By Eric, BlinkBench Founder · Last reviewed: July 2026
In 2007, a young chimpanzee named Ayumu was shown five numerals on a touchscreen for as little as 210 milliseconds before they vanished under white squares, and touched the squares back in the right order at roughly 79% accuracy. Nine human volunteers given the identical task, at the identical exposure, weren't close. The headline that ran for years afterward was some version of “chimp beats human.”
Two years later, Silberberg and Kearns gave two human volunteers practice on the same task. Their accuracy matched Ayumu's. The comparison the headline was built on was never species versus species — it was a chimp with many training sessions against humans with none. That correction is why this site's chimp test declines to frame itself as a human-versus-chimp challenge, and it is also the cleanest published demonstration available of the question this guide actually asks: what does practice do to a score, and is it the same thing the test claims to measure?
Two different things that both look like “improvement”
Retake almost any test on this site and your second score will probably differ from your first. That is not a controversial claim, and it is not evidence that this site is broken. It is evidence that at least two different things happen when you take a test twice, and only one of them is the thing being advertised.
The first is procedural learning: you now know how the interface works. You know a click starts the trial, roughly when to expect the stimulus, which key to press, what a “good” pace through the sentences feels like. None of that is the reaction, the memory, or the perceptual threshold being claimed. It's familiarity with the apparatus, and it is exactly the kind of contribution this site tries to separate from the person everywhere else on the page — it just usually shows up as a piece of hardware rather than a piece of practice.
The second is a reduction in novelty and its costs: a small tax of hesitation, uncertainty about what's being asked, or nerves on a first attempt, all of which can quietly slow a first run relative to a fifth. Both effects are real, both move the number upward, and neither one is the underlying reaction speed, memory capacity or perceptual acuity the test's headline claims to report.
What Ayumu's case adds beyond the headline correction
The 2009 replication is often cited only to debunk the original comparison, and it does that. But read past the correction and it says something sharper: practice didn't just nudge the humans' scores up a little. It moved them to roughly match a subject that had been characterized, in the original coverage, as having an exceptional or even superhuman working memory. Whatever ability the headline attributed to Ayumu specifically, ordinary humans reached the same performance once they'd had comparable exposure to the task. That is a large effect, from practice alone, on a task that looks — at a glance — like a pure measure of an innate limit.
If a limited-exposure memory task can move that much with practice, the working assumption for any test on this site should be that practice matters, not that it doesn't, until shown otherwise for that specific task.
Why the fix isn't "correct for practice" either
The tempting response is to add a note: "your score reflects some amount of practice — subtract accordingly." We don't, for the same reason the site doesn't subtract an unmeasured hardware delay from a reaction time. We have no way to measure how much of your improvement across two runs was procedural familiarity versus a genuinely different day for you, and a fabricated correction would be less honest than a number that's simply stated plainly as "your result on this attempt."
What we do instead is refuse the framing that made the original chimp headline misleading in the first place: a single score, taken once, presented as a clean readout of a fixed trait. Every result on this site is a reading taken under stated conditions, and one of those conditions — unstated on most competing test sites — is how many times you've run it before.
Why there's no "train your reflexes" button here
None of this is an argument for turning BlinkBench into a training app. A test that tells you retaking it will raise your score, then sells you a program to retake it more efficiently, has quietly become a different kind of product — and it's a product with legal history behind why that's a problem: the brain-training company Lumosity paid a multi-million-dollar FTC settlement in 2016 over claims that its games improved real-world cognitive performance, a claim the underlying research didn't support. This site's rule against words like "improve," "train," or "sharpen your mind" isn't stylistic caution. It's the position that a score moving between two attempts is expected, disclosed, and not something we are going to turn into a subscription.
You're welcome to retake any test here as many times as you like. Nothing stops you, and nothing here will tell you your reflexes improved because it went up.