One email each morning — yesterday's AI, sortedGet it in your inbox

METAL LAB

Grok 4.6 Debuts, Users Flood X With Game and 3D Model Demos Within 48 Hours

From games built without Unity to a 3D helicopter model, ten user-shared creations went viral

이미지: METAL LAB 생성

Summary

  • Less than 48 hours after xAI released Grok 4.6, users began sharing a wave of games, 3D models, and websites built with the model
  • One example claimed the same task cost $1.80 with Grok 4.6, cheaper than a rival model that cost $4.20
  • According to pricing revealed on August 12, Grok 4.6 charges $2 per million input tokens, keeping it cheaper than competing models
공개 시점
2026년 8월 13일 기준 48시간 이내(추정 8월 11~12일)
정리한 계정
Min Choi(X), 사례 10건 게시
비용 비교 사례
동일 작업 기준 1.80달러 vs 4.20달러
Grok 4.6 가격(배경)
100만 토큰당 입력 2달러 / 출력 6달러
Intelligence Index(배경)
61점, GPT-5.6 Sol과 동급

What happened

Less than two days after xAI released its large language model Grok 4.6, user-made creations went viral on X (formerly Twitter). Min Choi, who runs an AI news account, rounded up ten of them. The list included a game built without touching the Unity engine, a snowboarding game made from scratch, an interactive 3D simulation, an AI simulation faithfully recreating the sitcom "The Office," and a 3D recreation of an Airbus H145 rescue helicopter. Also featured were a 4D holographic wedding website, a 3D reconstruction of the pirate ship Queen Anne's Revenge, a 3D world generated from a single prompt, and a hand-drawn sketch turned into a finished website. One standout was a cost comparison: the same task reportedly cost $1.80 with Grok 4.6, versus $4.20 for a comparison model.

What this is about

Grok is the model family developed by xAI, led by Elon Musk. According to pricing revealed on August 12, Grok 4.6 is priced at $2 per million input tokens and $6 per million output tokens — notably lower than Anthropic's Claude Opus 5 ($5/$25) or OpenAI's GPT-5.6 Sol ($5/$30). On the Artificial Analysis Intelligence Index, it scored 61, putting it on par with GPT-5.6 Sol. The combination of low pricing and solid code-generation performance appears to be behind the sudden surge of "vibe coding" cases — game prototypes, 3D models, and websites produced from prompts alone — in such a short span. Musk has previously said the upcoming Grok 4.7 will be trained on real SpaceX engineering data and will surpass every existing model, making Grok 4.6's early reception notable as a potential stepping stone toward that.

What changes now

The examples shared this week show that game-like prototypes, 3D models, and finished websites can now be produced from text prompts alone, without writing a single line of code. With one case showing costs less than half those of a rival model, individual developers and hobbyist creators may be more inclined to experiment with quick ideas. Still, since these results were shared as individual posts, actual quality and reproducibility may vary from user to user.