“I Can't Wait for GTA6” — a 25-Year-Old AI Entrepreneur Starts Building His Own GTA6, With AI Alone

Unable to wait for GTA6's release, one person started building it himself. Ziwen Xu, a 25-year-old AI entrepreneur, has been developing a GTA6-style open-world crime game, “GT-Caliber,” since June 10 through “vibe coding” — running AI agents (mainly Claude). His stated goal: to release before the real GTA6 (out November 19). It's reckless, but the development process has become a talking point as a public experiment mirroring where AI stands today. GTA6 FEED has summarized the overview.

This article is based on information as of July 4, 2026.

What Is He Doing

Ziwen Xu is the founder of Hyperecho, an AI-agent startup (which helps companies adopt AI). On June 10, he kicked off the project with a post to the effect of “I've started making GTA6. I'm still typing and it hasn't sunk in yet,” and revealed he'd paid for Claude's higher-tier plan (Claude Max 20x) for this development.

His development method is what's called “vibe coding” — running AI agents continuously in a loop and letting them handle generating, testing, and fixing code. He posts progress on X almost daily, publishes the code openly on GitHub, and sometimes livestreams the AI at work. He is also recruiting human collaborators who can do modeling, composing, level design, and so on. It's not that he's dumping everything on the AI.

The trigger was a challenge thrown down by AI investor Matt Shumer. Shumer had posted, in essence, “Can you loop a model and build a GTA6-class open world with quality and scale beyond the initial trailer?” — and Xu reposted it and then set to work himself. Of the goal, he wrote: “Ambitious, and probably foolish. I'll do it anyway.”

Progress: From a “Blue Box” to a City in About Nine Days

The pace is fast — and chaotic.

Day one's footage was just a blue-box-like character bouncing around on a flat plane. By day two a humanoid character was running through a city; by day three NPCs were walking, cars were driving on roads, and there was even shooting and a smartphone with an Instagram-like social feed. By day seven, a stretch of cityscape with roads appeared, reaching the stage where he said it “looked like GTA for the first time.” Currently implemented are driving, NPCs, shooting, a phone (with map, contacts, and wallet), an intro screen with original BGM, weather changes, and more. The fine details of roads and sidewalks, he says, can't be nailed by AI alone and are being adjusted by hand.

The development engine initially started with the free Godot, but on day seven he moved to Unreal Engine. During this period, the engine choice reportedly flip-flopped several times.

Stumbles and “Incidents”

What people find amusing isn't the level of completion so much as the trouble that happens along the way.

The most emblematic is the mix-up over the setting. GTA6's Vice City is modeled on Miami (Florida), but at one point the AI agent generated an LA-style cluster of skyscrapers. This episode was cited all over as a prime example of how “AI can quickly build a generic box (sandbox), but it can't build the specific, deliberate, culturally accurate world that makes GTA GTA.”

The cost issue is vivid too. Xu wrote that he burned through 33% of Claude's weekly quota (on the 20x plan) in a single day — “the clock has started ticking.” Development that keeps running AI is also a battle against your usage quota. It's worth noting that this progress-posting took off largely because it coincided with the timing of Rockstar announcing GTA6 pre-orders (June 25).

Assessment: AI's “Speed” and “Absence of Vision”

How should we view this project? Bluntly put, the speed of spinning up a sandbox in a few days is genuinely astonishing as a milestone for agentic AI development. On the other hand, almost no observer believes the goal of matching — or beating to market — the real GTA6 will actually be achieved. The fair assessment is that the value lies in the “experiment,” not the “competition.”

The key is that LA mix-up from earlier. AI can mass-produce fragments of features and looks at high speed, but the creative vision of “what to make” — which city, which radio stations, which jokes and characters to place — still belongs to the human domain. Fast, but no vision. That's the shared read across much of the media.

Ironically, this experiment is also taken as material backing Take-Two CEO Zelnick's argument. Zelnick has repeatedly said that while AI is useful as a development tool, it doesn't have the creative vision to produce a billion-dollar blockbuster like GTA. Rockstar itself reportedly avoided using AI in developing GTA6. If GTA6 — hand-crafted over 13 years — arrives on November 19, the contrast will only stand out more.

Reactions are split, too.

  • “It runs this well in just a few days? I feel AI's potential.”
  • “He's so serious about it that it makes you want to root for him.”
  • “Interesting, but still a long way from GTA6.”
  • “In the end, isn't this just engagement-baiting for buzz?”

It's also been pointed out that works made to resemble GTA — even without directly reusing assets — can, in terms of trade dress like tone and the mold of the city, more easily draw the rights holder's gaze down the line. This is a point that fan-made works in the AI era may face going forward.

Summary

  • What he's doing: 25-year-old AI entrepreneur Ziwen Xu is making a GTA6-style game, “GT-Caliber,” by running AI (mainly Claude). The goal is to release before the real thing, but the view that it rivals GTA6 is not the majority.
  • Why it's drawing attention: because it shows, in real time as daily public updates, both the “speed” of AI-agent development and the “absence of creative vision.”

This project is a fan-made experiment by an individual unaffiliated with Rockstar. It isn't information about GTA6 itself, but while waiting for release it's a very of-the-moment topic that makes you consider “how far can AI build a game?”

Disclaimer

This article was compiled and organized by GTA6 FEED based on Ziwen Xu's own posts on X (formerly Twitter) and reporting from overseas media. “GT-Caliber” is a fan-made project by an individual with no relation whatsoever to Rockstar Games or Take-Two Interactive, and is unrelated to the official development or announcements of GTA6. The development status, implemented content, and goals are based on information as of July 4, 2026, and are subject to change. For the latest status, please check primary sources such as the creator's own X account.