Coilr report·Player counts·
Three things move a Steam game, and Counter-Strike got none of them
Coilr records how many people are inside a game every few minutes. Between May 25 and August 23 that is 1,890,595 readings, which is enough to ask what a summer actually does to a player count — not to one game, but to all of them at once.
Comparing four weeks in early summer against four weeks in late August, across the 70 games big enough to measure: the middle of the pack barely moved, at +3.4%. 40 games rose and 30 fell. But the spread around that middle runs from +1079% to -89.5%, and every large move on the board has a cause. There turn out to be only three kinds.
An update
The biggest moves of the summer were all patches, and their lift holds.
36,427 → 292,610
Palworld left early access on 10 July. It had been sitting near 36,000 players a day for a month; the next day it was at 292,610, and two days after that 644,290.
24,016 → 41,746
Project Zomboid shipped Build 42.20 to stable on 29 July, eleven years into early access. 24,016 the day before, 41,746 that day, 92,110 by 2 August.
29,017 → 70,606
HELLDIVERS 2 landed the Devoid of Liberty update and a Warhammer 40,000 warbond on the same day, 12 August. The count went from 29,017 to 70,606 in twenty-four hours.
A price cut
A deep enough discount moves a game like a patch does — and part of the lift survives the price going back up.
19,961 → 51,463
The Binding of Isaac: Rebirth fell from $14.99 to $1.49 on 25 June. The next day its player count went from 19,961 to 51,463, and it kept climbing to over 92,000.
23,844 → 34,362
Cyberpunk 2077 went from $59.99 to $17.99 on 19 June. Same-day: 23,844 to 34,362, and 51,678 two days later.
A season
The third kind has no day attached to it at all. Path of Exile 2 is the steepest fall on the board at -89.5% — from 200,245 players to 21,026 — and it did not happen on any particular date. It is a smooth slide from the first day of the window to the last, the shape of a season that began before we started watching and emptied out on schedule. Dota 2 is the same shape pointing the other way: no single day, no announcement in the middle of it, just a steady climb of +34.8% across the summer.
Games on a season are the ones whose charts are easiest to misread. A snapshot taken in the first week would call Path of Exile 2 one of the biggest games on Steam; a snapshot taken in the last would barely rank it.
The board
Every game with enough coverage in both windows, biggest riser first. The figures are mean concurrent players, not peaks — see the method below for why that distinction decides whether this table is real.
And then there is Counter-Strike
Counter-Strike 2 fell 11.0% over the same summer, from 931,003 concurrent players to 828,387. As a percentage that is unremarkable — 14 games on this board fell harder, most of them games nobody wrote about. As a number of people it is the largest loss here by a distance: 102,616 players, more than 62 of the other 69 games on this page have in total.
And none of the three engines is available to it. It cannot be discounted — Steam lists it as free, so there is no price to cut. It was not on a season. It shipped eleven updates during the window we watched, roughly one a fortnight, and not one of them shows up in the line: no day in thirteen weeks moves it by more than a normal weekend. What is left is drift.
The flattest week on the board
Sunday is Counter-Strike’s busiest day and Tuesday its quietest, and the gap between them is 10.9%. For most of Steam the weekend is a different order of magnitude from a Tuesday. Nobody plays this game as a weekend treat; they play it the way people go to the gym.
Its day is not flat at all. Between 14:00 UTC and 04:00 UTC the count swings by 130% — 1,240,945 down to 540,004. The week is a habit; the day is a timezone.
Method
Coilr polls Steam’s concurrent-player endpoint for a panel of games on a rolling schedule and keeps every reading. This report compares two four-week windows ten weeks apart — May 25 to June 21, against July 27 to August 23 — using the 1,890,595 readings taken between them. A game is on the board if it has at least 400 readings in the first window and 500 in the second, averaged at least 5,000 players in the first, and reached 20,000 in one of the two. That last clause is deliberate: a game that grew into significance over the summer belongs in a report about what moved.
Means, never peaks. Our sampling rate has changed twice — roughly daily in April, hourly from late May, every ten minutes from late July. A daily peak rises with sampling density on its own, because more samples means a better chance of catching the real high point, so a game would appear to grow purely because we started looking more often. A mean over a window with complete 24-hour coverage is unaffected. It is also why these windows start in late May and not in April, when an entire day could be represented by a single reading taken at four in the morning.
Blank readings are dropped. Steam’s endpoint answers 0 rather than failing about 9% of the time. For a game averaging tens of thousands of players that is plainly a hiccup and not an empty server, and leaving them in would drag a mean down by however often Steam happened to stumble that week.
The causes are not inferred from the player counts. Each was checked against a separate record: Coilr’s own price history for the two discounts, both of which land on the day the line moves, and Steam’s news feed for the three updates, where the studios’ own announcements carry the same date. Where no such record exists — the seasons — no cause is claimed.
Player counts are concurrent players, not unique players, and Steam reports them per game rather than per account. No individual account appears in any figure on this page.
Figures free to quote with attribution to Coilr. The live curve behind every row is on its own game page, and the weekly shape this report leans on is the subject of a separate report.