Every Chrome user has met the T-Rex: the little dinosaur that appears when the internet drops out. From August 5 to October 31, 2026 we logged every game played on chromedino.com, a browser copy of that game: the score, the duration, every jump and duck, what killed the dino, and the time of day in the player's own time zone. Then we removed the cheaters and looked at what was left. As far as we know it is the largest dataset ever collected on the dinosaur game, and this is what it says about how people play it.
“Google built this game for people with no internet. Ten years and twenty-two million runs later, I can report that it is mostly played by people with excellent internet, between 10 and 3, on a weekday.”
Alex, founder of chromedino.comchromedino.com went online in November 2016. Precise game logging only began in August 2026, so everything below covers 88 days, August 5 to October 31. Ten years of visits at today's rate of games per visit add up to roughly half a billion runs since the site started.
The first cactus appears at 41 to 43 points. A third of all games, 32%, end right there: the player never clears it. The median game scores 74 points and lasts about eight seconds. The average is almost twice as high, 136, because a few very long runs pull it up: one game in ten passes 291, one in a hundred passes 889. Count only the games that clear the first obstacle and the median rises to 136.
Share of all games by final score, in steps of 50. Hover for exact counts.
Per game the numbers are humbling; per player they are kinder. The median player's best score over the three months is 250. Four players in five eventually pass 100 and three in five pass 200. Then the air gets thin: one in five ever reaches 500, 4.9% reach 1,000, 0.4% reach 2,000. Past that the counts stop being percentages. Out of every million players, 177 passed 5,000, 41 passed 10,000 and 12 passed 20,000.
Share of all players whose personal best reached the threshold over the period.
A quarter of all games end within five seconds of the first keypress. Half are over within eight, nine in ten within thirty. Only one game in a hundred lasts longer than 82 seconds. Add it all up and the three months contain 98,600 hours of play, which is eleven years of one person jumping over cacti without a break. Players pressed jump 251 million times, about eleven per game, and ducked 12 million times: the duck button accounts for under 5% of all moves. Counting the game's own odometer, where every point is a metre, the dinosaur ran about 3 million kilometres, which is to the Moon and back almost four times.
Share of games shorter than N seconds. The x axis is logarithmic.
Within a session, the median first game scores 70. The second jumps to 82. From there it slides, game after game, down to 69 by the twentieth attempt. Warming up works exactly once; grinding does not. A typical session is four games and 30 seconds long, and 76% of players came to the site exactly once in three months. At the other end of the scale, 693 people played on twenty or more different days, and the most persistent audience, China, averages 23 games per player, almost twice the global figure of 12.
The last point (20+) includes all later attempts.
“I've run this site for ten years, and the data finally confirmed what I suspected: it isn't the dinosaur that's extinct, it's our patience. A third of all games are over before the first jump.”
Alex, founder of chromedino.comCacti cause 98% of all deaths. Pterodactyls, which start flying in at around 450 points, account for 2%. Look only at games that passed 500 and the picture flips: there, pterodactyls cause 43% of deaths, most of them the low-flying kind that has to be jumped at exactly the right moment. The high ones can be ducked, but as the duck counter shows, hardly anyone does.
Share of deaths by cause within each score band.
Share of all games.
Share of games that reached 500.
We reconstructed the local time of every game from the player's browser time zone. The result is a school-and-office day: play starts climbing at 8, peaks at 11 and again at 2 p.m., and 57% of everything happens between 9 and 4. Evenings after 8 p.m. carry 13% of games; the night, under 3%. Players are also at their best in the morning: the median score at 9 a.m. is 80, by 7 p.m. it has fallen to 67. Weekdays see about 290,000 games a day, weekends 170,000, and weekend players score lower (median 65 against 76). Four games in five are played on a desktop and 63% on Windows, the profile of a school or office PC. Among the sites that send us players are Canvas and Google Classroom, and more than one school proxy.
The busiest day was Tuesday, October 20, with 423,000 games. The busiest moment came the next morning, October 21, with 1,048 people playing at once. It was half past three in the morning in New York and mid-afternoon in Beijing and Jakarta: on this site, Asian school afternoons beat the American daytime.
Share of all games. Grey bars: 9:00 to 15:59, 57% of all games.
Total games over the period, player's local day.
Weekend players score lower.
The weekly rhythm is visible at a glance. The climb through September is the school year starting around the world; the dip in the first week of October is China's Golden Week.
Rankings by country are treacherous. A desktop keyboard scores a median of 81, a phone 52, and a mouse click just 40, because a click does not make the dinosaur jump. India has 32% of its games on phones, the Netherlands 9%, so a plain ranking mostly measures who owns a laptop. We therefore rank countries on desktop-and-keyboard games only, and only where we have at least 10,000 of them, which leaves 79 countries. On that basis Chile, France and Finland share the top (median 99), followed by Puerto Rico (98), Austria (95), and Switzerland, Germany and the Netherlands (94). The United States sits at 92, level with Belgium, Sweden and Lithuania; Japan scores 93, the UK, Italy and Canada 91, India 78, China 73. The gap is visible at the very first cactus: 23% of Chilean desktop games end there, against 30% in China and 40% in Kuwait, the last country in the table.
Desktop and keyboard games only, countries with at least 10,000 such games. The full table of 79 countries is below the chart.
Desktop and keyboard games, countries with at least 10,000 such games. Hover a country for its numbers.
By volume the picture is different again. The United States played 4.3 million games, China 4.1 million, India 2.1 million and Indonesia 1.6 million; 40% of games came from browsers set to American English and 18% from Simplified Chinese. The Chinese audience is also the most persistent one, at 23 games per player against 10 to 12 almost everywhere else.
Same game, very different results.
A mouse click does not make the dinosaur jump.
The game speeds up by a fixed amount every frame, from 6 at the start to a cap of 13, so the highest speed a player reached before dying is a clean measure of where reflexes give out. One game in ten never leaves the starting speed: the player dies before the first acceleration tick, almost always at the first cactus. More than half of all games, 55%, end in the next half-step, at speed 6.5 and a typical score of 54. From there every half-step of speed loses about half of the survivors: 17% of games end at 7, 8% at 7.5, 4% at 8. Only 0.35% of all games, about 79,000, ever reach the cap, which the engine hits at roughly 1,300 points. After that the game stops getting harder and the only opponent left is concentration.
Share of games by the maximum speed reached before the dinosaur died.
If every obstacle were an independent coin toss, the chance of clearing the next one would not depend on how many came before. It does. The first obstacle is cleared in 76% of games; once past it the odds improve fast, 85% for the second, 89% for the third, 93% by the tenth. Then comes a dip: between the fourteenth and the sixteenth obstacle the success rate slips back to 92%, which is where the speed passes 8 and the first pterodactyls appear, around 450 points. Past the twentieth obstacle the curve climbs steadily to 95% and beyond. The players who are still alive by then are simply better, and by the two-hundredth obstacle they clear 99 in 100.
Probability of clearing the next obstacle, given that the player has already cleared n of them.
Beginners press too much. In games that end under 100 points the median player jumps 1.5 times per obstacle cleared; past 250 points it is one jump per obstacle, and past 1,000 it drops below one, because by then players duck under the high pterodactyls instead of jumping over them. Ducking is a late skill: 11% of games under 100 points include a duck, against 59% of games past 1,000, and only 36% of all players ever pressed the down arrow. Among games that passed 500, those with at least one duck ended at a median of 771 points, those without at 645.
Median jumps per obstacle cleared, by final score. Beginners press twice.
Share of games with at least one duck, by final score.
Within a session the second game is the best one. Across days the question is different: does a player who comes back tomorrow start from a higher level? At first sight, yes: the median best score on a player's first day is 239, on the second day 285, on the tenth 448. But the players who come back are not the same people as the players who leave. Follow one fixed group, the 20,345 players who played on at least five different days, and the curve flattens: 369 on day one, 405 on day two, then about 400 for the rest of the week. Most of the apparent learning curve is selection. The real gain from coming back is about ten percent, and it arrives on the second day.
Median of the best score of the day, by how many days the player has played so far.
When do people stop? The chart shows the share of games that turned out to be the last one of the session, split by score and by whether that game set a personal record. The answer is the same at every level: a record is a reason to stop, not to continue. Among games under 100 points, 8% of ordinary games closed the session, against 13% of games that set a record; past 1,000 points, 14% against 35%. Dying at the very first cactus does not drive people away either: 10% of those games ended the session, fewer than the 12% for games that cleared it. Dinosaur players quit on a high, not in a rage.
Share of games that were the last one of the session, by final score and personal record.
The daily rhythm differs by country. In the Netherlands 72% of games happen between 9 and 4, with a peak at 11 a.m.; in China 65%, with a sharp peak at 3 p.m., the last lesson of the day. Turkey peaks at 8 p.m. and plays a quarter of its games after that hour; in India and South Korea one game in five is played after 8 p.m. Where the game is a school-time habit the curve peaks before noon, where it is an evening pastime, after dinner.
Share of each country's games by local hour. Grey: 9:00 to 15:59.
School calendars leave a mark. The Netherlands doubles in the last week of August as schools reopen; Poland triples on September 2, its first day of lessons; China almost doubles on September 1 and then loses two thirds of its games to Golden Week, from 61,000 on September 30 to 18,000 on October 5, before returning at a new high. The United States climbs through August as districts reopen one by one and levels off in October.
Games per day, each country indexed to its own average over the period.
Browsers tell the same story as devices. Edge players score a median of 86, mostly because Edge means a school or office Windows PC with a keyboard; Safari players 62, because half of them are on an iPhone. The most persistent players are a different species. One player in Chile played on every one of the 88 days of the study, 9,555 games in all, with a best of 5,868. The record for sheer volume belongs to a desktop in Kuwait: 42,670 games in 73 days, 133 hours, and a best score of only 1,626. A pattern like that looks more like a script than a person, and we leave it in the table as a reminder that our checks test physics, not intent. The longest single day was 12.5 hours and 138 games, in the United States on October 7.
Browsers with at least 5,000 games. Differences describe audiences, not software.
Longest streaks of consecutive days with at least one game. Anonymous device identifiers, no nicknames.
The dinosaur runs at a known speed, so a score implies a minimum duration, a minimum number of obstacles and at least as many jumps. We check every game against five such rules. In three months 0.08% of games failed, 17,932 in all: 1,313 claimed scores above the game's own cap of 99,999, 1,974 earned points faster than the engine can produce them, 864 ran faster than the maximum speed, 895 passed more obstacles than the distance allows, and 252 passed obstacles without a single jump, the signature of a collision hack. The remaining two thirds were caught by earlier versions of the checks, which did not record which rule fired. The exact thresholds stay unpublished. All flagged games are excluded from everything above. The claimed scores fall into two groups: fantasy numbers such as 10,000,000 or the cap itself, 99,999, and tiny ones, 78 to 80, right after the first cactus, where a tampered game fails the physics check at the very first obstacle. By share of games, Turkey (5.8% flagged), Pakistan (4.4%) and Brazil (3.5%) lead; China, at 0.3%, is the cleanest large audience. The best verified run of the period is 52,972 points, played on a desktop in Serbia on August 16: 46 minutes, 3,616 obstacles, 2,826 jumps. Second and third place were both set in the last week of October, 50,108 in the United States and 49,307 in Poland.
Flagged games by the rule they broke.
Claimed scores among flagged games.