The biggest GTA RP server is now NoPixel — we counted

NoPixel, and it is not close any more.

Across 13 complete censuses of every live Grand Theft Auto stream on Kick, Twitch and YouTube — 04:03:36 to 04:16:34 UTC on 22 September 2026 — streams whose titles named NoPixel carried a mean of 116,168 concurrent viewers. Windy City, which led this measurement when we first ran it, carried 12,666. That is a factor of 9.2×, and NoPixel was ahead in all 13 of the 13 censuses.

This post said the opposite when it was published on 15 August, and we are leaving the old numbers in it rather than quietly restating them. On that morning Windy City took 13,042 mean viewers against NoPixel’s 5,530 — Windy City ahead by 2.4×, on the same method, at the same hour of the day. Thirty-eight days later Windy City is essentially where it was, down 2.9%, and NoPixel has gone up twenty-one fold.

Something happened in between, and it has a date: 8 September 2026.

Server	Mean viewers	Channels	Platform split	Top stream’s share
NoPixel	116,168	152	Twitch 80%, Kick 20%	28%
Windy City	12,666	23	Kick 97%, Twitch 3%	39%
Prodigy RP	5,218	32	Kick 93%, Twitch 7%	72%
Headliner RP	3,310	67	Twitch 91%, Kick 9%	57%
FamilyRP	1,497	8	Kick 100%	40%
ChaseRP	1,445	27	Twitch 100%	77%
DovuX Life RP	1,184	3	Kick 100%	63%
Complexo (CPX)	1,122	32	Kick 66%, Twitch 34%	26%
District 10 (D10)	753	18	Twitch 55%, Kick 45%	45%
The Academy RP	446	4	Twitch 100%	62%
Prestige RP	386	20	Kick 64%, Twitch 36%	55%
New Day RP	74	2	Twitch 100%	86%
Cidade dos Anjos	63	7	Twitch 100%	54%

13 complete censuses, 04:03:36–04:16:34 UTC, 22 September 2026. Mean live GTA audience across the window was 320,788 viewers on 1,848 channels.

NoPixel is now 36.2% of every live GTA viewer we can see, on any platform, in any language, playing anything. It is 68.6% of everything our classifier files as roleplay. One server is carrying more than a third of the category.

What happened on 8 September

Rockstar opened its own roleplay server. nopixel V — lower-case n, Rockstar’s styling — went into closed beta on 8 September on the Rockstar Games Launcher, the first Rockstar-sanctioned GTA RP server, and the clearest consequence yet of Cfx.re, the team behind FiveM, becoming part of Rockstar. Rockstar’s own announcement is on the Newswire (confirmed: the page is live; we could not read its body text, which renders client-side, so the access details below are attributed to the reporting rather than to Rockstar).

Access is the part that matters for this measurement. As reported by Dexerto, the beta is an invite-only allowlist of nearly 450 content creators and established roleplayers, rolled out in phases, with NoPixel’s Koil quoted as saying that “while the first phase of invites is nearly complete, it’s just the first group” and that “over time, access will expand”. NoPixel Public and NoPixel 4.0 keep running for everyone else. There is no public application for the first wave and, as the reporting notes, no legitimate way to buy in.

So the biggest GTA roleplay server by live audience is one that almost nobody reading this can play, by design. Roughly 450 invited accounts are generating more watched hours than the rest of the named roleplay scene combined. That is a genuinely new shape for this category, and it is the sharpest illustration we have yet found of the distinction the first version of this post raised and then left hanging: being watched and being played are different things.

The platform flip is the finding nobody else has

Here is the part we did not expect, and it is not visible from any server list or any single platform’s directory.

In August, NoPixel was a Kick server by audience: 81% Kick, 19% Twitch. Today it is 80% Twitch, 20% Kick, and zero YouTube. Its Kick audience grew 5.2× over the 38 days. Its Twitch audience grew 88×.

That runs against the grain of the whole category. In the same window, GTA viewing overall was 59.4% Kick, 37.8% Twitch and 2.8% YouTube — Kick’s lead is the single most durable fact in our data, and we have written it up more than once. The biggest server inside that Kick-majority category is a Twitch island.

The likeliest explanation is sitting in plain sight, and it is dated, which makes it testable. Twitch ran GTA RP Week from 8 to 30 September to celebrate nopixel V’s launch, with a Drops campaign — one hour watched for GTA$250K, up to the Burger Shot Tracksuit at six hours — announced on Twitch’s own blog. Twitch Drops require watching on Twitch. If Drops are doing this work, NoPixel’s Twitch share should fall back materially once the campaign ends on 30 September, and we will re-run these censuses in the first week of October and publish the answer on this URL whichever way it goes. If the share holds, the audience moved for the content and not the currency, and that is a more interesting result.

We cannot separate those two explanations today. Anyone who tells you they can, on 14 days of data with a Drops campaign running through all of them, is guessing.

Is this just xQc?

Largely, one person is a third of it, and you should know that before you quote the 116,168.

xQc simulcasts to Twitch and Kick, and across those two channels he averaged 43,124 concurrent viewers in our window — 37.1% of NoPixel’s entire audience. The top ten NoPixel streams are 76% of the server’s total. Whatever this is, it is not broad.

The check that matters is whether the headline survives without him. It does: delete both of xQc’s channels and NoPixel still averages 73,045 viewers, which is 5.8× Windy City rather than 9.2×. The finding is not an artefact of one broadcaster having a good night. It would be a different post if it were.

Worth naming as a caveat: summing a simulcast across two platforms counts two real concurrent audiences, but one performer. We did the same in August, so the comparison is consistent, but a reader who cares about people watching one show should mentally treat xQc as one number, not two.

This is a level, not a blink

The category-wide series says 8 September moved the floor rather than causing a spike. Taking every complete UTC day in our 30-day hourly series, the mean concurrent GTA audience was 239,926 across 24 August to 7 September and 317,705 across 9 to 21 September — up 32.4%, and visible at every hour of the clock rather than at one peak.

Two honest observations from that chart, both of which cut against the simplest story.

The level is already fading. The peak daily mean was 358,923 on 13 September and the last two complete days are 253,912 and 270,923 — still well above the August baseline, but the top of the curve was a week and a half ago, and Drops end on the 30th.

And the arithmetic does not let NoPixel be purely additive. Comparing like with like at the hour we sample, the 04:00 UTC mean was 184,098 before 8 September and 265,092 after — a gain of about 81,000. NoPixel alone is 116,168. A server larger than the entire increase cannot be made only of people who were not watching GTA before; some tens of thousands of these viewers were already in this category, watching something else. That is a redistribution as much as a windfall, and it is why Windy City being flat rather than down is a slightly surprising result that we cannot fully explain.

Watched is not played, and the server lists are measuring a third thing

Search biggest gta rp server and page one is mostly Sportskeeda rankings from 2021 and 2023 plus FiveM server-list sites. The lists that are current rank by player count pulled from server browsers, which is why you will see names there — Los Angeles Cops & Robbers, Alpenland, RP4you — that do not appear anywhere in our data at all. Nobody streams them. They may be full.

GTA World remains the cleanest example. It is a text-based server reporting over 137,000 registered users and, by its own description, 300 to 500 players online at a time. In our window it carried 26 viewers. Both things are true. Reading somebody else’s chat log is not a broadcast.

So read the table above for what it is: a ranking of where the GTA roleplay audience is, which is the thing an outsider actually encounters when they open the GTA category, and a good proxy for which servers are alive and easy to learn by watching. It is not a headcount. A quiet server is not an empty one, and this month a whitelist of 450 people just outdrew everybody.

What this method still cannot see

Server attribution here is done by reading stream titles, and it only works where creators name their server. In English almost everybody does. In most other languages almost nobody does.

Language	Live viewers	Share attributed to a named server
English	186,804	74.6%
Spanish	61,607	3.9%
Portuguese	27,319	3.0%
Arabic	23,488	0.0%
Thai	2,119	0.0%

Single census, 04:16:34 UTC, 22 September 2026. Language is the platform's own self-declared tag, so it is imperfect — but not so imperfect that a zero becomes a half.

English attribution has risen sharply since August, from 42.2% to 74.6%, for a banal reason: the biggest server in the scene is one whose streamers write its name in the title. That is a measurement artefact of NoPixel’s growth, not a change in labelling habits.

Everything else on that table is the hole. Arabic is 23,488 live viewers and not one of them names a server we recognise. If a server in the Arabic or Thai scene is bigger than NoPixel, this method would never see it, and we cannot rule that out from a census. A further 379 channels and 21,176 viewers are unmistakably roleplay — the title says RP, roleplay, FiveM or whitelist — and name nothing we match.

The rest of the caveats, hardest first:

• One 13-minute window at 04:00 UTC, on a Tuesday. The August census was a Saturday. The GTA audience swings roughly 5× across the UTC day and we have written about that; a weekday-versus-weekend comparison at a fixed hour is a real confound for the small servers in that table. It is not a plausible explanation for a 21× change in the top row. Strefa RP, which we measured as the fourth-largest language scene yesterday, registers zero here, because 04:00 UTC is 06:00 in Warsaw. Every census this site publishes has that blind spot.
• “NoPixel” in a title is a server family, not a version. NoPixel Public and 4.0 are still running. Of the 152 matched channels, 127 explicitly name V or 5, accounting for 72,704 of the 116,168 viewers; the remaining ~43,000 say only “NoPixel” and we are not going to guess which server they are on. Read the headline as the NoPixel family.
• Channel counts are not comparable between the two dates. Kick’s public API returns only the top 100 channels in a category, so small Kick streams reach us only once we know them by name, and our discovery coverage has changed since August. Viewer totals are barely affected — Kick’s top 100 hold nearly all of Kick’s audience — but do not read Windy City’s 63 channels in August against 23 today as a collapse in breadth. That comparison is measuring our collector.
• Title matching is a floor inside a server too. A NoPixel streamer who writes only their character name is invisible to it.
• Overlap is not deduplicated. A handful of titles name two servers — one of The Academy RP’s channels also names NoPixel V — and they are counted in both rows. It is a rounding error at these magnitudes, not a correction.

Sources and how to check us

Everything above comes from 13 full censuses — the live board plus the tail, with the publish timestamp matched across both, so channel counts and viewer sums reconcile exactly against the platform summary in all 13, with no duplicate channel appearing twice — taken 04:03:36 to 04:16:34 UTC on 22 September 2026. Servers were attributed by matching a hand-built alias list against stream titles, and every match in the top rows was read by eye before publication. The 15 August figures are the ones originally published here, from 17 censuses taken 04:05 to 04:31 UTC that day.

If you want to quote any of this, the permalink is https://luciastream.com/blog/biggest-gta-rp-server/, the sampling window is 04:03:36 to 04:16:34 UTC on 22 September 2026, and the collection method is documented at luciastream.com/methodology. The underlying numbers are the same ones behind our live board, sampled every 60 seconds, and the leaderboard carries the 30-day version — both are where you can check whether any of this still holds by the time you read it.

It did not hold for five weeks last time. We said in the first version of this post that if the ranking changed, that was the finding and we would rather be told than believed, so: it changed, we were the ones who noticed, and the next scheduled check is the first week of October, after Twitch’s Drops campaign ends.