GTA VI shipped about a year after its original fall 2025 target, and it slipped twice on the way. The public side of those delays, the dates and the wording, is a useful reference for anyone who has to put a date on software.
Grand Theft Auto VI launched on 19 November 2026 on PS5 and Xbox Series X|S. The next day Take-Two announced $3.79 billion in day-one retail sales by its own estimate, plus 32.5 million units sold before release, and called it the biggest entertainment launch in history.
The same game missed its target twice. I have no inside knowledge of how Rockstar plans its work, so everything below comes from public statements and reporting. The parts anyone can see still map closely onto problems small web teams have with estimates: when to give a date, what to cut, how much time to keep for finishing, and how to tell people you're late.
The timeline
| Announced | What happened | Promised release |
|---|---|---|
| December 2023 | Official reveal with Trailer 1 | 2025 |
| 2024 (Take-Two earnings report) | Window narrowed | Fall 2025 |
| 2 May 2025 | First delay | 26 May 2026 |
| 6 November 2025 | Second delay | 19 November 2026 |
| 19 November 2026 | Launch | Shipped |
Sources: Game Informer for the reveal, the 2025 window and the first delay, Engadget for the narrowing to fall 2025, and VGC for the second delay.
Two details in that table matter more to me than the total slip. Both delays came months before the date they replaced: at least four months before the fall 2025 window, and about six and a half months before 26 May 2026. And both times the new target was an exact day. The first exact day didn't hold, and the second one did.
What the two delay notes say
The first note apologized, thanked players for their patience while the team finished the game, and said the studio needed "extra time to deliver at the level of quality you expect and deserve" (PC Gamer). The second apologized again and said the extra months would let the team finish the game "with the level of polish you have come to expect and deserve" (VGC). Take-Two reported quarterly results the same day, and a statement from CEO Strauss Zelnick that came with them repeated the new date.
Neither note gives a technical reason. There is no list of unfinished features and no explanation of what "polish" covers. For an audience of players that is a defensible choice, since they don't need to see the bug tracker. Each note is short: an apology, a brief reason, a new exact date, and thanks.
I'd keep that structure for client work and add what a consumer announcement can leave out: what is finished, what isn't, and what the new date depends on. A client or product owner has to plan marketing, content and budgets around your date, so they need more to go on than a promise.
Early estimates are wide whether you say so or not
Steve McConnell's cone of uncertainty describes how estimate accuracy improves as a project moves forward. In Jeff Atwood's summary of McConnell's book, an estimate made at the initial concept stage can be off by a factor of four in either direction, a 16x range, and that is the best case.
A year-wide window like "2025", published more than a year ahead, is honest about that uncertainty. Narrowing it to a season or a day is where a team takes on risk, because the promise can get more precise faster than the knowledge behind it.
For a small team, the practical answer is to give a range with a confidence level attached and narrow it as the work narrows. If you track how many tickets you close per week, a Monte Carlo forecast is a short script:
// Tickets closed per week over recent months (placeholder data, use your own)
const history = [4, 7, 3, 6, 5, 8, 2, 6, 5, 4];
function simulateWeeks(backlog: number, runs = 10_000): number[] {
const results: number[] = [];
for (let i = 0; i < runs; i++) {
let remaining = backlog;
let weeks = 0;
while (remaining > 0) {
remaining -= history[Math.floor(Math.random() * history.length)];
weeks++;
}
results.push(weeks);
}
return results.sort((a, b) => a - b);
}
const runs = simulateWeeks(42);
const percentile = (p: number) => runs[Math.floor(runs.length * p)];
console.log(
`50%: ${percentile(0.5)} weeks, 85%: ${percentile(0.85)} weeks, 95%: ${percentile(0.95)} weeks`
);Each run replays random weeks from your own history until the backlog is empty, and the sorted results give you percentiles. I'd quote the 85% figure as the date and say plainly that the 50% figure is a coin flip. The model assumes the backlog stays the same size, which it never does, so rerun it every week as tickets are added and let the range move with the backlog.
Scope is the other lever
When a date is at risk, the two levers you control are the date and the scope. GTA VI launched with some limits stated before release. Rockstar says the game targets 30 fps on every console, PS5 Pro included, and the Pro uses its extra headroom for PSSR upscaling and ray-traced shadows (Shacknews). Take-Two's Q2 FY27 earnings release lists the game for PS5 and Xbox Series X|S only. I don't know whether either was a cut or the plan from the start. Either way, both are public statements of what the launch version does and doesn't do.
The web version of this is writing down what v1 won't do. For example: support for the last two versions of evergreen browsers only, no offline mode, two languages at launch. Once it's written down, it is much harder to reopen in the final week.
Cutting a feature late is much cheaper if it was built behind a flag from the start:
export const flags = {
savedCarts: process.env.FLAG_SAVED_CARTS === "true",
giftCards: process.env.FLAG_GIFT_CARDS === "true",
} as const;import { CartSummary } from "@/components/cart/CartSummary";
import { flags } from "@/lib/flags";
import { SavedCarts } from "@/components/cart/SavedCarts";
export default function CartPage() {
return (
<main>
<CartSummary />
{flags.savedCarts && <SavedCarts />}
</main>
);
}Dropping saved carts from the release then becomes an environment variable and a redeploy. The alternative is a revert across a dozen files the night before launch.
My rule for choosing the lever: if the date matters more than any single feature (a campaign, a trade fair, a seasonal sale), cut scope. If the feature is the reason for the release, move the date. I'd guess a game like GTA VI sits closer to the second case: the product is one large world, and there's no obvious half of it to ship first.
Put the polish in the plan
Both delay notes talk about finishing the game, and they give quality and polish as the reason for the extra time. In my view the finishing stretch is the part most often left out of estimates. It covers the work between "feature complete" and "done": browser-specific bugs, loading and empty states, accessibility fixes, performance on a mid-range phone, and copy changes that arrive late.
I'd give it its own line in the estimate instead of spreading it thinly across every feature: a fixed block at the end, sized as a share of the build time, in which nobody starts new features. About a fifth of the build time is where I'd start, adjusted later from your own project history.
Polish also needs an end condition, or it expands to fill whatever time is left. Automated gates help here. A Lighthouse CI config that fails the build below your thresholds turns "is it good enough?" into a check:
{
"ci": {
"collect": {
"startServerCommand": "npm run start",
"url": ["http://localhost:3000/", "http://localhost:3000/blog"],
"numberOfRuns": 3
},
"assert": {
"assertions": {
"categories:performance": ["error", { "minScore": 0.9 }],
"categories:accessibility": ["error", { "minScore": 0.95 }],
"largest-contentful-paint": ["error", { "maxNumericValue": 2500 }],
"cumulative-layout-shift": ["error", { "maxNumericValue": 0.1 }]
}
}
}
}Run it with npx @lhci/cli autorun in CI. The LCP and CLS limits match the "good" thresholds Google publishes for Core Web Vitals. Combine the gates with a short list of release blockers, and the finishing phase is over when both are clear.
How to announce a delay
Rockstar's timing is the part I'd copy first. Both delays were announced months ahead, and that gives everyone downstream time to replan. On a web project that means telling the client in the same week you first see the date is in danger.
The part I wouldn't copy is slipping twice. The first new date didn't hold, which meant a second round of apologies. If you have to move, move once, to a date you can defend with numbers like the 85% forecast above.
Here is a template for a client project. The project and details are made up:
Subject: New launch date for the booking flow: 12 March
The booking flow won't be ready for 20 February. I'm sorry for the change.
Done: search, listing pages, payments (tested in Stripe test mode).
Left: email notifications, mobile layout fixes, accessibility pass.
Why: the webhook handling took longer than planned, and I don't want
to ship checkout without the finishing week in the plan.
New date: 12 March (85% confidence on our forecast).
Earliest realistic date: 5 March.
What could move it: email copy from your side by 26 February.
Next update: Friday, whatever the status.It has the same apology, reason and new date as Rockstar's notes. It adds what is done, what is left, a range around the date and the dependency on the client, which a client needs and a player audience doesn't.
Record sales don't make delays free
The $3.79 billion figure is Take-Two's own, "based on company estimates", as the press release footnote puts it. A third-party estimate from Sensor Tower, reported by VGC before Take-Two published its number, was lower: about 24.6 million copies and $2.36 billion on day one. The two figures come from different methods, so I wouldn't read much into the gap. The same press release also cites a Metacritic score of 99.
The delays still had a visible cost. Take-Two shares fell about 10% in pre-market trading on the day the first delay was announced, according to Windows Central, citing the Financial Times.
There is also survivorship bias in reading this story as "delays pay off". GTA VI started with a reserve of goodwill that few products have: according to the same press release, GTA V has "sold-in nearly 235 million units". A freelance developer or a five-person agency doesn't get that patience from a client after a second missed date. The habits worth taking from this launch are the ones that reduce how often you need patience in the first place:
- Give ranges with a confidence level, and narrow them as the work narrows.
- Decide early whether the date or the scope is fixed, and write the v1 limits down.
- Budget the finishing phase as its own block, with automated gates that say when it's done.
- If you have to slip, do it early and once, with a date backed by data and a list of what it depends on.