A bee understands zero. A nutcracker remembers thousands of buried seeds. Neither could do the other’s job, and no ranking survives that.
There is no ranking, and the reason is not diplomacy. Intelligence is not one quantity: an animal can be extraordinary at remembering thousands of cached locations and unable to solve a problem a bee solves. The useful question is which problems a species is built to solve, and the answers are far stranger than any league table.
Ask which animal is cleverest and the honest answer is that the question does not have the shape it appears to have. Cognitive abilities do not come as a package. A Clark’s nutcracker recovers thousands of separately cached seeds months later, which no ape can do. A honey bee learns to treat "nothing" as a quantity smaller than one, a concept human children take years to reach. An octopus opens a screw-top jar with no social life to have learned it from and a nervous system arranged unlike any vertebrate’s. Each of these is remarkable on one dimension and unremarkable on others, and any ranking is really a choice about which dimension to privilege — usually the one humans are best at. What replaces the ranking is a set of dimensions that can be measured separately: memory, causal inference, planning for a future state, self-control, numerical discrimination, social cognition, navigation, and the capacity to learn from others. Ecology predicts the answers well. Species that cache food have exceptional spatial memory because that is the problem their lives set them. Species living in large fluid social groups tend to be good at tracking relationships. Species that must extract hidden food tend to be good at tools. The most useful reframing of the question is that intelligence is not a height an animal reaches but a shape it has, and the shapes are worth knowing individually.
Developed coverage · 53% complete · reviewed 2026-08-31
What this page covers
Cognitive testing is heavily concentrated in a few groups — corvids, apes, cetaceans, parrots, honey bees and octopuses — which is a fact about researchers rather than about animals.
Often confused with: Brain size; Trainability
Quick facts
Is there a ranking?
No — the abilities are separable, and ecology predicts them
Understanding zero
Demonstrated in honey bees
Brain size
A poor predictor on its own; a bee has about a million neurons
Not because it would be unkind. Because the abilities come apart.
Every ranking of animal intelligence is a ranking on one dimension with the others quietly dropped. Choose spatial memory and a nutcracker beats a chimpanzee. Choose tool manufacture and a New Caledonian crow beats almost everything. Choose numerical concepts and a honey bee with a million neurons is in the conversation. These are not near-misses on a single scale; they are different abilities, dissociable in the same animal, and an individual can be outstanding at one and ordinary at another.
Brain size is the usual proxy, and it is a bad one. It correlates with body size before it correlates with anything interesting, the corrections for that are contested, and the clearest counterexamples are not marginal: an insect brain of about a million neurons supports numerical concepts, symbolic-scale learning and the ability to acquire a communicative display by watching. What predicts cognitive ability far better than any measure of the brain is the problem the animal has to solve to stay alive.
Diagram
Eight abilities, four animals
Each animal is exceptional on some dimensions and unremarkable on others. Any ranking is a decision about which column to read.
The same explanation in words
A grid of four animals against eight cognitive abilities. The New Caledonian crow is exceptional at tool manufacture and causal inference, ordinary at numerical tasks. The Clark’s nutcracker is exceptional at spatial memory and unexceptional at tool use. The honey bee is exceptional at numerical discrimination and navigation and has no tool use at all. The bottlenose dolphin is exceptional at social cognition and vocal learning and has minimal tool use outside one population. No row is best on every column.
Choose the array with fewer or more items, including none
Honey bee — treats zero as a quantity
Causal inference
Solve a problem requiring an understanding of what makes what happen
New Caledonian crow — bends a wire to make a hook
Planning
Select a tool now that is only useful later
Raven — chooses a tool for a task hours away
Social cognition
Track who did what to whom, and who is watching
Raven — caches differently when a competitor can see
Vocal learning
Acquire a novel signal by hearing it
Bottlenose dolphin — invents and learns whistles
Navigation
Return to a goal from an unfamiliar release point
Honey bee — integrates a vector and reads polarised light
Learning from others
Acquire a behaviour by watching another animal
Great tit — and then conforms to the local version
The pattern in that table is ecological rather than taxonomic. Caching species have the spatial memory; species that extract hidden food have the tools; species in large fluid groups have the social tracking. Intelligence looks less like a ladder and more like a set of tools each lineage assembled for the job in front of it.
The best predictor of what an animal can do is the problem it has to solve.
Once the abilities are separated, a pattern appears that a ranking hides completely: they track ecology. Species that cache food have extraordinary spatial memory, because a nutcracker that cannot find its own seeds in February dies. Species that extract food from inside things — under bark, inside nuts, down holes — are the ones that use and make tools. Species living in large, shifting social groups are the ones that track who did what to whom.
Animals that need better memory grow the brain region for it — within a single lifetime
Well supported
Good evidence backs this, though some details remain open.
Food-caching chickadees from harsher environments, and birds held on restricted food, show larger hippocampi with more neurons and better spatial memory performance than conspecifics facing lower demand.
Who this applies to
black-capped chickadees across populations and food regimes
Studied in
Poecile atricapillus
Why we rate it this way, and what the caveats are
Well supportedModerate confidence
The population comparison is correlational, but the food-restriction manipulation supplies an experimental component. Confidence is held at moderate because hippocampal volume is a coarse proxy for memory capacity.
How far it can be extended
Hippocampal responsiveness to spatial demand has been reported across several food-caching bird families; it was measured here in one species.
Caveats
Volume is a blunt measure; more neurons is not automatically more memory.
Population differences confound genetics with experience, which the food manipulation only partly separates.
Caching strategies differ between species, so this does not transfer uniformly.
Still unanswered
Does the same plasticity operate in non-caching animals facing spatial demand?
Pravosudov and Clayton, 2002 · Behavioral Neuroscience
Hippocampal size and spatial performance tracking environmental demand.
That reframing does more work than it first appears to. It explains why a bee with about a million neurons handles a numerical concept that took human mathematics until the seventh century to write down: a forager comparing patches has to compare quantities, and evolution equipped it to. It explains why an octopus, with no social life to learn from and a nervous system arranged unlike any vertebrate’s, is nonetheless good at manipulating objects — it hunts by feeling into crevices. And it explains why the animals people find most impressive are so often the ones whose problems resemble ours.
Many animals can tell "more" from "less" — far fewer can actually count
Well supported
Good evidence backs this, though some details remain open.
Discrimination of approximate quantity is documented across insects, birds and mammals, while precise numerical operations — ordering, and treating an empty set as a quantity — have been demonstrated in a much smaller set of species under trained conditions.
Who this applies to
the species in which numerical tasks have been directly tested
Approximate magnitude discrimination is robust and replicated across very distant lineages. The stronger claims rest on fewer species and heavily trained individuals, and the claim is worded to keep the two apart.
How far it can be extended
Built only from species with their own primary study. The distinction between approximate magnitude and precise number is drawn from what each task actually required.
Caveats
Most "counting" results are magnitude discrimination, which is a genuinely different ability.
Several headline findings come from small numbers of intensively trained individuals.
Controlling for surface area, density and spacing is a persistent methodological difficulty in this field.
Still unanswered
Does approximate number sense share a mechanism across lineages, or has it arisen repeatedly?
Which animals, if any, use precise number outside a training regime?
An early result often cited as counting that is better explained by sequential landmark matching.
What follows is a warning about tests. A species failing a task may be telling you about the species, or about the task. Apes fail some memory problems a scrub jay finds trivial, not because apes are worse at remembering but because they were never under selection to recall ten thousand locations. Almost every claim that one animal is smarter than another is a claim about which test was chosen.
Memory is not one thing animals have more or less of — it is several separate systems
Established
Specialists would state this without hedging. Multiple independent lines of evidence agree.
Distinguishable memory systems with different durations, capacities and mechanisms have been demonstrated within single species: honey bees show short-, mid- and long-term phases separable by protein-synthesis dependence, while spatial, associative and social recognition memory dissociate across tasks in birds and mammals.
Who this applies to
the species in which memory systems have been experimentally dissociated
Memory phases in bees are separable pharmacologically, and spatial memory dissociates from other kinds by lesion and by task across birds and mammals. The finding is old, replicated and mechanistically grounded.
How far it can be extended
Assembled from primary work in each species. The claim is about the architecture of memory rather than about any one animal, so its strength comes from the same organisation appearing in brains of very different sizes.
Caveats
Ranking species by "memory" is close to meaningless — the question is always which memory, for what.
Phase boundaries are defined by the interventions that disrupt them and are not sharp.
Most of this work uses narrow laboratory tasks.
Still unanswered
How far do the memory phases described in insects map onto vertebrate systems?
How complete this page is, and what it is still missing
NatureHQ publishes its own gaps. This page is at 53% completeness against what we would call a finished subject, and was last reviewed on 2026-08-31. It carries 9 claims and answers 0 mapped search questions.
1 high-priority search intent(s) not yet covered
no research from the last few years is attached — check for newer work
more experiments could be explained in plain English
no popular claim about this subject has been checked yet
Metacognition — whether an animal can represent its own uncertainty — is a substantial literature and is not covered.
Self-recognition and the mirror test are treated elsewhere and not integrated here.
Nothing here addresses consciousness or sentience, which is a separate question requiring different evidence.
Fish and reptile cognition are almost absent, reflecting where the research is rather than where the ability is.