Roblox does not empty out overnight so much as it breathes. Across a balanced panel of 91 games — every one of them with a player reading in all 216 hours of 6 to 14 August 2026 (UTC) — the quietest hours were 05:00 and 06:00 UTC, averaging about 961,000 players, and the busiest was 17:00 UTC at 1,399,158. That is 46% more people in game at the top of the day than at the bottom of it.
The curve below is the average across the nine days, so a single busy Saturday cannot bend it.
The shape of a Roblox day
| Hour (UTC) | Average concurrent players | Share of the day's peak |
|---|---|---|
| 00:00 | 1,207,264 | 86% |
| 01:00 | 1,184,827 | 85% |
| 02:00 | 1,127,931 | 81% |
| 03:00 | 1,058,501 | 76% |
| 04:00 | 999,452 | 71% |
| 05:00 | 961,491 | 69% |
| 06:00 | 960,893 | 69% |
| 07:00 | 989,354 | 71% |
| 08:00 | 1,024,970 | 73% |
| 09:00 | 1,069,497 | 76% |
| 10:00 | 1,115,958 | 80% |
| 11:00 | 1,181,073 | 84% |
| 12:00 | 1,269,210 | 91% |
| 13:00 | 1,345,991 | 96% |
| 14:00 | 1,384,899 | 99% |
| 15:00 | 1,366,715 | 98% |
| 16:00 | 1,363,945 | 97% |
| 17:00 | 1,399,158 | 100% |
| 18:00 | 1,361,182 | 97% |
| 19:00 | 1,359,762 | 97% |
| 20:00 | 1,320,781 | 94% |
| 21:00 | 1,264,467 | 90% |
| 22:00 | 1,223,294 | 87% |
| 23:00 | 1,208,407 | 86% |
The busy stretch is wide rather than sharp: every hour from 13:00 to 20:00 UTC sits at 90% of the peak or above. In local terms, 13:00 UTC is 9am in New York, 6am in Los Angeles and 2pm in London; 20:00 UTC is 4pm, 1pm and 9pm in those three. The floor at 05:00–06:00 UTC is the middle of the American night and the start of the European morning, which is the only stretch of the day when no large audience is awake and out of school.
The peak hour is not the same for everyone
Weight the panel by players, as the table does, and it tops out at 17:00 UTC. Give every game equal weight instead — normalise each of 452 games to its own average, then take the mean across games — and the peak moves to 14:00 UTC, where the typical game sits at 1.152 times its own daily average against 0.844 at 06:00.
Four hours of difference between those two answers, and the reason is size. The largest games run late: Brookhaven RP averaged 651,771 players at 19:00 UTC against 374,565 at 05:00, and Murder Mystery 2 averaged 984,856 at 19:00 against 698,222 at 05:00. They are heavy enough to drag the whole curve toward the evening. A mid-sized game is more likely to top out in the early UTC afternoon.
Some games run on a different clock entirely
Across the wider set of 452 games, 172 reached their daily high at 13:00 or 14:00 UTC. But 56 of them peaked between 22:00 and 03:00, and the ones that do are strikingly consistent about it — this is a property of a game's audience, not day-to-day noise.
- Evade peaked at 13:00 UTC on all nine days, without exception: 67,240 players at 13:00 against 39,968 at 22:00.
- Sol's RNG peaked between 08:00 and 10:00 UTC every day — the earliest clock among the large games here.
- Natural Disaster Survival peaked at 01:00 or 02:00 UTC on eight of the nine days.
- Jujutsu Shenanigans peaked between 23:00 and 01:00 on seven of the nine, averaging 176,672 players at midnight UTC against 106,186 at 07:00.
- Flee the Facility peaked at 23:00 or midnight UTC on eight of the nine days — 46,416 players at midnight against 10,866 at 09:00.
- BedWars peaked between 01:00 and 03:00 UTC on seven of the nine.
- Dress to Impress held to the evening: 110,793 at 19:00 UTC against 51,147 at 06:00.
How far a game swings between its high and low
The gap between a game's best hour and its worst is its own number, and it varies more than the peak hour does.
| Game | Peak hour (UTC) | Peak | Low | Ratio |
|---|---|---|---|---|
| Flee the Facility | 00:00 | 46,416 | 10,866 | 4.27x |
| Tower of Hell | 18:00 | 50,634 | 14,314 | 3.54x |
| Dress to Impress | 19:00 | 110,793 | 51,147 | 2.17x |
| Brookhaven RP | 19:00 | 651,771 | 374,565 | 1.74x |
| Evade | 13:00 | 67,240 | 39,968 | 1.68x |
| Jujutsu Shenanigans | 00:00 | 176,672 | 106,186 | 1.66x |
| Blox Fruits | 13:00 | 427,700 | 266,985 | 1.60x |
| Murder Mystery 2 | 19:00 | 984,856 | 698,222 | 1.41x |
| Bee Swarm Simulator | 14:00 | 21,401 | 18,397 | 1.16x |
Bee Swarm Simulator is the flat one: it holds within 16% of itself all day, which is what a long-session game with a spread-out audience looks like. At the other end, Flee the Facility has four times as many people in it at midnight UTC as it does at 09:00. A game whose audience sits in one part of the world swings hardest, because that part of the world sleeps all at once.
Weekends add about a sixth
Two of the nine days were a weekend. The panel averaged 1,352,558 players across Saturday and Sunday against 1,153,681 across the seven weekdays — 17% higher. On the wider 452-game set with each game weighted equally, a weekend hour ran 18% above a weekday hour.
Two weekend days is a thin basis for a rule, so treat that as a first reading rather than a settled figure.
What the curve is good for
If you want a full server, the hour that matters is the game's own peak, not the network's. Evade at 06:00 UTC is a different game from Evade at 13:00, and no amount of platform-wide traffic changes that.
If you want an empty one — a quiet lobby, an uncontested map, a private-feeling world — 05:00 to 07:00 UTC is the platform's floor, and games with evening audiences are at their thinnest then.
And if a game you follow looks like it is losing people, check the hour before concluding anything. A 30% drop between 19:00 and 05:00 UTC is the clock, not the game. The live charts and each game's analytics page carry the hourly history to compare against.
How this was measured
Bloxism reads Roblox's public concurrent-player count for each game on a repeating cycle and stores every reading with its timestamp. For this article, readings between 00:00 UTC on 6 August 2026 and 00:00 UTC on 15 August were averaged per game per clock hour per day, then summed across games — averaging first, because the sampler visits different games on different ticks and summing raw rows would measure the sampler rather than the platform.
The headline curve uses a balanced panel: the 91 games that have a reading in every one of the 216 (day, hour) cells in the window. That matters more than panel size. A wider cohort with occasional gaps produces phantom dips at whichever hours a large game happened to be missing, and one appeared at 21:00 in the 452-game set before the balanced panel ruled it out. Per-game peak hours and the equal-weight index use the wider set, because neither can be biased by which games are present in a given hour.
Two limits worth naming. The panel is a sample of games with pages on this site, not the whole platform, so the absolute totals are smaller than Roblox's own — the shape is the finding, not the level. And the window is August, when schools across the northern hemisphere are on holiday; a term-time curve would likely be flatter during the day and sharper in the evening.
