memory — to process and store abstract relationships between quantities. The mechanism is not yet fully mapped, but the behavioral evidence is consistent across multiple experimental replications.

This has direct implications for understanding how numerical cognition evolved. If distantly related species — vertebrates and insects separated by roughly 600 million years of evolution — independently developed the ability to discriminate small quantities, it suggests numerical processing confers a survival advantage substantial enough to appear across radically different neural architectures. Foraging efficiency is the leading candidate: a bee that can assess which flower patch holds more blooms gains a measurable caloric advantage.

Technical constraints and open questions

The experimental designs used in bee counting studies rely on operant conditioning with food rewards, which introduces a methodological limitation: researchers cannot fully rule out that bees are responding to low-level perceptual features — total surface area, spatial density, or edge count — rather than discrete numerosity. Multiple studies have controlled for these variables by varying element size, shape, and arrangement, but the debate over what bees are truly computing remains active in the literature.

The upper bound of 4 also needs qualification. Bees can be trained to respond to larger quantities in some experimental formats, but their error rates increase sharply above 4, and transfer to novel stimuli becomes unreliable. Whether this reflects a hard cognitive ceiling or a training artifact is unresolved.

Engineers working on bee-inspired navigation chips for insect-sized robots have drawn on bee spatial cognition for years; the numerical cognition data adds another layer to what minimal neural hardware can theoretically support, with potential relevance to ultra-low-power edge AI design.

If a minimal counting circuit can be characterized in Apis mellifera , it may offer a blueprint compact enough to implement in neuromorphic hardware — a direction that would carry the science well beyond apiology.

This study was first published in the journal Science and Proceedings of the Royal Society B .